Figures (sim, capacity sweep, two-node), paper results v0.1 with two-node evidence, LOG

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

Simon-Pierre Boucher committed 2 days ago (Nov 1, 2026) parent 5edaf3e Browse files

12 changed files +264 −2

modified benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl +83 −0

		`@@ -112,3 +112,86 @@`
112	112	{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "none", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor none --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "2574612"}, "seed": 0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:56:12", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 251.57787499483675, "prompt_tps": 71.65770317737694, "gen_tps": 48.42080705339218, "decode_tps_wall": 47.664664722850894, "itl_ms_p50": 20.775957964360714, "itl_ms_p95": 26.101749972440302, "itl_ms_p99": 30.63241601921618, "itl_ms_mean": 20.979902110180802, "total_s": 1.5733117079362273, "peak_mem_gb": 2.472523792, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.7912429076116048, "cache_misses": 1950, "cache_hits": 7391, "stall_s": 0.6353897843509912, "hits_device": 0, "layers_complete": 0, "prefetch": null, "ssd_bytes": 6900940800, "ssd_read_s": 1.1909447448560968}
113	113	{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "none", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor none --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "2574612"}, "seed": 0, "warmup": true, "repeat": -1, "family": "reasoning", "prompt": "Q: What is the smallest positive integer divisible by 2, 3, ", "ts": "2026-11-01T01:56:13", "prompt_tokens": 29, "gen_tokens": 64, "ttft_ms": 289.09962507896125, "prompt_tps": 100.43671230577584, "gen_tps": 49.18196558897762, "decode_tps_wall": 48.41389732634681, "itl_ms_p50": 20.339249982498586, "itl_ms_p95": 26.802208041772246, "itl_ms_p99": 34.39366596285254, "itl_ms_mean": 20.655226189687493, "total_s": 1.5903788750292733, "peak_mem_gb": 2.597377892, "text": "The least common multiple (LCM) of 2, 3, 4, 5, and 6 is the smallest positive integer that is divisible by all of these numbers. The prime factorization of 2 is 2. The prime factorization of 3 is 3. T", "cache_hit_rate": 0.7793997550020416, "cache_misses": 2161, "cache_hits": 7635, "stall_s": 0.6567125713918358, "hits_device": 0, "layers_complete": 0, "prefetch": null, "ssd_bytes": 7647657984, "ssd_read_s": 1.286987041006796}
114	114	{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "none", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor none --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "2574612"}, "seed": 0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:56:15", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 199.78074997197837, "prompt_tps": 95.28834680741699, "gen_tps": 56.62403522263462, "decode_tps_wall": 55.73988672118333, "itl_ms_p50": 17.17858307529241, "itl_ms_p95": 24.96608393266797, "itl_ms_p99": 26.742333080619574, "itl_ms_mean": 17.940474206596495, "total_s": 1.3300306249875575, "peak_mem_gb": 2.469248884, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.8563977006599958, "cache_misses": 1349, "cache_hits": 8045, "stall_s": 0.4672505249036476, "hits_device": 0, "layers_complete": 0, "prefetch": null, "ssd_bytes": 4774035456, "ssd_read_s": 0.7979402884375304}
	115	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "none", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor none --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "2574612"}, "seed": 0, "warmup": false, "repeat": 0, "family": "reasoning", "prompt": "Q: A tank fills in 6 hours with pipe A and 9 hours with pipe", "ts": "2026-11-01T01:56:16", "prompt_tokens": 32, "gen_tokens": 64, "ttft_ms": 284.16120901238173, "prompt_tps": 112.75508099171603, "gen_tps": 44.10160803690054, "decode_tps_wall": 43.41288564870694, "itl_ms_p50": 22.570249973796308, "itl_ms_p95": 30.478207976557314, "itl_ms_p99": 33.801999990828335, "itl_ms_mean": 23.034635570920294, "total_s": 1.7353432499803603, "peak_mem_gb": 2.755937396, "text": "If pipe A fills the tank in 6 hours, then the rate of pipe A is 1 / 6 of the tank per hour. If pipe B fills the tank in 9 hours, then the rate of pipe B is 1 / 9 of the tank per hour. The combined rat", "cache_hit_rate": 0.7463080684596577, "cache_misses": 2594, "cache_hits": 7631, "stall_s": 0.7904208408435807, "hits_device": 0, "layers_complete": 0, "prefetch": null, "ssd_bytes": 9180020736, "ssd_read_s": 1.5752222624141723}
	116	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "none", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor none --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "2574612"}, "seed": 0, "warmup": false, "repeat": 0, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:56:18", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 238.3841669652611, "prompt_tps": 75.60417725501566, "gen_tps": 51.33413095018941, "decode_tps_wall": 50.53244724218592, "itl_ms_p50": 19.192834035493433, "itl_ms_p95": 24.936707923188806, "itl_ms_p99": 27.704833075404167, "itl_ms_mean": 19.78926520632018, "total_s": 1.4851078749634326, "peak_mem_gb": 2.464595216, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.7930628412375549, "cache_misses": 1933, "cache_hits": 7408, "stall_s": 0.5997709431685507, "hits_device": 0, "layers_complete": 0, "prefetch": null, "ssd_bytes": 6840778752, "ssd_read_s": 1.1616856374312192}
	117	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "none", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor none --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "2574612"}, "seed": 0, "warmup": false, "repeat": 0, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:56:20", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 221.29808401223272, "prompt_tps": 63.352071222316766, "gen_tps": 44.45762609825912, "decode_tps_wall": 43.763362020621706, "itl_ms_p50": 22.22233300562948, "itl_ms_p95": 30.22420802153647, "itl_ms_p99": 33.73758401721716, "itl_ms_mean": 22.850164014565213, "total_s": 1.660858416929841, "peak_mem_gb": 2.374838524, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.7287056029364137, "cache_misses": 2513, "cache_hits": 6750, "stall_s": 0.744502259651199, "hits_device": 0, "layers_complete": 0, "prefetch": null, "ssd_bytes": 8893366272, "ssd_read_s": 1.5194422950735316}
	118	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "none", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor none --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "2574612"}, "seed": 0, "warmup": false, "repeat": 0, "family": "reasoning", "prompt": "Q: What is the smallest positive integer divisible by 2, 3, ", "ts": "2026-11-01T01:56:21", "prompt_tokens": 29, "gen_tokens": 64, "ttft_ms": 269.4497909396887, "prompt_tps": 107.76387631799321, "gen_tps": 49.1663455435908, "decode_tps_wall": 48.398579948683256, "itl_ms_p50": 20.051375031471252, "itl_ms_p95": 28.881374979391694, "itl_ms_p99": 36.583874956704676, "itl_ms_mean": 20.661763238927556, "total_s": 1.5711408749921247, "peak_mem_gb": 2.607646564, "text": "The least common multiple (LCM) of 2, 3, 4, 5, and 6 is the smallest positive integer that is divisible by all of these numbers. The prime factorization of 2 is 2. The prime factorization of 3 is 3. T", "cache_hit_rate": 0.7755206206614945, "cache_misses": 2199, "cache_hits": 7597, "stall_s": 0.6573501937091351, "hits_device": 0, "layers_complete": 0, "prefetch": null, "ssd_bytes": 7782137856, "ssd_read_s": 1.309832967002876}
	119	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "none", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor none --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "2574612"}, "seed": 0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:56:23", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 235.84695789031684, "prompt_tps": 101.89591833638708, "gen_tps": 56.34368446101921, "decode_tps_wall": 55.463920645825475, "itl_ms_p50": 17.546750023029745, "itl_ms_p95": 24.633875000290573, "itl_ms_p99": 26.978000067174435, "itl_ms_mean": 18.029738762711602, "total_s": 1.3717204999411479, "peak_mem_gb": 2.480622872, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.8375391032325339, "cache_misses": 1558, "cache_hits": 8032, "stall_s": 0.4964398378506303, "hits_device": 0, "layers_complete": 0, "prefetch": null, "ssd_bytes": 5513674752, "ssd_read_s": 0.9026410088408738}
	120	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "none", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor none --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "2574612"}, "seed": 0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:56:24", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 190.15433301683515, "prompt_tps": 100.08026174828318, "gen_tps": 58.397335353996795, "decode_tps_wall": 57.48555662623313, "itl_ms_p50": 16.422374988906085, "itl_ms_p95": 23.83612492121756, "itl_ms_p99": 25.87599994149059, "itl_ms_mean": 17.39567395166627, "total_s": 1.2860817919718102, "peak_mem_gb": 2.469347188, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.8592718756653183, "cache_misses": 1322, "cache_hits": 8072, "stall_s": 0.44590369798243046, "hits_device": 0, "layers_complete": 0, "prefetch": null, "ssd_bytes": 4678483968, "ssd_read_s": 0.7657993045868352}
	121	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "none", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor none --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "2574612"}, "seed": 0, "warmup": false, "repeat": 1, "family": "reasoning", "prompt": "Q: What is the smallest positive integer divisible by 2, 3, ", "ts": "2026-11-01T01:56:25", "prompt_tokens": 29, "gen_tokens": 64, "ttft_ms": 274.3655410595238, "prompt_tps": 105.82126424171258, "gen_tps": 49.77887291230627, "decode_tps_wall": 49.00150359764567, "itl_ms_p50": 20.00466699246317, "itl_ms_p95": 26.791749987751245, "itl_ms_p99": 34.54133297782391, "itl_ms_mean": 20.40753704643557, "total_s": 1.5600403749849647, "peak_mem_gb": 2.57641866, "text": "The least common multiple (LCM) of 2, 3, 4, 5, and 6 is the smallest positive integer that is divisible by all of these numbers. The prime factorization of 2 is 2. The prime factorization of 3 is 3. T", "cache_hit_rate": 0.7832788893425888, "cache_misses": 2123, "cache_hits": 7673, "stall_s": 0.6371748555684462, "hits_device": 0, "layers_complete": 0, "prefetch": null, "ssd_bytes": 7513178112, "ssd_read_s": 1.2614294610684738}
	122	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "none", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor none --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "2574612"}, "seed": 0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:56:27", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 198.3139580115676, "prompt_tps": 95.95129713290473, "gen_tps": 57.90602971523838, "decode_tps_wall": 57.00180675362896, "itl_ms_p50": 16.657334053888917, "itl_ms_p95": 24.488208000548184, "itl_ms_p99": 25.874500046484172, "itl_ms_mean": 17.543303571449268, "total_s": 1.3035420830128714, "peak_mem_gb": 2.469248884, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.8563977006599958, "cache_misses": 1349, "cache_hits": 8045, "stall_s": 0.4601672978606075, "hits_device": 0, "layers_complete": 0, "prefetch": null, "ssd_bytes": 4774035456, "ssd_read_s": 0.802551290136762}
	123	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "none", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor none --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "2574612"}, "seed": 0, "warmup": false, "repeat": 1, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:56:28", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 226.38579201884568, "prompt_tps": 79.614709048029, "gen_tps": 52.61680569524645, "decode_tps_wall": 51.79508333273456, "itl_ms_p50": 19.09966696985066, "itl_ms_p95": 24.101708899252117, "itl_ms_p99": 27.171374997124076, "itl_ms_mean": 19.306851841051074, "total_s": 1.4427174580050632, "peak_mem_gb": 2.453978384, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.7957392142168933, "cache_misses": 1908, "cache_hits": 7433, "stall_s": 0.5772590979468077, "hits_device": 0, "layers_complete": 0, "prefetch": null, "ssd_bytes": 6752305152, "ssd_read_s": 1.1212810397846624}
	124	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "none", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor none --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "2574612"}, "seed": 0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:56:30", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 236.9219169486314, "prompt_tps": 101.43647058367213, "gen_tps": 57.36168942730434, "decode_tps_wall": 56.4659760563488, "itl_ms_p50": 17.01224991120398, "itl_ms_p95": 24.26766697317362, "itl_ms_p99": 26.685958029702306, "itl_ms_mean": 17.7097797619238, "total_s": 1.352638041949831, "peak_mem_gb": 2.48770076, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.8338894681960375, "cache_misses": 1593, "cache_hits": 7997, "stall_s": 0.4866465798113495, "hits_device": 0, "layers_complete": 0, "prefetch": null, "ssd_bytes": 5637537792, "ssd_read_s": 0.898337729042396}
	125	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "none", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor none --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "2574612"}, "seed": 0, "warmup": false, "repeat": 1, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:56:31", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 209.65387509204447, "prompt_tps": 66.86458074667341, "gen_tps": 45.7667499358963, "decode_tps_wall": 45.052025726211426, "itl_ms_p50": 21.982959005981684, "itl_ms_p95": 28.87145895510912, "itl_ms_p99": 31.690999981947243, "itl_ms_mean": 22.19656017416763, "total_s": 1.6080371660646051, "peak_mem_gb": 2.36304204, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.7323761200475009, "cache_misses": 2479, "cache_hits": 6784, "stall_s": 0.7163165680831298, "hits_device": 0, "layers_complete": 0, "prefetch": null, "ssd_bytes": 8773042176, "ssd_read_s": 1.480862574884668}
	126	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "none", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor none --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "2574612"}, "seed": 0, "warmup": false, "repeat": 1, "family": "reasoning", "prompt": "Q: A tank fills in 6 hours with pipe A and 9 hours with pipe", "ts": "2026-11-01T01:56:33", "prompt_tokens": 32, "gen_tokens": 64, "ttft_ms": 273.5227909870446, "prompt_tps": 117.1461191764391, "gen_tps": 47.4691289863635, "decode_tps_wall": 46.72785852638877, "itl_ms_p50": 20.936249988153577, "itl_ms_p95": 27.268707985058427, "itl_ms_p99": 31.281375093385577, "itl_ms_mean": 21.40050992140517, "total_s": 1.6217549160355702, "peak_mem_gb": 2.751677372, "text": "If pipe A fills the tank in 6 hours, then the rate of pipe A is 1 / 6 of the tank per hour. If pipe B fills the tank in 9 hours, then the rate of pipe B is 1 / 9 of the tank per hour. The combined rat", "cache_hit_rate": 0.7433740831295843, "cache_misses": 2624, "cache_hits": 7601, "stall_s": 0.7309220866300166, "hits_device": 0, "layers_complete": 0, "prefetch": null, "ssd_bytes": 9286189056, "ssd_read_s": 1.5045044647995383}
	127	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "reasoning", "prompt": "Q: A tank fills in 6 hours with pipe A and 9 hours with pipe", "ts": "2026-11-01T01:56:40", "prompt_tokens": 32, "gen_tokens": 64, "ttft_ms": 537.12837502826, "prompt_tps": 60.01188509066157, "gen_tps": 38.37134619658702, "decode_tps_wall": 37.77377745861214, "itl_ms_p50": 26.52791701257229, "itl_ms_p95": 31.685000052675605, "itl_ms_p99": 41.86441597994417, "itl_ms_mean": 26.47339152393951, "total_s": 2.2049520410364494, "peak_mem_gb": 2.8607033, "text": "If pipe A fills the tank in 6 hours, then the rate of pipe A is 1 / 6 of the tank per hour. If pipe B fills the tank in 9 hours, then the rate of pipe B is 1 / 9 of the tank per hour. The combined rat", "cache_hit_rate": 0.7713447432762837, "cache_misses": 2338, "cache_hits": 7887, "stall_s": 0.896944917505607, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 1456, "installed": 1456, "late": 32, "dup": 6824, "predictor": "transition", "m": 8}, "ssd_bytes": 13426753536, "ssd_read_s": 2.999714838573709}
	128	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:56:42", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 248.5780839342624, "prompt_tps": 96.6808910407994, "gen_tps": 47.010754168751824, "decode_tps_wall": 46.27650006579437, "itl_ms_p50": 21.070166025310755, "itl_ms_p95": 28.713042032904923, "itl_ms_p99": 30.693084001541138, "itl_ms_mean": 21.6092400803482, "total_s": 1.609960208996199, "peak_mem_gb": 2.487737752, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.8579770594369135, "cache_misses": 1362, "cache_hits": 8228, "stall_s": 0.46157957590185106, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 397, "installed": 397, "late": 14, "dup": 7755, "predictor": "transition", "m": 8}, "ssd_bytes": 6225002496, "ssd_read_s": 1.0489242697367445}
	129	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:56:44", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 216.25570906326175, "prompt_tps": 64.83709027208769, "gen_tps": 35.52815117461518, "decode_tps_wall": 34.97319774721887, "itl_ms_p50": 28.280625003390014, "itl_ms_p95": 36.04683291632682, "itl_ms_p99": 41.89708305057138, "itl_ms_mean": 28.593324729064037, "total_s": 2.017635166994296, "peak_mem_gb": 2.377201144, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.7943430853935011, "cache_misses": 1905, "cache_hits": 7358, "stall_s": 0.733634035452269, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 2211, "installed": 2207, "late": 68, "dup": 5837, "predictor": "transition", "m": 8}, "ssd_bytes": 14566293504, "ssd_read_s": 2.5604033486451954}
	130	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:56:46", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 234.04616699554026, "prompt_tps": 77.01186831209532, "gen_tps": 42.0842544781391, "decode_tps_wall": 41.42692295140082, "itl_ms_p50": 23.884625057689846, "itl_ms_p95": 29.785625054500997, "itl_ms_p99": 34.03600002638996, "itl_ms_mean": 24.13889154097036, "total_s": 1.7547963340766728, "peak_mem_gb": 2.50075918, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.8352424793919281, "cache_misses": 1539, "cache_hits": 7802, "stall_s": 0.5556992036290467, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 1089, "installed": 1092, "late": 27, "dup": 6983, "predictor": "transition", "m": 8}, "ssd_bytes": 9300344832, "ssd_read_s": 1.6137432872783393}
	131	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "reasoning", "prompt": "Q: What is the smallest positive integer divisible by 2, 3, ", "ts": "2026-11-01T01:56:48", "prompt_tokens": 29, "gen_tokens": 64, "ttft_ms": 281.386791029945, "prompt_tps": 103.17887315821685, "gen_tps": 33.53500694342829, "decode_tps_wall": 33.01123075692832, "itl_ms_p50": 29.71879206597805, "itl_ms_p95": 39.29099999368191, "itl_ms_p99": 41.45362495910376, "itl_ms_mean": 30.292720903479864, "total_s": 2.1898282079491764, "peak_mem_gb": 2.594692576, "text": "The least common multiple (LCM) of 2, 3, 4, 5, and 6 is the smallest positive integer that is divisible by all of these numbers. The prime factorization of 2 is 2. The prime factorization of 3 is 3. T", "cache_hit_rate": 0.81472029399755, "cache_misses": 1815, "cache_hits": 7981, "stall_s": 0.6964973034337163, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 952, "installed": 950, "late": 21, "dup": 7224, "predictor": "transition", "m": 8}, "ssd_bytes": 9792258048, "ssd_read_s": 1.8330684097018093}
	132	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:56:49", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 234.94666605256498, "prompt_tps": 81.0111470370424, "gen_tps": 44.660945356483474, "decode_tps_wall": 43.96355397279772, "itl_ms_p50": 21.836041007190943, "itl_ms_p95": 29.456667019985616, "itl_ms_p99": 33.34979200735688, "itl_ms_mean": 22.746113760929024, "total_s": 1.6679518329910934, "peak_mem_gb": 2.476345452, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.8776878858846072, "cache_misses": 1149, "cache_hits": 8245, "stall_s": 0.4518067493336275, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 331, "installed": 334, "late": 2, "dup": 7765, "predictor": "transition", "m": 8}, "ssd_bytes": 5237637120, "ssd_read_s": 0.9524213743861765}
	133	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "reasoning", "prompt": "Q: A tank fills in 6 hours with pipe A and 9 hours with pipe", "ts": "2026-11-01T01:56:52", "prompt_tokens": 32, "gen_tokens": 64, "ttft_ms": 336.71350008808076, "prompt_tps": 95.1377192722343, "gen_tps": 31.770734437612802, "decode_tps_wall": 31.27466280226716, "itl_ms_p50": 31.3452499685809, "itl_ms_p95": 39.24054198432714, "itl_ms_p99": 50.376374972984195, "itl_ms_mean": 31.97476520602192, "total_s": 2.351123708067462, "peak_mem_gb": 2.769020348, "text": "If pipe A fills the tank in 6 hours, then the rate of pipe A is 1 / 6 of the tank per hour. If pipe B fills the tank in 9 hours, then the rate of pipe B is 1 / 9 of the tank per hour. The combined rat", "cache_hit_rate": 0.7899266503667481, "cache_misses": 2148, "cache_hits": 8077, "stall_s": 0.8536091272253543, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 1285, "installed": 1285, "late": 28, "dup": 6995, "predictor": "transition", "m": 8}, "ssd_bytes": 12149194752, "ssd_read_s": 2.438564954791218}
	134	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:56:54", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 265.23241598624736, "prompt_tps": 68.02827140271641, "gen_tps": 35.15052388005113, "decode_tps_wall": 34.60156932921162, "itl_ms_p50": 28.241583961062133, "itl_ms_p95": 37.11599996313453, "itl_ms_p99": 44.96050009038299, "itl_ms_mean": 28.900423286748783, "total_s": 2.0859590830514207, "peak_mem_gb": 2.463500932, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.8331013810084573, "cache_misses": 1559, "cache_hits": 7782, "stall_s": 0.6608698770869523, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 1048, "installed": 1047, "late": 32, "dup": 7024, "predictor": "transition", "m": 8}, "ssd_bytes": 9226027008, "ssd_read_s": 1.84350956894923}
	135	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:56:56", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 239.99174998607486, "prompt_tps": 58.42973215935761, "gen_tps": 33.53723287921878, "decode_tps_wall": 33.0134125450079, "itl_ms_p50": 30.094875022768974, "itl_ms_p95": 38.700582925230265, "itl_ms_p99": 46.733499970287085, "itl_ms_mean": 30.29071892027758, "total_s": 2.1483070419635624, "peak_mem_gb": 2.382968316, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.7921839576810968, "cache_misses": 1925, "cache_hits": 7338, "stall_s": 0.7835988189326599, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 2218, "installed": 2215, "late": 75, "dup": 5830, "predictor": "transition", "m": 8}, "ssd_bytes": 14661844992, "ssd_read_s": 2.7195660427678376}
	136	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "reasoning", "prompt": "Q: What is the smallest positive integer divisible by 2, 3, ", "ts": "2026-11-01T01:56:58", "prompt_tokens": 29, "gen_tokens": 64, "ttft_ms": 279.3260000180453, "prompt_tps": 103.95477738535655, "gen_tps": 42.96203901905487, "decode_tps_wall": 42.29114396375833, "itl_ms_p50": 24.486916954629123, "itl_ms_p95": 30.314750038087368, "itl_ms_p99": 32.26291702594608, "itl_ms_mean": 23.64561244446252, "total_s": 1.7689995840191841, "peak_mem_gb": 2.625694176, "text": "The least common multiple (LCM) of 2, 3, 4, 5, and 6 is the smallest positive integer that is divisible by all of these numbers. The prime factorization of 2 is 2. The prime factorization of 3 is 3. T", "cache_hit_rate": 0.8138015516537362, "cache_misses": 1824, "cache_hits": 7972, "stall_s": 0.6315422409679741, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 978, "installed": 979, "late": 28, "dup": 7198, "predictor": "transition", "m": 8}, "ssd_bytes": 9916121088, "ssd_read_s": 1.745780982542783}
	137	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:56:59", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 246.90954200923443, "prompt_tps": 97.33786021979385, "gen_tps": 47.21045821123445, "decode_tps_wall": 46.47319188429845, "itl_ms_p50": 21.33366698399186, "itl_ms_p95": 27.76787499897182, "itl_ms_p99": 31.163415987975895, "itl_ms_mean": 21.517781745864166, "total_s": 1.602529791998677, "peak_mem_gb": 2.487737752, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.8604796663190823, "cache_misses": 1338, "cache_hits": 8252, "stall_s": 0.4553022999316454, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 399, "installed": 402, "late": 10, "dup": 7753, "predictor": "transition", "m": 8}, "ssd_bytes": 6147145728, "ssd_read_s": 1.0380624668905511}
	138	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:57:01", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 200.95729094464332, "prompt_tps": 94.68678367729349, "gen_tps": 47.840757566434014, "decode_tps_wall": 47.093690131907024, "itl_ms_p50": 20.65054199192673, "itl_ms_p95": 28.029291075654328, "itl_ms_p99": 32.08879102021456, "itl_ms_mean": 21.234267206478215, "total_s": 1.5387161249527708, "peak_mem_gb": 2.47646014, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.8797104534809452, "cache_misses": 1130, "cache_hits": 8264, "stall_s": 0.4055240408051759, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 325, "installed": 325, "late": 3, "dup": 7771, "predictor": "transition", "m": 8}, "ssd_bytes": 5149163520, "ssd_read_s": 0.8579142205417156}
	139	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "reasoning", "prompt": "Q: What is the smallest positive integer divisible by 2, 3, ", "ts": "2026-11-01T01:57:03", "prompt_tokens": 29, "gen_tokens": 64, "ttft_ms": 282.0582080166787, "prompt_tps": 102.92806675704358, "gen_tps": 40.68185545852985, "decode_tps_wall": 40.04648252502245, "itl_ms_p50": 24.982333066873252, "itl_ms_p95": 30.436999979428947, "itl_ms_p99": 33.80570805165917, "itl_ms_mean": 24.970982142443223, "total_s": 1.8552300829906017, "peak_mem_gb": 2.575310304, "text": "The least common multiple (LCM) of 2, 3, 4, 5, and 6 is the smallest positive integer that is divisible by all of these numbers. The prime factorization of 2 is 2. The prime factorization of 3 is 3. T", "cache_hit_rate": 0.8164556962025317, "cache_misses": 1798, "cache_hits": 7998, "stall_s": 0.6233157982351258, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 935, "installed": 932, "late": 27, "dup": 7241, "predictor": "transition", "m": 8}, "ssd_bytes": 9671933952, "ssd_read_s": 1.675455072778277}
	140	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:57:04", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 204.97587497811764, "prompt_tps": 92.85052972381513, "gen_tps": 47.64513307839783, "decode_tps_wall": 46.901075079288255, "itl_ms_p50": 20.554249989800155, "itl_ms_p95": 28.412916930392385, "itl_ms_p99": 29.660750064067543, "itl_ms_mean": 21.32147287262515, "total_s": 1.548228665953502, "peak_mem_gb": 2.479884396, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.8776878858846072, "cache_misses": 1149, "cache_hits": 8245, "stall_s": 0.41758362122345716, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 332, "installed": 335, "late": 4, "dup": 7764, "predictor": "transition", "m": 8}, "ssd_bytes": 5241176064, "ssd_read_s": 0.9039581266697496}
	141	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:57:06", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 229.15124997962266, "prompt_tps": 78.66005497161797, "gen_tps": 41.96813046903063, "decode_tps_wall": 41.31264592547789, "itl_ms_p50": 24.0199159597978, "itl_ms_p95": 30.005750013515353, "itl_ms_p99": 33.73841708526015, "itl_ms_mean": 24.20566336525279, "total_s": 1.7541080419905484, "peak_mem_gb": 2.450329484, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.8359918638261428, "cache_misses": 1532, "cache_hits": 7809, "stall_s": 0.5551461785798892, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 1046, "installed": 1046, "late": 31, "dup": 7026, "predictor": "transition", "m": 8}, "ssd_bytes": 9123397632, "ssd_read_s": 1.5742781347362325}
	142	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:57:08", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 250.88149996008724, "prompt_tps": 95.79490313426386, "gen_tps": 46.538314452759366, "decode_tps_wall": 45.81149890277602, "itl_ms_p50": 21.1615419248119, "itl_ms_p95": 28.868083958514035, "itl_ms_p99": 31.803624937310815, "itl_ms_mean": 21.82858068281637, "total_s": 1.6260820829775184, "peak_mem_gb": 2.49481564, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.8584984358706986, "cache_misses": 1357, "cache_hits": 8233, "stall_s": 0.4639157010242343, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 411, "installed": 411, "late": 12, "dup": 7741, "predictor": "transition", "m": 8}, "ssd_bytes": 6256852992, "ssd_read_s": 1.0651860805228353}
	143	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:57:10", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 216.87979099806398, "prompt_tps": 64.6351479425608, "gen_tps": 35.75213519549105, "decode_tps_wall": 35.19379641972889, "itl_ms_p50": 28.78604200668633, "itl_ms_p95": 36.30054194945842, "itl_ms_p99": 41.629457962699234, "itl_ms_mean": 28.41409855514824, "total_s": 2.006967999972403, "peak_mem_gb": 2.384213624, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.7944510417791212, "cache_misses": 1904, "cache_hits": 7359, "stall_s": 0.7393082856433466, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 2211, "installed": 2208, "late": 75, "dup": 5837, "predictor": "transition", "m": 8}, "ssd_bytes": 14562754560, "ssd_read_s": 2.5910886902129278}
	144	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "reasoning", "prompt": "Q: A tank fills in 6 hours with pipe A and 9 hours with pipe", "ts": "2026-11-01T01:57:12", "prompt_tokens": 32, "gen_tokens": 64, "ttft_ms": 317.04166601412, "prompt_tps": 101.03133513549277, "gen_tps": 38.2362503826078, "decode_tps_wall": 37.63909005929106, "itl_ms_p50": 26.37291606515646, "itl_ms_p95": 31.681708991527557, "itl_ms_p99": 40.97141698002815, "itl_ms_mean": 26.568123682712514, "total_s": 1.9908334580250084, "peak_mem_gb": 2.802509684, "text": "If pipe A fills the tank in 6 hours, then the rate of pipe A is 1 / 6 of the tank per hour. If pipe B fills the tank in 9 hours, then the rate of pipe B is 1 / 9 of the tank per hour. The combined rat", "cache_hit_rate": 0.7899266503667481, "cache_misses": 2148, "cache_hits": 8077, "stall_s": 0.7414987228112295, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 1326, "installed": 1329, "late": 26, "dup": 6954, "predictor": "transition", "m": 8}, "ssd_bytes": 12294291456, "ssd_read_s": 2.2050836889538914}
	145	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "reasoning", "prompt": "Q: A tank fills in 6 hours with pipe A and 9 hours with pipe", "ts": "2026-11-01T01:57:19", "prompt_tokens": 32, "gen_tokens": 64, "ttft_ms": 526.2467919383198, "prompt_tps": 61.317796123046456, "gen_tps": 38.66756407637023, "decode_tps_wall": 38.06518490863805, "itl_ms_p50": 26.244333013892174, "itl_ms_p95": 31.38954099267721, "itl_ms_p99": 42.13595809414983, "itl_ms_mean": 26.270724873664598, "total_s": 2.1813024589791894, "peak_mem_gb": 2.837896772, "text": "If pipe A fills the tank in 6 hours, then the rate of pipe A is 1 / 6 of the tank per hour. If pipe B fills the tank in 9 hours, then the rate of pipe B is 1 / 9 of the tank per hour. The combined rat", "cache_hit_rate": 0.7598044009779951, "cache_misses": 2456, "cache_hits": 7769, "stall_s": 0.8876701433910057, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 1171, "installed": 1171, "late": 9, "dup": 7077, "predictor": "transition2", "m": 8}, "ssd_bytes": 12835749888, "ssd_read_s": 2.7585903315339237}
	146	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:57:21", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 246.94054201245308, "prompt_tps": 97.32212104346677, "gen_tps": 47.01821700718641, "decode_tps_wall": 46.283937078079965, "itl_ms_p50": 21.34562493301928, "itl_ms_p95": 28.52187503594905, "itl_ms_p99": 30.362334102392197, "itl_ms_mean": 21.605767856632905, "total_s": 1.6081039169803262, "peak_mem_gb": 2.491276696, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.8531803962460897, "cache_misses": 1408, "cache_hits": 8182, "stall_s": 0.4605353972874582, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 340, "installed": 340, "late": 4, "dup": 7812, "predictor": "transition2", "m": 8}, "ssd_bytes": 6186074112, "ssd_read_s": 1.021657341858372}
	147	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:57:23", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 220.02695803530514, "prompt_tps": 63.71451033254018, "gen_tps": 35.766929823448926, "decode_tps_wall": 35.2083281655901, "itl_ms_p50": 28.530958923511207, "itl_ms_p95": 37.50645800027996, "itl_ms_p99": 42.41429199464619, "itl_ms_mean": 28.402371032695687, "total_s": 2.0093763330951333, "peak_mem_gb": 2.377201272, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.7828997085177588, "cache_misses": 2011, "cache_hits": 7252, "stall_s": 0.7473375742556527, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 2012, "installed": 2007, "late": 55, "dup": 6036, "predictor": "transition2", "m": 8}, "ssd_bytes": 14237171712, "ssd_read_s": 2.4921520388452336}
	148	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:57:24", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 239.7110000019893, "prompt_tps": 75.1862686549162, "gen_tps": 42.304296297558295, "decode_tps_wall": 41.64360362790369, "itl_ms_p50": 24.068666039966047, "itl_ms_p95": 30.433833017013967, "itl_ms_p99": 32.19966706819832, "itl_ms_mean": 24.01329166743727, "total_s": 1.7525483750505373, "peak_mem_gb": 2.479410828, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.8257145915854833, "cache_misses": 1628, "cache_hits": 7713, "stall_s": 0.561451684916392, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 930, "installed": 934, "late": 32, "dup": 7142, "predictor": "transition2", "m": 8}, "ssd_bytes": 9052618752, "ssd_read_s": 1.5275575995910913}
	149	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "reasoning", "prompt": "Q: What is the smallest positive integer divisible by 2, 3, ", "ts": "2026-11-01T01:57:26", "prompt_tokens": 29, "gen_tokens": 64, "ttft_ms": 264.22424998600036, "prompt_tps": 109.88906891185485, "gen_tps": 42.673507455874024, "decode_tps_wall": 42.00696497735514, "itl_ms_p50": 24.298166972585022, "itl_ms_p95": 29.134584008716047, "itl_ms_p99": 32.61645801831037, "itl_ms_mean": 23.805576064328235, "total_s": 1.7639755420386791, "peak_mem_gb": 2.604460512, "text": "The least common multiple (LCM) of 2, 3, 4, 5, and 6 is the smallest positive integer that is divisible by all of these numbers. The prime factorization of 2 is 2. The prime factorization of 3 is 3. T", "cache_hit_rate": 0.8126786443446304, "cache_misses": 1835, "cache_hits": 7961, "stall_s": 0.6109398454427719, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 802, "installed": 803, "late": 22, "dup": 7374, "predictor": "transition2", "m": 8}, "ssd_bytes": 9332195328, "ssd_read_s": 1.5725603667087853}
	150	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:57:28", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 204.57962492946535, "prompt_tps": 93.01282012853194, "gen_tps": 49.11216022914396, "decode_tps_wall": 48.345155234164295, "itl_ms_p50": 20.258666947484016, "itl_ms_p95": 26.490667019970715, "itl_ms_p99": 29.32937501464039, "itl_ms_mean": 20.68459590534783, "total_s": 1.5077091669663787, "peak_mem_gb": 2.472921196, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.8742814562486694, "cache_misses": 1181, "cache_hits": 8213, "stall_s": 0.4042675201781094, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 240, "installed": 240, "late": 5, "dup": 7856, "predictor": "transition2", "m": 8}, "ssd_bytes": 5028839424, "ssd_read_s": 0.8246123712742701}
	151	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "reasoning", "prompt": "Q: A tank fills in 6 hours with pipe A and 9 hours with pipe", "ts": "2026-11-01T01:57:30", "prompt_tokens": 32, "gen_tokens": 64, "ttft_ms": 298.8200420513749, "prompt_tps": 107.19331348912667, "gen_tps": 38.62533536907778, "decode_tps_wall": 38.022065019358564, "itl_ms_p50": 26.01483406033367, "itl_ms_p95": 31.988083035685122, "itl_ms_p99": 39.039624971337616, "itl_ms_mean": 26.300517856956475, "total_s": 1.9557526670396328, "peak_mem_gb": 2.769020532, "text": "If pipe A fills the tank in 6 hours, then the rate of pipe A is 1 / 6 of the tank per hour. If pipe B fills the tank in 9 hours, then the rate of pipe B is 1 / 9 of the tank per hour. The combined rat", "cache_hit_rate": 0.7831784841075795, "cache_misses": 2217, "cache_hits": 8008, "stall_s": 0.7259550300659612, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 1062, "installed": 1062, "late": 10, "dup": 7218, "predictor": "transition2", "m": 8}, "ssd_bytes": 11604197376, "ssd_read_s": 2.00498158717528}
	152	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:57:31", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 235.45412498060614, "prompt_tps": 76.54730555019555, "gen_tps": 42.16143564477407, "decode_tps_wall": 41.502934349015625, "itl_ms_p50": 23.894666926935315, "itl_ms_p95": 30.284166103228927, "itl_ms_p99": 31.48029197473079, "itl_ms_mean": 24.094681874553245, "total_s": 1.7534190830774605, "peak_mem_gb": 2.468040332, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.8249652071512686, "cache_misses": 1635, "cache_hits": 7706, "stall_s": 0.5660235554678366, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 909, "installed": 908, "late": 29, "dup": 7163, "predictor": "transition2", "m": 8}, "ssd_bytes": 9003073536, "ssd_read_s": 1.5280321226455271}
	153	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:57:33", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 221.29662509541959, "prompt_tps": 63.34772376805383, "gen_tps": 35.47670939707307, "decode_tps_wall": 34.92266006485693, "itl_ms_p50": 28.77699991222471, "itl_ms_p95": 36.66425007395446, "itl_ms_p99": 43.07016695383936, "itl_ms_mean": 28.634703030721063, "total_s": 2.0252829160308465, "peak_mem_gb": 2.381919772, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.7798769297203929, "cache_misses": 2039, "cache_hits": 7224, "stall_s": 0.7604859068524092, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 2011, "installed": 2007, "late": 66, "dup": 6037, "predictor": "transition2", "m": 8}, "ssd_bytes": 14332723200, "ssd_read_s": 2.531662316294387}
	154	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "reasoning", "prompt": "Q: What is the smallest positive integer divisible by 2, 3, ", "ts": "2026-11-01T01:57:35", "prompt_tokens": 29, "gen_tokens": 64, "ttft_ms": 277.9518330935389, "prompt_tps": 104.46504005461514, "gen_tps": 41.49031706884608, "decode_tps_wall": 40.842303347646805, "itl_ms_p50": 24.697249988093972, "itl_ms_p95": 29.536165995523334, "itl_ms_p99": 32.76883403304964, "itl_ms_mean": 24.484417332883275, "total_s": 1.8204701250651851, "peak_mem_gb": 2.630277664, "text": "The least common multiple (LCM) of 2, 3, 4, 5, and 6 is the smallest positive integer that is divisible by all of these numbers. The prime factorization of 2 is 2. The prime factorization of 3 is 3. T", "cache_hit_rate": 0.810432829726419, "cache_misses": 1857, "cache_hits": 7939, "stall_s": 0.6138930609449744, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 810, "installed": 815, "late": 24, "dup": 7366, "predictor": "transition2", "m": 8}, "ssd_bytes": 9438363648, "ssd_read_s": 1.5867449769284576}
	155	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:57:37", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 242.76633304543793, "prompt_tps": 98.9883257059173, "gen_tps": 46.73172294766079, "decode_tps_wall": 46.00191204773223, "itl_ms_p50": 21.043957909569144, "itl_ms_p95": 29.10587494261563, "itl_ms_p99": 30.490832985378802, "itl_ms_mean": 21.73822685810072, "total_s": 1.6122746251057833, "peak_mem_gb": 2.487737752, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.8558915537017727, "cache_misses": 1382, "cache_hits": 8208, "stall_s": 0.45742425217758864, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 337, "installed": 337, "late": 5, "dup": 7815, "predictor": "transition2", "m": 8}, "ssd_bytes": 6083444736, "ssd_read_s": 1.007557698409073}
	156	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:57:38", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 202.49391696415842, "prompt_tps": 93.97415458566383, "gen_tps": 48.16626846601871, "decode_tps_wall": 47.4140570839877, "itl_ms_p50": 20.457208971492946, "itl_ms_p95": 28.0318750301376, "itl_ms_p99": 29.554625041782856, "itl_ms_mean": 21.090791666037624, "total_s": 1.5312137919245288, "peak_mem_gb": 2.476525676, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.8752395145837769, "cache_misses": 1172, "cache_hits": 8222, "stall_s": 0.40988460066728294, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 234, "installed": 234, "late": 6, "dup": 7862, "predictor": "transition2", "m": 8}, "ssd_bytes": 4975755264, "ssd_read_s": 0.8229692546883598}
	157	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "reasoning", "prompt": "Q: What is the smallest positive integer divisible by 2, 3, ", "ts": "2026-11-01T01:57:40", "prompt_tokens": 29, "gen_tokens": 64, "ttft_ms": 250.01370802056044, "prompt_tps": 116.12952296055184, "gen_tps": 41.96681521643548, "decode_tps_wall": 41.311333190355455, "itl_ms_p50": 24.232625029981136, "itl_ms_p95": 30.34666704479605, "itl_ms_p99": 33.43629091978073, "itl_ms_mean": 24.206432539762723, "total_s": 1.775018958025612, "peak_mem_gb": 2.579209696, "text": "The least common multiple (LCM) of 2, 3, 4, 5, and 6 is the smallest positive integer that is divisible by all of these numbers. The prime factorization of 2 is 2. The prime factorization of 3 is 3. T", "cache_hit_rate": 0.81472029399755, "cache_misses": 1815, "cache_hits": 7981, "stall_s": 0.6051625002874061, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 788, "installed": 788, "late": 20, "dup": 7388, "predictor": "transition2", "m": 8}, "ssd_bytes": 9211871232, "ssd_read_s": 1.5440555774839595}
	158	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:57:42", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 207.83304201904684, "prompt_tps": 91.55174209258396, "gen_tps": 48.629504350818316, "decode_tps_wall": 47.87007910969626, "itl_ms_p50": 20.405165967531502, "itl_ms_p95": 27.231291984207928, "itl_ms_p99": 29.89404194522649, "itl_ms_mean": 20.88987565089372, "total_s": 1.5238952080253512, "peak_mem_gb": 2.472921196, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.8742814562486694, "cache_misses": 1181, "cache_hits": 8213, "stall_s": 0.4092693250859156, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 240, "installed": 240, "late": 5, "dup": 7856, "predictor": "transition2", "m": 8}, "ssd_bytes": 5028839424, "ssd_read_s": 0.8329627900384367}
	159	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:57:44", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 230.12087505776435, "prompt_tps": 78.33347006973082, "gen_tps": 42.08169372605458, "decode_tps_wall": 41.42447254162209, "itl_ms_p50": 23.665833054110408, "itl_ms_p95": 30.684374971315265, "itl_ms_p99": 31.604208052158356, "itl_ms_mean": 24.14031944511133, "total_s": 1.7509610000997782, "peak_mem_gb": 2.453983116, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.8271063055347393, "cache_misses": 1615, "cache_hits": 7726, "stall_s": 0.564033625065349, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 906, "installed": 905, "late": 32, "dup": 7166, "predictor": "transition2", "m": 8}, "ssd_bytes": 8921677824, "ssd_read_s": 1.512469703797251}
	160	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:57:45", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 245.58254098519683, "prompt_tps": 97.85678351701762, "gen_tps": 46.77768021154988, "decode_tps_wall": 46.04708745056892, "itl_ms_p50": 21.15079201757908, "itl_ms_p95": 28.85829098522663, "itl_ms_p99": 29.13104195613414, "itl_ms_mean": 21.71690014213146, "total_s": 1.613747249939479, "peak_mem_gb": 2.491276696, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.8530761209593326, "cache_misses": 1409, "cache_hits": 8181, "stall_s": 0.45696721330750734, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 345, "installed": 346, "late": 4, "dup": 7807, "predictor": "transition2", "m": 8}, "ssd_bytes": 6207307776, "ssd_read_s": 1.0156202514190227}
	161	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:57:47", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 210.71124996524304, "prompt_tps": 66.53070136420489, "gen_tps": 36.08053993946833, "decode_tps_wall": 35.516990081721616, "itl_ms_p50": 28.451750054955482, "itl_ms_p95": 35.19320790655911, "itl_ms_p99": 39.47437496390194, "itl_ms_mean": 28.155539016653265, "total_s": 1.9845102080143988, "peak_mem_gb": 2.3688782, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.7826837957465184, "cache_misses": 2013, "cache_hits": 7250, "stall_s": 0.7126202502986416, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 2012, "installed": 2007, "late": 46, "dup": 6036, "predictor": "transition2", "m": 8}, "ssd_bytes": 14244249600, "ssd_read_s": 2.3624545776983723}
	162	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition2", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition2 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "reasoning", "prompt": "Q: A tank fills in 6 hours with pipe A and 9 hours with pipe", "ts": "2026-11-01T01:57:49", "prompt_tokens": 32, "gen_tokens": 64, "ttft_ms": 313.0920419935137, "prompt_tps": 102.30909516845405, "gen_tps": 38.29381507229609, "decode_tps_wall": 37.695758040872974, "itl_ms_p50": 26.349458028562367, "itl_ms_p95": 32.40716701839119, "itl_ms_p99": 42.20050002913922, "itl_ms_mean": 26.528183858664264, "total_s": 1.9843676250893623, "peak_mem_gb": 2.796152436, "text": "If pipe A fills the tank in 6 hours, then the rate of pipe A is 1 / 6 of the tank per hour. If pipe B fills the tank in 9 hours, then the rate of pipe B is 1 / 9 of the tank per hour. The combined rat", "cache_hit_rate": 0.7817114914425428, "cache_misses": 2232, "cache_hits": 7993, "stall_s": 0.7305938890203834, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 1084, "installed": 1089, "late": 10, "dup": 7196, "predictor": "transition2", "m": 8}, "ssd_bytes": 11735138304, "ssd_read_s": 2.0255970461294055}
	163	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "reasoning", "prompt": "Q: A tank fills in 6 hours with pipe A and 9 hours with pipe", "ts": "2026-11-01T01:57:57", "prompt_tokens": 32, "gen_tokens": 64, "ttft_ms": 530.6949169607833, "prompt_tps": 60.711212878555976, "gen_tps": 37.61759079615561, "decode_tps_wall": 37.03165427681308, "itl_ms_p50": 26.682166033424437, "itl_ms_p95": 32.583207939751446, "itl_ms_p99": 40.29008303768933, "itl_ms_mean": 27.003924602583524, "total_s": 2.2319421669235453, "peak_mem_gb": 2.834357828, "text": "If pipe A fills the tank in 6 hours, then the rate of pipe A is 1 / 6 of the tank per hour. If pipe B fills the tank in 9 hours, then the rate of pipe B is 1 / 9 of the tank per hour. The combined rat", "cache_hit_rate": 0.7527628361858191, "cache_misses": 2528, "cache_hits": 7697, "stall_s": 0.9040113523369655, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 1104, "installed": 1104, "late": 10, "dup": 7104, "predictor": "transition3", "m": 8}, "ssd_bytes": 12853444608, "ssd_read_s": 2.7384828088106588}
	164	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:57:58", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 250.18979201558977, "prompt_tps": 96.07380080962646, "gen_tps": 47.3216806774956, "decode_tps_wall": 46.58264682391963, "itl_ms_p50": 21.394332987256348, "itl_ms_p95": 28.589375084266067, "itl_ms_p99": 29.91420798934996, "itl_ms_mean": 21.46722155527049, "total_s": 1.6026247499976307, "peak_mem_gb": 2.487737752, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.8495307612095934, "cache_misses": 1443, "cache_hits": 8147, "stall_s": 0.46130490279756486, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 277, "installed": 277, "late": 2, "dup": 7875, "predictor": "transition3", "m": 8}, "ssd_bytes": 6086983680, "ssd_read_s": 0.9874504569452256}
	165	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:58:00", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 218.65341695956886, "prompt_tps": 64.1226758520243, "gen_tps": 35.4168779949671, "decode_tps_wall": 34.86369104309848, "itl_ms_p50": 28.66358298342675, "itl_ms_p95": 36.35525004938245, "itl_ms_p99": 42.23512497264892, "itl_ms_mean": 28.68313623947047, "total_s": 2.0256910000462085, "peak_mem_gb": 2.363045368, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.7735075029688006, "cache_misses": 2098, "cache_hits": 7165, "stall_s": 0.7597819989314303, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 1827, "installed": 1824, "late": 38, "dup": 6221, "predictor": "transition3", "m": 8}, "ssd_bytes": 13890355200, "ssd_read_s": 2.45023709803354}
	166	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:58:02", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 241.28454201854765, "prompt_tps": 74.71593684323653, "gen_tps": 42.30884779451468, "decode_tps_wall": 41.64811623298193, "itl_ms_p50": 23.504958022385836, "itl_ms_p95": 31.18704108055681, "itl_ms_p99": 33.52537506725639, "itl_ms_mean": 24.010689809016647, "total_s": 1.7539579999865964, "peak_mem_gb": 2.48298254, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.8210041751418478, "cache_misses": 1672, "cache_hits": 7669, "stall_s": 0.5712962452089414, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 807, "installed": 810, "late": 19, "dup": 7265, "predictor": "transition3", "m": 8}, "ssd_bytes": 8773042176, "ssd_read_s": 1.5135142977815121}
	167	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "reasoning", "prompt": "Q: What is the smallest positive integer divisible by 2, 3, ", "ts": "2026-11-01T01:58:04", "prompt_tokens": 29, "gen_tokens": 64, "ttft_ms": 288.3212079759687, "prompt_tps": 100.69813039139494, "gen_tps": 40.75601669777822, "decode_tps_wall": 40.11950729849327, "itl_ms_p50": 25.002750102430582, "itl_ms_p95": 31.589207937940955, "itl_ms_p99": 32.96391700860113, "itl_ms_mean": 24.925530429870363, "total_s": 1.8586296250578016, "peak_mem_gb": 2.597382624, "text": "The least common multiple (LCM) of 2, 3, 4, 5, and 6 is the smallest positive integer that is divisible by all of these numbers. The prime factorization of 2 is 2. The prime factorization of 3 is 3. T", "cache_hit_rate": 0.8037974683544303, "cache_misses": 1922, "cache_hits": 7874, "stall_s": 0.6321207407163456, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 761, "installed": 761, "late": 15, "dup": 7415, "predictor": "transition3", "m": 8}, "ssd_bytes": 9494986752, "ssd_read_s": 1.616611955803819}
	168	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:58:05", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 204.7907910309732, "prompt_tps": 92.92915099835191, "gen_tps": 47.14587731748471, "decode_tps_wall": 46.4096218191413, "itl_ms_p50": 20.818791002966464, "itl_ms_p95": 28.576500015333295, "itl_ms_p99": 32.01054106466472, "itl_ms_mean": 21.547255952591225, "total_s": 1.5622679160442203, "peak_mem_gb": 2.470075244, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.8687460080902704, "cache_misses": 1233, "cache_hits": 8161, "stall_s": 0.43462368892505765, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 193, "installed": 193, "late": 1, "dup": 7903, "predictor": "transition3", "m": 8}, "ssd_bytes": 5046534144, "ssd_read_s": 0.8415085544111207}
	169	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "reasoning", "prompt": "Q: A tank fills in 6 hours with pipe A and 9 hours with pipe", "ts": "2026-11-01T01:58:07", "prompt_tokens": 32, "gen_tokens": 64, "ttft_ms": 306.4874999690801, "prompt_tps": 104.52459468294853, "gen_tps": 37.95241014117154, "decode_tps_wall": 37.35966681644194, "itl_ms_p50": 26.526625035330653, "itl_ms_p95": 32.675832975655794, "itl_ms_p99": 37.84616698976606, "itl_ms_mean": 26.766834000775976, "total_s": 1.9927980420179665, "peak_mem_gb": 2.769020532, "text": "If pipe A fills the tank in 6 hours, then the rate of pipe A is 1 / 6 of the tank per hour. If pipe B fills the tank in 9 hours, then the rate of pipe B is 1 / 9 of the tank per hour. The combined rat", "cache_hit_rate": 0.7764303178484108, "cache_misses": 2286, "cache_hits": 7939, "stall_s": 0.7414712503086776, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 1000, "installed": 1000, "late": 10, "dup": 7280, "predictor": "transition3", "m": 8}, "ssd_bytes": 11628969984, "ssd_read_s": 2.018930197576992}
	170	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:58:09", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 237.16158303432167, "prompt_tps": 76.00340327565287, "gen_tps": 41.39204240567406, "decode_tps_wall": 40.74555418429973, "itl_ms_p50": 24.041750002652407, "itl_ms_p95": 31.232250039465725, "itl_ms_p99": 35.08116595912725, "itl_ms_mean": 24.542554887750793, "total_s": 1.7833425409626216, "peak_mem_gb": 2.4611099, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.8191842415158976, "cache_misses": 1689, "cache_hits": 7652, "stall_s": 0.5718261183938012, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 780, "installed": 780, "late": 18, "dup": 7292, "predictor": "transition3", "m": 8}, "ssd_bytes": 8737652736, "ssd_read_s": 1.4828760590171441}
	171	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:58:11", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 225.4374170443043, "prompt_tps": 62.18391002055594, "gen_tps": 34.95960197813111, "decode_tps_wall": 34.41356027824513, "itl_ms_p50": 29.0010420139879, "itl_ms_p95": 36.68762499000877, "itl_ms_p99": 43.111167033202946, "itl_ms_mean": 29.058312825370756, "total_s": 2.056111125042662, "peak_mem_gb": 2.393716288, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.7700528986289539, "cache_misses": 2130, "cache_hits": 7133, "stall_s": 0.7766643203794956, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 1823, "installed": 1820, "late": 53, "dup": 6225, "predictor": "transition3", "m": 8}, "ssd_bytes": 13989445632, "ssd_read_s": 2.487749135354534}
	172	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "reasoning", "prompt": "Q: What is the smallest positive integer divisible by 2, 3, ", "ts": "2026-11-01T01:58:13", "prompt_tokens": 29, "gen_tokens": 64, "ttft_ms": 271.2649169843644, "prompt_tps": 107.03557416160317, "gen_tps": 40.652507686234756, "decode_tps_wall": 40.01769460473437, "itl_ms_p50": 25.02000005915761, "itl_ms_p95": 31.01287502795458, "itl_ms_p99": 33.342791022732854, "itl_ms_mean": 24.988945762050292, "total_s": 1.845568499993533, "peak_mem_gb": 2.615958048, "text": "The least common multiple (LCM) of 2, 3, 4, 5, and 6 is the smallest positive integer that is divisible by all of these numbers. The prime factorization of 2 is 2. The prime factorization of 3 is 3. T", "cache_hit_rate": 0.8019599836668028, "cache_misses": 1940, "cache_hits": 7856, "stall_s": 0.6440316166263074, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 776, "installed": 779, "late": 13, "dup": 7400, "predictor": "transition3", "m": 8}, "ssd_bytes": 9611771904, "ssd_read_s": 1.659277145168744}
	173	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:58:15", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 247.1101249102503, "prompt_tps": 97.25624293572443, "gen_tps": 46.4478535200105, "decode_tps_wall": 45.72248326328945, "itl_ms_p50": 21.429833956062794, "itl_ms_p95": 29.485541977919638, "itl_ms_p99": 31.328708049841225, "itl_ms_mean": 21.871078048004872, "total_s": 1.6249880419345573, "peak_mem_gb": 2.484198808, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.8527632950990616, "cache_misses": 1412, "cache_hits": 8178, "stall_s": 0.46597758354619145, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 278, "installed": 278, "late": 1, "dup": 7874, "predictor": "transition3", "m": 8}, "ssd_bytes": 5980815360, "ssd_read_s": 0.9928303242195398}
	174	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:58:16", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 206.4854169730097, "prompt_tps": 92.16865543735817, "gen_tps": 46.84942783945552, "decode_tps_wall": 46.11781202715002, "itl_ms_p50": 20.866874954663217, "itl_ms_p95": 27.801500051282346, "itl_ms_p99": 32.90587500669062, "itl_ms_mean": 21.683595904577817, "total_s": 1.5725519589614123, "peak_mem_gb": 2.469382252, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.8703427719821163, "cache_misses": 1218, "cache_hits": 8176, "stall_s": 0.4356100804870948, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 186, "installed": 186, "late": 1, "dup": 7910, "predictor": "transition3", "m": 8}, "ssd_bytes": 4968677376, "ssd_read_s": 0.8366910853656009}
	175	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "reasoning", "prompt": "Q: What is the smallest positive integer divisible by 2, 3, ", "ts": "2026-11-01T01:58:18", "prompt_tokens": 29, "gen_tokens": 64, "ttft_ms": 259.0983749832958, "prompt_tps": 112.0695527214353, "gen_tps": 40.55835547103742, "decode_tps_wall": 39.924911584296126, "itl_ms_p50": 25.12725000269711, "itl_ms_p95": 31.321458984166384, "itl_ms_p99": 32.69941604230553, "itl_ms_mean": 25.047018523475835, "total_s": 1.8370605419622734, "peak_mem_gb": 2.57750576, "text": "The least common multiple (LCM) of 2, 3, 4, 5, and 6 is the smallest positive integer that is divisible by all of these numbers. The prime factorization of 2 is 2. The prime factorization of 3 is 3. T", "cache_hit_rate": 0.8061453654552879, "cache_misses": 1899, "cache_hits": 7897, "stall_s": 0.635601923102513, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 750, "installed": 750, "late": 14, "dup": 7426, "predictor": "transition3", "m": 8}, "ssd_bytes": 9374662656, "ssd_read_s": 1.6130917553091422}
	176	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:58:20", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 208.04695901460946, "prompt_tps": 91.47820347533639, "gen_tps": 47.61138101982919, "decode_tps_wall": 46.86780188914181, "itl_ms_p50": 20.600584102794528, "itl_ms_p95": 27.790957945398986, "itl_ms_p99": 30.509707983583212, "itl_ms_mean": 21.33660977669356, "total_s": 1.5522533749463037, "peak_mem_gb": 2.472921196, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.8687460080902704, "cache_misses": 1233, "cache_hits": 8161, "stall_s": 0.4324378327000886, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 193, "installed": 193, "late": 1, "dup": 7903, "predictor": "transition3", "m": 8}, "ssd_bytes": 5046534144, "ssd_read_s": 0.8399806651286781}
	177	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:58:21", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 235.4157919762656, "prompt_tps": 76.56716786281245, "gen_tps": 42.18419781437584, "decode_tps_wall": 41.52534796820358, "itl_ms_p50": 23.598375031724572, "itl_ms_p95": 31.155375065281987, "itl_ms_p99": 33.42212503775954, "itl_ms_mean": 24.08167658861549, "total_s": 1.7525614170590416, "peak_mem_gb": 2.453983116, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.8219676694144096, "cache_misses": 1663, "cache_hits": 7678, "stall_s": 0.5653333888622001, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 781, "installed": 781, "late": 24, "dup": 7291, "predictor": "transition3", "m": 8}, "ssd_bytes": 8649179136, "ssd_read_s": 1.4717162100132555}
	178	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:58:23", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 248.1124580372125, "prompt_tps": 96.84620965944319, "gen_tps": 46.84358843337115, "decode_tps_wall": 46.1120736147709, "itl_ms_p50": 21.135417046025395, "itl_ms_p95": 28.927042032592, "itl_ms_p99": 30.076291994191706, "itl_ms_mean": 21.68629431749679, "total_s": 1.6143490000395104, "peak_mem_gb": 2.491276696, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.8495307612095934, "cache_misses": 1443, "cache_hits": 8147, "stall_s": 0.4654328323667869, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 284, "installed": 284, "late": 1, "dup": 7868, "predictor": "transition3", "m": 8}, "ssd_bytes": 6111756288, "ssd_read_s": 1.0083087478997186}
	179	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:58:25", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 219.43383407779038, "prompt_tps": 63.89447060001168, "gen_tps": 35.38798280098549, "decode_tps_wall": 34.83523820612727, "itl_ms_p50": 28.87949999421835, "itl_ms_p95": 36.70649998821318, "itl_ms_p99": 41.11674998421222, "itl_ms_mean": 28.706564142974834, "total_s": 2.027947375085205, "peak_mem_gb": 2.363045368, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.7730756774263198, "cache_misses": 2102, "cache_hits": 7161, "stall_s": 0.7645913262385875, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 1827, "installed": 1824, "late": 51, "dup": 6221, "predictor": "transition3", "m": 8}, "ssd_bytes": 13904510976, "ssd_read_s": 2.467960340087302}
	180	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "transition3", "m": 8, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor transition3 --m 8 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "reasoning", "prompt": "Q: A tank fills in 6 hours with pipe A and 9 hours with pipe", "ts": "2026-11-01T01:58:27", "prompt_tokens": 32, "gen_tokens": 64, "ttft_ms": 325.7768329931423, "prompt_tps": 98.32125429356284, "gen_tps": 37.235080040657586, "decode_tps_wall": 36.65352092509186, "itl_ms_p50": 27.147167013026774, "itl_ms_p95": 33.30558305606246, "itl_ms_p99": 39.70529208891094, "itl_ms_mean": 27.282508603844143, "total_s": 2.044574875035323, "peak_mem_gb": 2.79700422, "text": "If pipe A fills the tank in 6 hours, then the rate of pipe A is 1 / 6 of the tank per hour. If pipe B fills the tank in 9 hours, then the rate of pipe B is 1 / 9 of the tank per hour. The combined rat", "cache_hit_rate": 0.7754523227383863, "cache_misses": 2296, "cache_hits": 7929, "stall_s": 0.7595669905422255, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 1027, "installed": 1030, "late": 16, "dup": 7253, "predictor": "transition3", "m": 8}, "ssd_bytes": 11759910912, "ssd_read_s": 2.0804126370931044}
	181	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "prev_rank", "m": 16, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor prev_rank --m 16 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "reasoning", "prompt": "Q: A tank fills in 6 hours with pipe A and 9 hours with pipe", "ts": "2026-11-01T01:58:33", "prompt_tokens": 32, "gen_tokens": 64, "ttft_ms": 500.6392920622602, "prompt_tps": 64.60762933221953, "gen_tps": 29.399985349876175, "decode_tps_wall": 28.941701880083457, "itl_ms_p50": 33.084416994825006, "itl_ms_p95": 44.03874999843538, "itl_ms_p99": 54.268999956548214, "itl_ms_mean": 34.55221825390167, "total_s": 2.6774290420580655, "peak_mem_gb": 2.805456452, "text": "If pipe A fills the tank in 6 hours, then the rate of pipe A is 1 / 6 of the tank per hour. If pipe B fills the tank in 9 hours, then the rate of pipe B is 1 / 9 of the tank per hour. The combined rat", "cache_hit_rate": 0.7409290953545232, "cache_misses": 2649, "cache_hits": 7576, "stall_s": 1.0457963303197175, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 6833, "installed": 6692, "late": 136, "dup": 9807, "predictor": "prev_rank", "m": 16}, "ssd_bytes": 33343930368, "ssd_read_s": 5.073906094650738}
	182	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "prev_rank", "m": 16, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor prev_rank --m 16 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:58:36", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 261.2464999547228, "prompt_tps": 92.07228825285056, "gen_tps": 35.33606704753488, "decode_tps_wall": 34.78411063980914, "itl_ms_p50": 27.58912497665733, "itl_ms_p95": 38.29137491993606, "itl_ms_p99": 45.17758300062269, "itl_ms_mean": 28.748758602887392, "total_s": 2.0724182919366285, "peak_mem_gb": 2.926215792, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.8583941605839416, "cache_misses": 1358, "cache_hits": 8232, "stall_s": 0.569021784584038, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 5858, "installed": 5904, "late": 97, "dup": 10782, "predictor": "prev_rank", "m": 16}, "ssd_bytes": 25434390528, "ssd_read_s": 3.4979454677086323}
	183	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "prev_rank", "m": 16, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor prev_rank --m 16 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:58:38", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 237.22312494646758, "prompt_tps": 59.15862697198368, "gen_tps": 28.355220546618618, "decode_tps_wall": 27.912391798049345, "itl_ms_p50": 35.38383392151445, "itl_ms_p95": 44.570500031113625, "itl_ms_p99": 56.429959018714726, "itl_ms_mean": 35.826381602664554, "total_s": 2.4942851659143344, "peak_mem_gb": 2.616736684, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.7543992227140235, "cache_misses": 2275, "cache_hits": 6988, "stall_s": 0.913522737333551, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 7008, "installed": 6971, "late": 175, "dup": 9632, "predictor": "prev_rank", "m": 16}, "ssd_bytes": 32742309888, "ssd_read_s": 4.747185734449886}
	184	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "prev_rank", "m": 16, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor prev_rank --m 16 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:58:41", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 266.4266660576686, "prompt_tps": 67.70304582379634, "gen_tps": 28.715373458201757, "decode_tps_wall": 28.266818869885107, "itl_ms_p50": 34.86925002653152, "itl_ms_p95": 44.409708003513515, "itl_ms_p99": 46.300375019200146, "itl_ms_mean": 35.37716800051312, "total_s": 2.4951882500899956, "peak_mem_gb": 2.7801883, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.7618028048388823, "cache_misses": 2225, "cache_hits": 7116, "stall_s": 0.9020266789011657, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 6843, "installed": 6849, "late": 214, "dup": 9797, "predictor": "prev_rank", "m": 16}, "ssd_bytes": 32069910528, "ssd_read_s": 4.624020031420514}
	185	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "prev_rank", "m": 16, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor prev_rank --m 16 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "reasoning", "prompt": "Q: What is the smallest positive integer divisible by 2, 3, ", "ts": "2026-11-01T01:58:43", "prompt_tokens": 29, "gen_tokens": 64, "ttft_ms": 255.62995800282806, "prompt_tps": 113.72955992030182, "gen_tps": 30.8851192655119, "decode_tps_wall": 30.402738563273047, "itl_ms_p50": 33.51187508087605, "itl_ms_p95": 43.52287494111806, "itl_ms_p99": 47.02941689174622, "itl_ms_mean": 32.89177380908754, "total_s": 2.327811707975343, "peak_mem_gb": 2.985711408, "text": "The least common multiple (LCM) of 2, 3, 4, 5, and 6 is the smallest positive integer that is divisible by all of these numbers. The prime factorization of 2 is 2. The prime factorization of 3 is 3. T", "cache_hit_rate": 0.7978766843609637, "cache_misses": 1980, "cache_hits": 7816, "stall_s": 0.8053798165638, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 6488, "installed": 6471, "late": 136, "dup": 10152, "predictor": "prev_rank", "m": 16}, "ssd_bytes": 29932388352, "ssd_read_s": 4.2504302660236135}
	186	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "prev_rank", "m": 16, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor prev_rank --m 16 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:58:45", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 230.87162501178682, "prompt_tps": 82.47392821817878, "gen_tps": 34.96123161300227, "decode_tps_wall": 34.41512295969738, "itl_ms_p50": 27.76787499897182, "itl_ms_p95": 41.148124961182475, "itl_ms_p99": 44.63662498164922, "itl_ms_mean": 29.056993379656756, "total_s": 2.0614622079301625, "peak_mem_gb": 2.785869828, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.8629976580796253, "cache_misses": 1287, "cache_hits": 8107, "stall_s": 0.5752399601042271, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 5915, "installed": 5938, "late": 104, "dup": 10725, "predictor": "prev_rank", "m": 16}, "ssd_bytes": 25544097792, "ssd_read_s": 3.5176863505039364}
	187	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "prev_rank", "m": 16, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor prev_rank --m 16 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "reasoning", "prompt": "Q: A tank fills in 6 hours with pipe A and 9 hours with pipe", "ts": "2026-11-01T01:58:47", "prompt_tokens": 32, "gen_tokens": 64, "ttft_ms": 329.61108395829797, "prompt_tps": 97.24129752398822, "gen_tps": 29.483137569646768, "decode_tps_wall": 29.02276827374092, "itl_ms_p50": 33.503124956041574, "itl_ms_p95": 44.09025004133582, "itl_ms_p99": 47.23362496588379, "itl_ms_mean": 34.455707001071126, "total_s": 2.500320625025779, "peak_mem_gb": 3.057408768, "text": "If pipe A fills the tank in 6 hours, then the rate of pipe A is 1 / 6 of the tank per hour. If pipe B fills the tank in 9 hours, then the rate of pipe B is 1 / 9 of the tank per hour. The combined rat", "cache_hit_rate": 0.776919315403423, "cache_misses": 2281, "cache_hits": 7944, "stall_s": 0.9247209813911468, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 6707, "installed": 6689, "late": 162, "dup": 9933, "predictor": "prev_rank", "m": 16}, "ssd_bytes": 31984975872, "ssd_read_s": 4.602123213931918}
	188	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "prev_rank", "m": 16, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor prev_rank --m 16 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:58:50", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 263.6612090282142, "prompt_tps": 68.40654715527157, "gen_tps": 28.099618415991355, "decode_tps_wall": 27.66067320788268, "itl_ms_p50": 35.72041599545628, "itl_ms_p95": 45.54595809895545, "itl_ms_p99": 56.83133297134191, "itl_ms_mean": 36.15241004745402, "total_s": 2.5412630420178175, "peak_mem_gb": 2.893237372, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.7609463654854941, "cache_misses": 2233, "cache_hits": 7108, "stall_s": 0.9181513079674914, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 6900, "installed": 6913, "late": 217, "dup": 9740, "predictor": "prev_rank", "m": 16}, "ssd_bytes": 32112377856, "ssd_read_s": 4.674717596964911}
	189	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "prev_rank", "m": 16, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor prev_rank --m 16 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:58:53", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 243.5570830712095, "prompt_tps": 57.60687653338281, "gen_tps": 21.666581909471056, "decode_tps_wall": 21.32817394693354, "itl_ms_p50": 41.80737503338605, "itl_ms_p95": 92.98520907759666, "itl_ms_p99": 100.21262499503791, "itl_ms_mean": 46.88633928474571, "total_s": 3.1973964580101892, "peak_mem_gb": 2.64143886, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.7498650545179747, "cache_misses": 2317, "cache_hits": 6946, "stall_s": 0.9944427234586328, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 7330, "installed": 7331, "late": 68, "dup": 9310, "predictor": "prev_rank", "m": 16}, "ssd_bytes": 34122498048, "ssd_read_s": 5.207796155358665}
	190	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "prev_rank", "m": 16, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor prev_rank --m 16 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "reasoning", "prompt": "Q: What is the smallest positive integer divisible by 2, 3, ", "ts": "2026-11-01T01:58:56", "prompt_tokens": 29, "gen_tokens": 64, "ttft_ms": 391.23033289797604, "prompt_tps": 74.2604917942824, "gen_tps": 25.224987267726544, "decode_tps_wall": 24.830876603496257, "itl_ms_p50": 40.87741696275771, "itl_ms_p95": 59.650499955751, "itl_ms_p99": 69.46666701696813, "itl_ms_mean": 40.272440476756955, "total_s": 2.9283940829336643, "peak_mem_gb": 2.985253408, "text": "The least common multiple (LCM) of 2, 3, 4, 5, and 6 is the smallest positive integer that is divisible by all of these numbers. The prime factorization of 2 is 2. The prime factorization of 3 is 3. T", "cache_hit_rate": 0.7932829726418946, "cache_misses": 2025, "cache_hits": 7771, "stall_s": 0.9602265820140019, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 6554, "installed": 6535, "late": 103, "dup": 10086, "predictor": "prev_rank", "m": 16}, "ssd_bytes": 30293360640, "ssd_read_s": 4.962919001583941}
	191	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "prev_rank", "m": 16, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor prev_rank --m 16 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:58:58", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 279.0375000331551, "prompt_tps": 86.19413802457292, "gen_tps": 32.43897022737971, "decode_tps_wall": 31.932382419124952, "itl_ms_p50": 29.314124956727028, "itl_ms_p95": 42.27933392394334, "itl_ms_p99": 49.144042073749006, "itl_ms_mean": 31.316172619838092, "total_s": 2.2519563750829548, "peak_mem_gb": 2.85726476, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.8607924921793535, "cache_misses": 1335, "cache_hits": 8255, "stall_s": 0.6195322249550372, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 5867, "installed": 5917, "late": 119, "dup": 10773, "predictor": "prev_rank", "m": 16}, "ssd_bytes": 25671499776, "ssd_read_s": 3.7687663243850693}
	192	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "prev_rank", "m": 16, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor prev_rank --m 16 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:59:01", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 238.3127500070259, "prompt_tps": 79.94098405491975, "gen_tps": 32.777170610735205, "decode_tps_wall": 32.26559880647992, "itl_ms_p50": 29.837792040780187, "itl_ms_p95": 45.05579103715718, "itl_ms_p99": 47.20070806797594, "itl_ms_mean": 30.992761237679847, "total_s": 2.1908567079808563, "peak_mem_gb": 2.717012376, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.85863317010858, "cache_misses": 1328, "cache_hits": 8066, "stall_s": 0.5923730889335275, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 5842, "installed": 5818, "late": 100, "dup": 10798, "predictor": "prev_rank", "m": 16}, "ssd_bytes": 25303449600, "ssd_read_s": 3.6563023164635524}
	193	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "prev_rank", "m": 16, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor prev_rank --m 16 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "reasoning", "prompt": "Q: What is the smallest positive integer divisible by 2, 3, ", "ts": "2026-11-01T01:59:03", "prompt_tokens": 29, "gen_tokens": 64, "ttft_ms": 284.04720895923674, "prompt_tps": 102.30366980706786, "gen_tps": 28.65849259889012, "decode_tps_wall": 28.210895248779455, "itl_ms_p50": 36.56758298166096, "itl_ms_p95": 45.93091597780585, "itl_ms_p99": 50.75000005308539, "itl_ms_mean": 35.44729761964094, "total_s": 2.5172269589966163, "peak_mem_gb": 2.941917728, "text": "The least common multiple (LCM) of 2, 3, 4, 5, and 6 is the smallest positive integer that is divisible by all of these numbers. The prime factorization of 2 is 2. The prime factorization of 3 is 3. T", "cache_hit_rate": 0.7983870967741935, "cache_misses": 1975, "cache_hits": 7821, "stall_s": 0.8224571474129334, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 6510, "installed": 6484, "late": 119, "dup": 10130, "predictor": "prev_rank", "m": 16}, "ssd_bytes": 29907615744, "ssd_read_s": 4.474754717899486}
	194	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "prev_rank", "m": 16, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor prev_rank --m 16 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:59:05", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 245.52683299407363, "prompt_tps": 77.56659956353012, "gen_tps": 32.47101369578315, "decode_tps_wall": 31.96403250418268, "itl_ms_p50": 29.763375059701502, "itl_ms_p95": 42.682374943979084, "itl_ms_p99": 47.419541981071234, "itl_ms_mean": 31.28516403145142, "total_s": 2.216492166975513, "peak_mem_gb": 2.789474308, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.8635299127102406, "cache_misses": 1282, "cache_hits": 8112, "stall_s": 0.5976252818945795, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 5908, "installed": 5933, "late": 109, "dup": 10732, "predictor": "prev_rank", "m": 16}, "ssd_bytes": 25526403072, "ssd_read_s": 3.7437403085641563}
	195	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "prev_rank", "m": 16, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor prev_rank --m 16 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:59:08", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 261.61037501879036, "prompt_tps": 68.9416892824878, "gen_tps": 27.464052531956753, "decode_tps_wall": 27.0350746300655, "itl_ms_p50": 36.600374965928495, "itl_ms_p95": 46.41845799051225, "itl_ms_p99": 47.61833290103823, "itl_ms_mean": 36.98898611095039, "total_s": 2.591916500008665, "peak_mem_gb": 2.727021948, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.7674767155550798, "cache_misses": 2172, "cache_hits": 7169, "stall_s": 0.9160170473624021, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 6864, "installed": 6853, "late": 224, "dup": 9776, "predictor": "prev_rank", "m": 16}, "ssd_bytes": 31938969600, "ssd_read_s": 4.686025946168229}
	196	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "prev_rank", "m": 16, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor prev_rank --m 16 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:59:10", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 261.63404202088714, "prompt_tps": 91.9208585718617, "gen_tps": 33.073144288813886, "decode_tps_wall": 32.556587418479594, "itl_ms_p50": 29.125166009180248, "itl_ms_p95": 38.984958082437515, "itl_ms_p99": 46.841374947689474, "itl_ms_mean": 30.715750000025665, "total_s": 2.196726292022504, "peak_mem_gb": 2.8326179, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.8581856100104275, "cache_misses": 1360, "cache_hits": 8230, "stall_s": 0.5971727652940899, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 5856, "installed": 5895, "late": 114, "dup": 10784, "predictor": "prev_rank", "m": 16}, "ssd_bytes": 25653805056, "ssd_read_s": 3.7076285743387416}
	197	+{"backend": "titan", "capacity": 32, "policy": "lru", "predictor": "prev_rank", "m": 16, "prefetch_workers": 2, "nocache": true, "max_tokens": 64, "split": "heldout", "tag": "cold_nvme", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_decode.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --backend titan --capacity 32 --predictor prev_rank --m 16 --nocache --traces benchmarks/raw/gateC/traces_olmoe_v2_laptop --families general,code,reasoning --split heldout --repeats 2 --max-tokens 64 --tag cold_nvme --out benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl", "env": {"host": "MacBookPro-de-simon-pierre-2.local", "chip": "Apple M5 Max", "ram_gb": 48, "macos": "27.0", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": "5edaf3e"}, "seed": 0, "warmup": false, "repeat": 1, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:59:13", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 246.18675000965595, "prompt_tps": 56.988426417342126, "gen_tps": 26.874429803645274, "decode_tps_wall": 26.45488250478592, "itl_ms_p50": 37.13241696823388, "itl_ms_p95": 47.65595798380673, "itl_ms_p99": 52.26895899977535, "itl_ms_mean": 37.800205682980874, "total_s": 2.627599708037451, "peak_mem_gb": 2.603629484, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.7534276152434416, "cache_misses": 2284, "cache_hits": 6979, "stall_s": 0.9120199183234945, "hits_device": 0, "layers_complete": 0, "prefetch": {"issued": 7093, "installed": 7056, "late": 134, "dup": 9547, "predictor": "prev_rank", "m": 16}, "ssd_bytes": 33067892736, "ssd_read_s": 4.923615432111546}

added benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl +56 −0

		`@@ -0,0 +1,56 @@`
	1	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:55:14", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 178.5065419971943, "prompt_tps": 135.2013967648807, "gen_tps": 85.29149785612425, "decode_tps_wall": 83.96721550863145, "itl_ms_p50": 11.420542025007308, "itl_ms_p95": 16.61008410155773, "itl_ms_p99": 18.366832984611392, "itl_ms_mean": 11.909410047034422, "total_s": 0.9287993749603629, "peak_mem_gb": 2.4887002, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.8315087060786154, "cache_misses": 1616, "stall_s": 0.30019399186130613, "remote": null}
	2	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:55:15", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 137.19691603910178, "prompt_tps": 102.26237237442251, "gen_tps": 67.556206611936, "decode_tps_wall": 66.50156215173993, "itl_ms_p50": 14.862500014714897, "itl_ms_p95": 19.776417058892548, "itl_ms_p99": 22.61562494095415, "itl_ms_mean": 15.037240745085809, "total_s": 1.0845430829795077, "peak_mem_gb": 2.378082608, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.7312304202225343, "cache_misses": 2488, "stall_s": 0.4498609454603866, "remote": null}
	3	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:55:16", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 138.96504207514226, "prompt_tps": 137.0431559884101, "gen_tps": 86.86643912128626, "decode_tps_wall": 85.51017618663688, "itl_ms_p50": 11.230833944864571, "itl_ms_p95": 15.840583015233278, "itl_ms_p99": 16.847082995809615, "itl_ms_mean": 11.694514554821783, "total_s": 0.8757194590289146, "peak_mem_gb": 2.491316616, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.8505428997232276, "cache_misses": 1404, "stall_s": 0.28127925912849605, "remote": null}
	4	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Write a short motivational note for someone starting to lear", "ts": "2026-11-01T01:55:17", "prompt_tokens": 16, "gen_tokens": 64, "ttft_ms": 158.30483299214393, "prompt_tps": 101.30691236803094, "gen_tps": 70.68724291593257, "decode_tps_wall": 69.58378267048148, "itl_ms_p50": 13.993666972965002, "itl_ms_p95": 18.955791951157153, "itl_ms_p99": 20.24804102256894, "itl_ms_mean": 14.371164682661258, "total_s": 1.0636882079998031, "peak_mem_gb": 2.451297796, "text": "\n\nDear [Name],\n\nWelcome to the wonderful world of piano playing! You've made an exciting decision to learn this beautiful instrument, and I'm thrilled to be a part of your journey.\n\nAs an adult, you b", "cache_hit_rate": 0.7533808966031307, "cache_misses": 2316, "stall_s": 0.42419996042735875, "remote": null}
	5	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:55:18", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 165.02429090905935, "prompt_tps": 109.29177412541864, "gen_tps": 76.24919105377168, "decode_tps_wall": 75.05894503518664, "itl_ms_p50": 12.891583028249443, "itl_ms_p95": 16.64870895911008, "itl_ms_p99": 19.467083970084786, "itl_ms_mean": 13.322862445391596, "total_s": 1.00436462496873, "peak_mem_gb": 2.458187812, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.7915640723691254, "cache_misses": 1947, "stall_s": 0.38373546046204865, "remote": null}
	6	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "general", "prompt": "What should a visitor know before hiking in Gatineau Park in", "ts": "2026-11-01T01:55:19", "prompt_tokens": 16, "gen_tokens": 64, "ttft_ms": 173.93787496257573, "prompt_tps": 92.15448546106387, "gen_tps": 69.8509005899028, "decode_tps_wall": 68.76032461231404, "itl_ms_p50": 13.8498330488801, "itl_ms_p95": 20.000332966446877, "itl_ms_p99": 21.816540975123644, "itl_ms_mean": 14.543270492660147, "total_s": 1.090163916000165, "peak_mem_gb": 2.48167362, "text": "\n\n1. Temperature and Weather Conditions: The temperature in Gatineau Park can drop to as low as -20\u00b0C (-4\u00b0F) during winter. Visitors should dress warmly, including layers of clothing, a hat, glove", "cache_hit_rate": 0.7343583173179585, "cache_misses": 2488, "stall_s": 0.44438852614257485, "remote": null}
	7	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a Go function that reads lines from stdin and prints t", "ts": "2026-11-01T01:55:20", "prompt_tokens": 21, "gen_tokens": 64, "ttft_ms": 167.44037496391684, "prompt_tps": 125.65955638304263, "gen_tps": 87.24190006967281, "decode_tps_wall": 85.8797892789634, "itl_ms_p50": 11.386458994820714, "itl_ms_p95": 16.809249995276332, "itl_ms_p99": 18.317250069230795, "itl_ms_mean": 11.64418320533716, "total_s": 0.9010239169001579, "peak_mem_gb": 2.62909018, "text": "package main\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n)\n\nfunc main() {\n\tfmt.Println(\"Enter your lines (press Ctrl+D to finish):\")\n\tvar line string\n\tfmt.Scan(&line)\n\n\tif len(line) >", "cache_hit_rate": 0.839597947858863, "cache_misses": 1532, "stall_s": 0.29442348168231547, "remote": null}
	8	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Implement Dijkstra's shortest path in Python using a binary ", "ts": "2026-11-01T01:55:21", "prompt_tokens": 20, "gen_tokens": 64, "ttft_ms": 139.64537507854402, "prompt_tps": 143.54796750430953, "gen_tps": 85.5902706529226, "decode_tps_wall": 84.25424651708104, "itl_ms_p50": 11.529083945788443, "itl_ms_p95": 15.486916992813349, "itl_ms_p99": 17.92583300266415, "itl_ms_mean": 11.868837967677603, "total_s": 0.887382167042233, "peak_mem_gb": 2.462204416, "text": "import heapq\n\ndef dijkstra(graph, start):\n # Create a priority queue (binary heap)\n pq = []\n for node, distance in graph.items():\n heapq.heappush(pq, (distance, node))\n\n # Initializ", "cache_hit_rate": 0.8534923339011925, "cache_misses": 1376, "stall_s": 0.29426933045033365, "remote": null}
	9	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 0, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:55:22", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 146.9164160080254, "prompt_tps": 95.46034900302631, "gen_tps": 66.82504844253033, "decode_tps_wall": 65.78184570894004, "itl_ms_p50": 15.321042039431632, "itl_ms_p95": 19.980457960627973, "itl_ms_p99": 22.824499988928437, "itl_ms_mean": 15.201762571768274, "total_s": 1.1046274580294266, "peak_mem_gb": 2.378082704, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.730690288430377, "cache_misses": 2493, "stall_s": 0.46624807722400874, "remote": null}
	10	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 0, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:55:23", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 162.1235409984365, "prompt_tps": 111.23470520221872, "gen_tps": 77.66610518487911, "decode_tps_wall": 76.45371273766366, "itl_ms_p50": 12.691083014942706, "itl_ms_p95": 16.15708298049867, "itl_ms_p99": 18.63858290016651, "itl_ms_mean": 13.079809523852807, "total_s": 0.9861515410011634, "peak_mem_gb": 2.47575146, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.7912429076116048, "cache_misses": 1950, "stall_s": 0.36670958204194903, "remote": null}
	11	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:55:24", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 162.7222920069471, "prompt_tps": 147.76865457730722, "gen_tps": 85.17920156787372, "decode_tps_wall": 83.84935992799959, "itl_ms_p50": 11.171665973961353, "itl_ms_p95": 16.503208898939192, "itl_ms_p99": 18.431083066388965, "itl_ms_mean": 11.926149476378683, "total_s": 0.9140697090188041, "peak_mem_gb": 2.488716588, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.833385465540611, "cache_misses": 1598, "stall_s": 0.30661741935182363, "remote": null}
	12	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:55:25", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 122.92812496889383, "prompt_tps": 154.90782678547015, "gen_tps": 87.17073754227556, "decode_tps_wall": 85.80985863220397, "itl_ms_p50": 10.944292065687478, "itl_ms_p95": 16.057790955528617, "itl_ms_p99": 17.15020788833499, "itl_ms_mean": 11.653672619205382, "total_s": 0.8571094999788329, "peak_mem_gb": 2.470363016, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.8594847775175644, "cache_misses": 1320, "stall_s": 0.268574655870907, "remote": null}
	13	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 0, "family": "general", "prompt": "What should a visitor know before hiking in Gatineau Park in", "ts": "2026-11-01T01:55:26", "prompt_tokens": 16, "gen_tokens": 64, "ttft_ms": 175.50287500489503, "prompt_tps": 91.32452674904759, "gen_tps": 69.83237998653466, "decode_tps_wall": 68.74226793595986, "itl_ms_p50": 14.033625018782914, "itl_ms_p95": 19.981416990049183, "itl_ms_p99": 22.00574998278171, "itl_ms_mean": 14.547090604162168, "total_s": 1.0919695830671117, "peak_mem_gb": 2.478134676, "text": "\n\n1. Temperature and Weather Conditions: The temperature in Gatineau Park can drop to as low as -20\u00b0C (-4\u00b0F) during winter. Visitors should dress warmly, including layers of clothing, a hat, glove", "cache_hit_rate": 0.7351057014734145, "cache_misses": 2481, "stall_s": 0.4473360065603629, "remote": null}
	14	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 0, "family": "general", "prompt": "Write a short motivational note for someone starting to lear", "ts": "2026-11-01T01:55:27", "prompt_tokens": 16, "gen_tokens": 64, "ttft_ms": 160.18208302557468, "prompt_tps": 100.0729437649106, "gen_tps": 70.45778798608657, "decode_tps_wall": 69.35762952120434, "itl_ms_p50": 13.85816594120115, "itl_ms_p95": 19.32783389929682, "itl_ms_p99": 20.50099999178201, "itl_ms_mean": 14.418024475508858, "total_s": 1.0685176249826327, "peak_mem_gb": 2.472564228, "text": "\n\nDear [Name],\n\nWelcome to the wonderful world of piano playing! You've made an exciting decision to learn this beautiful instrument, and I'm thrilled to be a part of your journey.\n\nAs an adult, you b", "cache_hit_rate": 0.7527419870088382, "cache_misses": 2322, "stall_s": 0.43018227932043374, "remote": null}
	15	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Write a Go function that reads lines from stdin and prints t", "ts": "2026-11-01T01:55:28", "prompt_tokens": 21, "gen_tokens": 64, "ttft_ms": 166.05354193598032, "prompt_tps": 126.69017593141277, "gen_tps": 87.04919884972608, "decode_tps_wall": 85.69026911704765, "itl_ms_p50": 11.284708976745605, "itl_ms_p95": 16.361374990083277, "itl_ms_p99": 18.399417051114142, "itl_ms_mean": 11.66993650859074, "total_s": 0.9012595419771969, "peak_mem_gb": 2.62909018, "text": "package main\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n)\n\nfunc main() {\n\tfmt.Println(\"Enter your lines (press Ctrl+D to finish):\")\n\tvar line string\n\tfmt.Scan(&line)\n\n\tif len(line) >", "cache_hit_rate": 0.839597947858863, "cache_misses": 1532, "stall_s": 0.293128167046234, "remote": null}
	16	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Implement Dijkstra's shortest path in Python using a binary ", "ts": "2026-11-01T01:55:29", "prompt_tokens": 20, "gen_tokens": 64, "ttft_ms": 135.34779206383973, "prompt_tps": 148.09882797281205, "gen_tps": 87.69233512390521, "decode_tps_wall": 86.32334979907448, "itl_ms_p50": 11.204999987967312, "itl_ms_p95": 15.035209013149142, "itl_ms_p99": 17.13545899838209, "itl_ms_mean": 11.584351190351068, "total_s": 0.865161917055957, "peak_mem_gb": 2.462204416, "text": "import heapq\n\ndef dijkstra(graph, start):\n # Create a priority queue (binary heap)\n pq = []\n for node, distance in graph.items():\n heapq.heappush(pq, (distance, node))\n\n # Initializ", "cache_hit_rate": 0.8534923339011925, "cache_misses": 1376, "stall_s": 0.2728264428442344, "remote": null}
	17	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:55:30", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 134.4117079861462, "prompt_tps": 141.659076615189, "gen_tps": 85.04810557075245, "decode_tps_wall": 83.72053616230828, "itl_ms_p50": 11.741708032786846, "itl_ms_p95": 16.16783393546939, "itl_ms_p99": 17.03070802614093, "itl_ms_mean": 11.944500666614326, "total_s": 0.8869152499828488, "peak_mem_gb": 2.470526856, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.8544815839897807, "cache_misses": 1367, "stall_s": 0.29486278689000756, "remote": null}
	18	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:55:30", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 152.79633307363838, "prompt_tps": 157.37992946371142, "gen_tps": 84.87378422386163, "decode_tps_wall": 83.54890553921419, "itl_ms_p50": 11.48695801384747, "itl_ms_p95": 16.662833048030734, "itl_ms_p99": 18.45091604627669, "itl_ms_mean": 11.96903769769484, "total_s": 0.9068457080284134, "peak_mem_gb": 2.462681856, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.8391200083411532, "cache_misses": 1543, "stall_s": 0.2983368367422372, "remote": null}
	19	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 1, "family": "general", "prompt": "What should a visitor know before hiking in Gatineau Park in", "ts": "2026-11-01T01:55:32", "prompt_tokens": 16, "gen_tokens": 64, "ttft_ms": 167.2900830162689, "prompt_tps": 95.79478321726408, "gen_tps": 68.34150218383628, "decode_tps_wall": 67.27441447089066, "itl_ms_p50": 14.28854197729379, "itl_ms_p95": 20.15958307310939, "itl_ms_p99": 22.961582988500595, "itl_ms_mean": 14.864492063809719, "total_s": 1.1037530830362812, "peak_mem_gb": 2.474595732, "text": "\n\n1. Temperature and Weather Conditions: The temperature in Gatineau Park can drop to as low as -20\u00b0C (-4\u00b0F) during winter. Visitors should dress warmly, including layers of clothing, a hat, glove", "cache_hit_rate": 0.7362801622891308, "cache_misses": 2470, "stall_s": 0.45875864988192916, "remote": null}
	20	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Implement Dijkstra's shortest path in Python using a binary ", "ts": "2026-11-01T01:55:32", "prompt_tokens": 20, "gen_tokens": 64, "ttft_ms": 151.3802499976009, "prompt_tps": 132.38114582665884, "gen_tps": 87.38064732483016, "decode_tps_wall": 86.01648937387795, "itl_ms_p50": 11.2885000417009, "itl_ms_p95": 15.679541043937206, "itl_ms_p99": 16.871374915353954, "itl_ms_mean": 11.625677905237627, "total_s": 0.8837979580275714, "peak_mem_gb": 2.484714128, "text": "import heapq\n\ndef dijkstra(graph, start):\n # Create a priority queue (binary heap)\n pq = []\n for node, distance in graph.items():\n heapq.heappush(pq, (distance, node))\n\n # Initializ", "cache_hit_rate": 0.843164395229983, "cache_misses": 1473, "stall_s": 0.2850979488575831, "remote": null}
	21	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 1, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:55:33", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 166.17045796010643, "prompt_tps": 108.493019146244, "gen_tps": 77.70834053769316, "decode_tps_wall": 76.49541319997908, "itl_ms_p50": 12.561082956381142, "itl_ms_p95": 16.49658300448209, "itl_ms_p99": 18.97879200987518, "itl_ms_mean": 13.072679238763476, "total_s": 0.9897492500022054, "peak_mem_gb": 2.478372888, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.7926346215608607, "cache_misses": 1937, "stall_s": 0.3649376155808568, "remote": null}
	22	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 1, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:55:35", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 152.44841703679413, "prompt_tps": 92.00715268714727, "gen_tps": 67.63118307781752, "decode_tps_wall": 66.57519917654966, "itl_ms_p50": 14.890999998897314, "itl_ms_p95": 19.589249975979328, "itl_ms_p99": 21.93804201669991, "itl_ms_mean": 15.020608460338462, "total_s": 1.0987467500381172, "peak_mem_gb": 2.376903056, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.7274494976774333, "cache_misses": 2523, "stall_s": 0.45728433376643807, "remote": null}
	23	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Write a Go function that reads lines from stdin and prints t", "ts": "2026-11-01T01:55:35", "prompt_tokens": 21, "gen_tokens": 64, "ttft_ms": 162.5065830303356, "prompt_tps": 129.47208525668592, "gen_tps": 86.91246514746022, "decode_tps_wall": 85.55557139831683, "itl_ms_p50": 11.230209027417004, "itl_ms_p95": 16.04699995368719, "itl_ms_p99": 18.92783399671316, "itl_ms_mean": 11.688309523927431, "total_s": 0.8988700830377638, "peak_mem_gb": 2.639395204, "text": "package main\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n)\n\nfunc main() {\n\tfmt.Println(\"Enter your lines (press Ctrl+D to finish):\")\n\tvar line string\n\tfmt.Scan(&line)\n\n\tif len(line) >", "cache_hit_rate": 0.8392838446235996, "cache_misses": 1535, "stall_s": 0.2996750872116536, "remote": null}
	24	+{"backend": "titan-distributed", "capacity": 32, "placement": "none", "workers": ["w0"], "rtt_ms": {"w0": [0.650166068226099, 0.46366697642952204, 0.5771249998360872, 0.6172090070322156, 0.4903330700471997, 0.5491249030455947, 0.6382910069078207, 0.6402080180123448, 0.47025003004819155, 0.5896659567952156]}, "n_remote_experts": 0, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement none --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 1, "family": "general", "prompt": "Write a short motivational note for someone starting to lear", "ts": "2026-11-01T01:55:37", "prompt_tokens": 16, "gen_tokens": 64, "ttft_ms": 154.26137496251613, "prompt_tps": 103.90959860802235, "gen_tps": 70.71857391390355, "decode_tps_wall": 69.61445834315857, "itl_ms_p50": 13.978666975162923, "itl_ms_p95": 18.903166987001896, "itl_ms_p99": 21.142125013284385, "itl_ms_mean": 14.36483201622549, "total_s": 1.059245791984722, "peak_mem_gb": 2.461947396, "text": "\n\nDear [Name],\n\nWelcome to the wonderful world of piano playing! You've made an exciting decision to learn this beautiful instrument, and I'm thrilled to be a part of your journey.\n\nAs an adult, you b", "cache_hit_rate": 0.7514641678202534, "cache_misses": 2334, "stall_s": 0.4296938021434471, "remote": null}
	25	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:55:44", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 374.1029159864411, "prompt_tps": 64.45335044114691, "gen_tps": 15.068215016015037, "decode_tps_wall": 14.833604919487318, "itl_ms_p50": 66.3045410765335, "itl_ms_p95": 83.44208297785372, "itl_ms_p99": 103.98220794741064, "itl_ms_mean": 67.41449603300896, "total_s": 4.621216166066006, "peak_mem_gb": 2.39195148, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.9510503773200082, "cache_misses": 240, "stall_s": 0.11324438010342419, "remote": {"remote_pairs": 5498, "local_pairs": 5766, "layers_remote": 1086, "remote_wait_s": 2.4414072384824976, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 1102, "exec_wall_s": 3.0890125607838854, "bytes_sent": 6237027, "bytes_recv": 23909140}}}}
	26	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:55:48", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 406.81245899759233, "prompt_tps": 34.559545843670264, "gen_tps": 15.57887635546551, "decode_tps_wall": 15.335682886660065, "itl_ms_p50": 64.76954102981836, "itl_ms_p95": 82.00216700788587, "itl_ms_p99": 85.42329201009125, "itl_ms_mean": 65.20739946115229, "total_s": 4.514878625050187, "peak_mem_gb": 2.234789604, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.9911631730374024, "cache_misses": 43, "stall_s": 0.07472425198648125, "remote": {"remote_pairs": 4766, "local_pairs": 5218, "layers_remote": 1072, "remote_wait_s": 2.3485147501342, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 2174, "exec_wall_s": 5.5557810658356175, "bytes_sent": 11478587, "bytes_recv": 43536815}}}}
	27	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:55:53", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 394.3847499322146, "prompt_tps": 48.270068576301135, "gen_tps": 15.415767616154488, "decode_tps_wall": 15.17506713280049, "itl_ms_p50": 65.56804210413247, "itl_ms_p95": 80.68816701415926, "itl_ms_p99": 82.00950000900775, "itl_ms_mean": 65.8975667948465, "total_s": 4.545931458007544, "peak_mem_gb": 2.118866256, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.9997953336062219, "cache_misses": 1, "stall_s": 0.0035461250226944685, "remote": {"remote_pairs": 5080, "local_pairs": 5544, "layers_remote": 1076, "remote_wait_s": 2.454924648278393, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 3250, "exec_wall_s": 8.056348544429056, "bytes_sent": 17042206, "bytes_recv": 64450919}}}}
	28	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Write a short motivational note for someone starting to lear", "ts": "2026-11-01T01:55:57", "prompt_tokens": 16, "gen_tokens": 64, "ttft_ms": 412.4626669799909, "prompt_tps": 38.93484011208967, "gen_tps": 15.165112081698279, "decode_tps_wall": 14.928277181055577, "itl_ms_p50": 65.68445800803602, "itl_ms_p95": 83.55854195542634, "itl_ms_p99": 85.41866601444781, "itl_ms_mean": 66.98696627022906, "total_s": 4.632641542004421, "peak_mem_gb": 2.1928547, "text": "\n\nDear [Name],\n\nWelcome to the wonderful world of piano playing! You've made an exciting decision to learn this beautiful instrument, and I'm thrilled to be a part of your journey.\n\nAs an adult, you b", "cache_hit_rate": 0.9984779299847792, "cache_misses": 7, "stall_s": 0.012387834023684263, "remote": {"remote_pairs": 5268, "local_pairs": 4972, "layers_remote": 1077, "remote_wait_s": 2.486839544842951, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 4327, "exec_wall_s": 10.594491942902096, "bytes_sent": 22410562, "bytes_recv": 86135018}}}}
	29	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:56:02", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 394.823707989417, "prompt_tps": 45.78161614812507, "gen_tps": 15.38288234416988, "decode_tps_wall": 15.142698753768581, "itl_ms_p50": 65.66441594623029, "itl_ms_p95": 76.68329204898328, "itl_ms_p99": 84.53599992208183, "itl_ms_mean": 66.03842658833379, "total_s": 4.555244583054446, "peak_mem_gb": 2.118837564, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.9997890740350137, "cache_misses": 1, "stall_s": 0.0036434170324355364, "remote": {"remote_pairs": 5185, "local_pairs": 5311, "layers_remote": 1070, "remote_wait_s": 2.5084352880949154, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 5397, "exec_wall_s": 13.145984808332287, "bytes_sent": 27896423, "bytes_recv": 107478069}}}}
	30	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "general", "prompt": "What should a visitor know before hiking in Gatineau Park in", "ts": "2026-11-01T01:56:06", "prompt_tokens": 16, "gen_tokens": 64, "ttft_ms": 362.0685419300571, "prompt_tps": 44.37987789118793, "gen_tps": 15.281103511109274, "decode_tps_wall": 15.04245030328018, "itl_ms_p50": 65.74237497989088, "itl_ms_p95": 78.6904999986291, "itl_ms_p99": 82.10070908535272, "itl_ms_mean": 66.47853107960333, "total_s": 4.550215999945067, "peak_mem_gb": 2.1181566, "text": "\n\n1. Temperature and Weather Conditions: The temperature in Gatineau Park can drop to as low as -20\u00b0C (-4\u00b0F) during winter. Visitors should dress warmly, including layers of clothing, a hat, glove", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5195, "local_pairs": 5045, "layers_remote": 1079, "remote_wait_s": 2.4603631681529805, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 6476, "exec_wall_s": 15.646278485888615, "bytes_sent": 33276690, "bytes_recv": 128862963}}}}
	31	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a Go function that reads lines from stdin and prints t", "ts": "2026-11-01T01:56:11", "prompt_tokens": 21, "gen_tokens": 64, "ttft_ms": 427.8730839723721, "prompt_tps": 49.26016405710952, "gen_tps": 14.369898072449228, "decode_tps_wall": 14.145553689009986, "itl_ms_p50": 71.40662497840822, "itl_ms_p95": 81.9948329590261, "itl_ms_p99": 83.02629203535616, "itl_ms_mean": 70.69359192188593, "total_s": 4.8815693750511855, "peak_mem_gb": 2.117360616, "text": "package main\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n)\n\nfunc main() {\n\tfmt.Println(\"Enter your lines (press Ctrl+D to finish):\")\n\tvar line string\n\tfmt.Scan(&line)\n\n\tif len(line) >", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5303, "local_pairs": 5577, "layers_remote": 1087, "remote_wait_s": 2.586499751312658, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 7563, "exec_wall_s": 18.2761108946288, "bytes_sent": 38986233, "bytes_recv": 150691037}}}}
	32	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Implement Dijkstra's shortest path in Python using a binary ", "ts": "2026-11-01T01:56:16", "prompt_tokens": 20, "gen_tokens": 64, "ttft_ms": 432.6194580644369, "prompt_tps": 46.37712077172855, "gen_tps": 14.705547687114, "decode_tps_wall": 14.475975847372972, "itl_ms_p50": 68.95387498661876, "itl_ms_p95": 83.08483299333602, "itl_ms_p99": 86.98933303821832, "itl_ms_mean": 69.07997157106855, "total_s": 4.784657667041756, "peak_mem_gb": 2.117329368, "text": "import heapq\n\ndef dijkstra(graph, start):\n # Create a priority queue (binary heap)\n pq = []\n for node, distance in graph.items():\n heapq.heappush(pq, (distance, node))\n\n # Initializ", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5276, "local_pairs": 5476, "layers_remote": 1077, "remote_wait_s": 2.546213002759032, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 8640, "exec_wall_s": 20.86342571384739, "bytes_sent": 44616881, "bytes_recv": 172407476}}}}
	33	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 0, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:56:21", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 365.9865000518039, "prompt_tps": 38.39828847339567, "gen_tps": 15.06037976991632, "decode_tps_wall": 14.825232712502277, "itl_ms_p50": 67.21037498209625, "itl_ms_p95": 78.49420898128301, "itl_ms_p99": 83.23866699356586, "itl_ms_mean": 67.45256680906529, "total_s": 4.615498209022917, "peak_mem_gb": 2.11646588, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 4766, "local_pairs": 5218, "layers_remote": 1072, "remote_wait_s": 2.51390994573012, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 9712, "exec_wall_s": 23.418756291270256, "bytes_sent": 49858441, "bytes_recv": 192034295}}}}
	34	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 0, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:56:25", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 397.6970419753343, "prompt_tps": 45.304906767709255, "gen_tps": 14.576662683119048, "decode_tps_wall": 14.349083711392803, "itl_ms_p50": 68.59983294270933, "itl_ms_p95": 82.47004193253815, "itl_ms_p99": 86.76099998410791, "itl_ms_mean": 69.69086111094506, "total_s": 4.788221291964874, "peak_mem_gb": 2.117292472, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5185, "local_pairs": 5311, "layers_remote": 1070, "remote_wait_s": 2.5479154620552436, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 10782, "exec_wall_s": 26.008038206840865, "bytes_sent": 55344302, "bytes_recv": 213377595}}}}
	35	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:56:30", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 502.1301250671968, "prompt_tps": 47.937756852362824, "gen_tps": 15.149045451555484, "decode_tps_wall": 14.912525461834464, "itl_ms_p50": 68.03220906294882, "itl_ms_p95": 78.64704099483788, "itl_ms_p99": 82.98891701269895, "itl_ms_mean": 67.0577228893452, "total_s": 4.726766667095944, "peak_mem_gb": 2.118710312, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5498, "local_pairs": 5766, "layers_remote": 1086, "remote_wait_s": 2.558194422046654, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 11868, "exec_wall_s": 28.607590664876625, "bytes_sent": 61247737, "bytes_recv": 236005152}}}}
	36	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:56:35", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 441.87854102347046, "prompt_tps": 43.15306990152925, "gen_tps": 14.869198533365013, "decode_tps_wall": 14.63703053702119, "itl_ms_p50": 67.03179189935327, "itl_ms_p95": 84.04724998399615, "itl_ms_p99": 87.05008297692984, "itl_ms_mean": 68.31986839616937, "total_s": 4.746030249982141, "peak_mem_gb": 2.117214664, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5080, "local_pairs": 5544, "layers_remote": 1076, "remote_wait_s": 2.52509659784846, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 12944, "exec_wall_s": 31.177086664014496, "bytes_sent": 66811356, "bytes_recv": 256919569}}}}
	37	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 0, "family": "general", "prompt": "What should a visitor know before hiking in Gatineau Park in", "ts": "2026-11-01T01:56:40", "prompt_tokens": 16, "gen_tokens": 64, "ttft_ms": 391.6577079799026, "prompt_tps": 41.00100971960321, "gen_tps": 14.84831934271548, "decode_tps_wall": 14.616513156490033, "itl_ms_p50": 68.51250003091991, "itl_ms_p95": 80.63674997538328, "itl_ms_p99": 83.12412502709776, "itl_ms_mean": 68.41576984152198, "total_s": 4.701851207995787, "peak_mem_gb": 2.116729528, "text": "\n\n1. Temperature and Weather Conditions: The temperature in Gatineau Park can drop to as low as -20\u00b0C (-4\u00b0F) during winter. Visitors should dress warmly, including layers of clothing, a hat, glove", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5195, "local_pairs": 5045, "layers_remote": 1079, "remote_wait_s": 2.526873570284806, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 14023, "exec_wall_s": 33.74513076571748, "bytes_sent": 72191623, "bytes_recv": 278305340}}}}
	38	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 0, "family": "general", "prompt": "Write a short motivational note for someone starting to lear", "ts": "2026-11-01T01:56:44", "prompt_tokens": 16, "gen_tokens": 64, "ttft_ms": 399.7366250259802, "prompt_tps": 40.171990553010986, "gen_tps": 15.13912083515291, "decode_tps_wall": 14.902680175827347, "itl_ms_p50": 66.29099999554455, "itl_ms_p95": 81.93741692230105, "itl_ms_p99": 86.64470899384469, "itl_ms_mean": 67.10202381059173, "total_s": 4.627164125093259, "peak_mem_gb": 2.118217908, "text": "\n\nDear [Name],\n\nWelcome to the wonderful world of piano playing! You've made an exciting decision to learn this beautiful instrument, and I'm thrilled to be a part of your journey.\n\nAs an adult, you b", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5268, "local_pairs": 4972, "layers_remote": 1077, "remote_wait_s": 2.5310352966189384, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 15100, "exec_wall_s": 36.31421813927591, "bytes_sent": 77559979, "bytes_recv": 299989938}}}}
	39	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Write a Go function that reads lines from stdin and prints t", "ts": "2026-11-01T01:56:49", "prompt_tokens": 21, "gen_tokens": 64, "ttft_ms": 546.0713329957798, "prompt_tps": 38.562928030022064, "gen_tps": 14.132192827300631, "decode_tps_wall": 13.91158632885405, "itl_ms_p50": 69.87091701012105, "itl_ms_p95": 82.87633303552866, "itl_ms_p99": 91.89437492750585, "itl_ms_mean": 71.88252844507733, "total_s": 5.074670625035651, "peak_mem_gb": 2.11737188, "text": "package main\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n)\n\nfunc main() {\n\tfmt.Println(\"Enter your lines (press Ctrl+D to finish):\")\n\tvar line string\n\tfmt.Scan(&line)\n\n\tif len(line) >", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5303, "local_pairs": 5577, "layers_remote": 1087, "remote_wait_s": 2.737311046337709, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 16187, "exec_wall_s": 39.09460177505389, "bytes_sent": 83269522, "bytes_recv": 321818871}}}}
	40	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 0, "family": "code", "prompt": "Implement Dijkstra's shortest path in Python using a binary ", "ts": "2026-11-01T01:56:54", "prompt_tokens": 20, "gen_tokens": 64, "ttft_ms": 428.98575006984174, "prompt_tps": 46.75799994121783, "gen_tps": 14.561354691058426, "decode_tps_wall": 14.333987616562041, "itl_ms_p50": 70.06154197733849, "itl_ms_p95": 81.88924996647984, "itl_ms_p99": 85.67404199857265, "itl_ms_mean": 69.76425728487176, "total_s": 4.824133959016763, "peak_mem_gb": 2.11717372, "text": "import heapq\n\ndef dijkstra(graph, start):\n # Create a priority queue (binary heap)\n pq = []\n for node, distance in graph.items():\n heapq.heappush(pq, (distance, node))\n\n # Initializ", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5276, "local_pairs": 5476, "layers_remote": 1077, "remote_wait_s": 2.5831803614273667, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 17264, "exec_wall_s": 41.71912443486508, "bytes_sent": 88900170, "bytes_recv": 343536203}}}}
	41	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:56:59", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 440.9932499984279, "prompt_tps": 43.23506776748733, "gen_tps": 15.491728960104247, "decode_tps_wall": 15.249880640954956, "itl_ms_p50": 64.38850006088614, "itl_ms_p95": 81.6039580386132, "itl_ms_p99": 92.41149993613362, "itl_ms_mean": 65.57428373009084, "total_s": 4.572173124994151, "peak_mem_gb": 2.117112264, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5080, "local_pairs": 5544, "layers_remote": 1076, "remote_wait_s": 2.4575695286039263, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 18340, "exec_wall_s": 44.21957937953994, "bytes_sent": 94463789, "bytes_recv": 364450393}}}}
	42	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:57:04", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 513.9315000269562, "prompt_tps": 46.8238549984884, "gen_tps": 14.645398013728665, "decode_tps_wall": 14.41674374490024, "itl_ms_p50": 69.07674996182323, "itl_ms_p95": 82.37241697497666, "itl_ms_p99": 92.20283408649266, "itl_ms_mean": 69.36379099848665, "total_s": 4.883850332931615, "peak_mem_gb": 2.117523608, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5498, "local_pairs": 5766, "layers_remote": 1086, "remote_wait_s": 2.676588977687061, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 19426, "exec_wall_s": 46.939388262690045, "bytes_sent": 100367224, "bytes_recv": 387077694}}}}
	43	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 1, "family": "general", "prompt": "What should a visitor know before hiking in Gatineau Park in", "ts": "2026-11-01T01:57:08", "prompt_tokens": 16, "gen_tokens": 64, "ttft_ms": 393.809667089954, "prompt_tps": 40.760860879697425, "gen_tps": 15.02024873134101, "decode_tps_wall": 14.785734177562771, "itl_ms_p50": 67.60754203423858, "itl_ms_p95": 81.05779101606458, "itl_ms_p99": 84.4417919870466, "itl_ms_mean": 67.6327592523266, "total_s": 4.654673499986529, "peak_mem_gb": 2.116823, "text": "\n\n1. Temperature and Weather Conditions: The temperature in Gatineau Park can drop to as low as -20\u00b0C (-4\u00b0F) during winter. Visitors should dress warmly, including layers of clothing, a hat, glove", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5195, "local_pairs": 5045, "layers_remote": 1079, "remote_wait_s": 2.504336794721894, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 20505, "exec_wall_s": 49.4849353521131, "bytes_sent": 105747491, "bytes_recv": 408463236}}}}
	44	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Implement Dijkstra's shortest path in Python using a binary ", "ts": "2026-11-01T01:57:13", "prompt_tokens": 20, "gen_tokens": 64, "ttft_ms": 454.0538330329582, "prompt_tps": 44.188226381154706, "gen_tps": 15.854698623434825, "decode_tps_wall": 15.607095096973826, "itl_ms_p50": 62.409750069491565, "itl_ms_p95": 79.17545805685222, "itl_ms_p99": 88.79329194314778, "itl_ms_mean": 64.07342261878685, "total_s": 4.49067945801653, "peak_mem_gb": 2.117079512, "text": "import heapq\n\ndef dijkstra(graph, start):\n # Create a priority queue (binary heap)\n pq = []\n for node, distance in graph.items():\n heapq.heappush(pq, (distance, node))\n\n # Initializ", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5276, "local_pairs": 5476, "layers_remote": 1077, "remote_wait_s": 2.4161494338186458, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 21582, "exec_wall_s": 51.94137500633951, "bytes_sent": 111378139, "bytes_recv": 430180340}}}}
	45	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 1, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:57:17", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 411.3634170498699, "prompt_tps": 43.89156868685017, "gen_tps": 15.400196822417758, "decode_tps_wall": 15.159722416776361, "itl_ms_p50": 64.80091693811119, "itl_ms_p95": 82.2217499371618, "itl_ms_p99": 83.59787496738136, "itl_ms_mean": 65.9642685075394, "total_s": 4.567112333024852, "peak_mem_gb": 2.11709804, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5185, "local_pairs": 5311, "layers_remote": 1070, "remote_wait_s": 2.487515496555716, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 22652, "exec_wall_s": 54.466830635094084, "bytes_sent": 116864000, "bytes_recv": 451524030}}}}
	46	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 1, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:57:22", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 373.53870808146894, "prompt_tps": 37.61048500312227, "gen_tps": 15.478581724182211, "decode_tps_wall": 15.236935404847973, "itl_ms_p50": 65.55262499023229, "itl_ms_p95": 81.47870807442814, "itl_ms_p99": 85.38237493485212, "itl_ms_mean": 65.62999536519841, "total_s": 4.5082284160889685, "peak_mem_gb": 2.117269908, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 4766, "local_pairs": 5218, "layers_remote": 1072, "remote_wait_s": 2.442976040299982, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 23724, "exec_wall_s": 56.953110640635714, "bytes_sent": 122105560, "bytes_recv": 471151661}}}}
	47	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 1, "family": "code", "prompt": "Write a Go function that reads lines from stdin and prints t", "ts": "2026-11-01T01:57:26", "prompt_tokens": 21, "gen_tokens": 64, "ttft_ms": 456.90874999854714, "prompt_tps": 46.10961558148383, "gen_tps": 15.486656536334875, "decode_tps_wall": 15.244791884474834, "itl_ms_p50": 65.32587495166808, "itl_ms_p95": 77.03037501778454, "itl_ms_p99": 94.00350006762892, "itl_ms_mean": 65.59617261934493, "total_s": 4.589467625017278, "peak_mem_gb": 2.117360616, "text": "package main\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n)\n\nfunc main() {\n\tfmt.Println(\"Enter your lines (press Ctrl+D to finish):\")\n\tvar line string\n\tfmt.Scan(&line)\n\n\tif len(line) >", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5303, "local_pairs": 5577, "layers_remote": 1087, "remote_wait_s": 2.525287412921898, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 24811, "exec_wall_s": 59.52095779264346, "bytes_sent": 127815103, "bytes_recv": 492980378}}}}
	48	+{"backend": "titan-distributed", "capacity": 32, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5602079909294844, 0.5876249633729458, 0.6379169644787908, 0.5096249515190721, 0.5739580374211073, 0.5437079817056656, 0.6351670017465949, 0.5011250032112002, 0.5718749016523361, 0.6378750549629331]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 32 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": false, "repeat": 1, "family": "general", "prompt": "Write a short motivational note for someone starting to lear", "ts": "2026-11-01T01:57:31", "prompt_tokens": 16, "gen_tokens": 64, "ttft_ms": 396.3244999758899, "prompt_tps": 40.49894701700236, "gen_tps": 15.485025943324581, "decode_tps_wall": 15.243234540567924, "itl_ms_p50": 64.10962494555861, "itl_ms_p95": 83.66962510626763, "itl_ms_p99": 86.45658299792558, "itl_ms_mean": 65.60287433343807, "total_s": 4.529305582982488, "peak_mem_gb": 2.116628824, "text": "\n\nDear [Name],\n\nWelcome to the wonderful world of piano playing! You've made an exciting decision to learn this beautiful instrument, and I'm thrilled to be a part of your journey.\n\nAs an adult, you b", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5268, "local_pairs": 4972, "layers_remote": 1077, "remote_wait_s": 2.49132607888896, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 25888, "exec_wall_s": 62.05094386078417, "bytes_sent": 133183459, "bytes_recv": 514664735}}}}
	49	+{"backend": "titan-distributed", "capacity": 64, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5988340126350522, 0.534667051397264, 0.5446670111268759, 0.547999981790781, 0.5813329480588436, 0.5235420539975166, 0.634583062492311, 0.5202499451115727, 0.6310830358415842, 0.5090000340715051]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 64 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a C function that computes the dot product of two floa", "ts": "2026-11-01T01:57:36", "prompt_tokens": 24, "gen_tokens": 64, "ttft_ms": 257.6604589121416, "prompt_tps": 93.49550484693123, "gen_tps": 16.955016001077553, "decode_tps_wall": 16.690376494227106, "itl_ms_p50": 59.50458301231265, "itl_ms_p95": 69.95866599027067, "itl_ms_p99": 70.94483310356736, "itl_ms_mean": 59.91476587396825, "total_s": 4.032290708972141, "peak_mem_gb": 3.976988316, "text": "#include <arm_neon.h>\n\nfloat dot_product_neon(const float a, const float b, int len) {\n // TODO: Implement the dot product computation using NEON intrinsics\n // Start by loading the arrays int", "cache_hit_rate": 0.9481977120656162, "cache_misses": 240, "stall_s": 0.09956279234029353, "remote": {"remote_pairs": 5498, "local_pairs": 5766, "layers_remote": 1055, "remote_wait_s": 2.1914026216836646, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 1071, "exec_wall_s": 2.412593872868456, "bytes_sent": 6234822, "bytes_recv": 23907886}}}}
	50	+{"backend": "titan-distributed", "capacity": 64, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5988340126350522, 0.534667051397264, 0.5446670111268759, 0.547999981790781, 0.5813329480588436, 0.5235420539975166, 0.634583062492311, 0.5202499451115727, 0.6310830358415842, 0.5090000340715051]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 64 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Summarize the history of the Rideau Canal in one paragraph.", "ts": "2026-11-01T01:57:40", "prompt_tokens": 14, "gen_tokens": 64, "ttft_ms": 310.83316600415856, "prompt_tps": 45.24359833855363, "gen_tps": 17.448054181829573, "decode_tps_wall": 17.175550862291246, "itl_ms_p50": 57.15133307967335, "itl_ms_p95": 67.94600002467632, "itl_ms_p99": 72.43295793887228, "itl_ms_mean": 58.22229563509897, "total_s": 3.978837791015394, "peak_mem_gb": 3.934340388, "text": "\n\nThe Rideau Canal, officially opened in 1832, was constructed by the British to connect Ottawa to Montreal via the Ottawa River. It was designed by Colonel John By and took approximately three years ", "cache_hit_rate": 0.9909110124709364, "cache_misses": 43, "stall_s": 0.07054520992096514, "remote": {"remote_pairs": 4766, "local_pairs": 5218, "layers_remote": 1055, "remote_wait_s": 2.1445465096039698, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 2126, "exec_wall_s": 4.672204844537191, "bytes_sent": 11474989, "bytes_recv": 43533820}}}}
	51	+{"backend": "titan-distributed", "capacity": 64, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5988340126350522, 0.534667051397264, 0.5446670111268759, 0.547999981790781, 0.5813329480588436, 0.5235420539975166, 0.634583062492311, 0.5202499451115727, 0.6310830358415842, 0.5090000340715051]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 64 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a bash script that lists the ten largest files under a", "ts": "2026-11-01T01:57:44", "prompt_tokens": 19, "gen_tokens": 64, "ttft_ms": 335.90412500780076, "prompt_tps": 56.82127588201756, "gen_tps": 17.344942175429484, "decode_tps_wall": 17.074052969998572, "itl_ms_p50": 57.24125006236136, "itl_ms_p95": 68.56862490531057, "itl_ms_p99": 72.3577500320971, "itl_ms_mean": 58.56840211033289, "total_s": 4.025713457958773, "peak_mem_gb": 3.930807632, "text": "#!/bin/bash\n\n# Specify the directory\ndir=\"/path/to/directory\"\n\n# Get the size of each file and store it in an array\nfiles=( $(ls -l \"$dir\" \| awk '{print $5, $9}') )\n\n# Sort the array", "cache_hit_rate": 0.999786233433091, "cache_misses": 1, "stall_s": 0.003519207937642932, "remote": {"remote_pairs": 5080, "local_pairs": 5544, "layers_remote": 1053, "remote_wait_s": 2.2477200055727735, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 3179, "exec_wall_s": 6.963823046302423, "bytes_sent": 17036896, "bytes_recv": 64445736}}}}
	52	+{"backend": "titan-distributed", "capacity": 64, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5988340126350522, 0.534667051397264, 0.5446670111268759, 0.547999981790781, 0.5813329480588436, 0.5235420539975166, 0.634583062492311, 0.5202499451115727, 0.6310830358415842, 0.5090000340715051]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 64 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Write a short motivational note for someone starting to lear", "ts": "2026-11-01T01:57:49", "prompt_tokens": 16, "gen_tokens": 64, "ttft_ms": 332.25095900706947, "prompt_tps": 48.379872822764185, "gen_tps": 16.568221478682467, "decode_tps_wall": 16.309467399036027, "itl_ms_p50": 59.867250034585595, "itl_ms_p95": 70.0176659738645, "itl_ms_p99": 83.56362499762326, "itl_ms_mean": 61.31408068292316, "total_s": 4.1950380420312285, "peak_mem_gb": 3.930772072, "text": "\n\nDear [Name],\n\nWelcome to the wonderful world of piano playing! You've made an exciting decision to learn this beautiful instrument, and I'm thrilled to be a part of your journey.\n\nAs an adult, you b", "cache_hit_rate": 0.9984170058796924, "cache_misses": 7, "stall_s": 0.014119333121925592, "remote": {"remote_pairs": 5268, "local_pairs": 4972, "layers_remote": 1053, "remote_wait_s": 2.3574230516096577, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 4232, "exec_wall_s": 9.374994189362042, "bytes_sent": 22403332, "bytes_recv": 86127723}}}}
	53	+{"backend": "titan-distributed", "capacity": 64, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5988340126350522, 0.534667051397264, 0.5446670111268759, 0.547999981790781, 0.5813329480588436, 0.5235420539975166, 0.634583062492311, 0.5202499451115727, 0.6310830358415842, 0.5090000340715051]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 64 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "general", "prompt": "Describe the main differences between a Mac Studio and a Mac", "ts": "2026-11-01T01:57:53", "prompt_tokens": 18, "gen_tokens": 64, "ttft_ms": 351.0849999729544, "prompt_tps": 51.5028949665602, "gen_tps": 17.201231904025644, "decode_tps_wall": 16.932477445892992, "itl_ms_p50": 59.86454198136926, "itl_ms_p95": 65.34862506669015, "itl_ms_p99": 70.08058298379183, "itl_ms_mean": 59.058103174532924, "total_s": 4.071745499968529, "peak_mem_gb": 3.930772376, "text": "\n\nMac Studio:\n- More powerful hardware options, including multiple GPUs and more RAM.\n- Higher-end CPU options, often with more cores and higher clock speeds.\n- Larger display with a higher resolution", "cache_hit_rate": 0.9997826086956522, "cache_misses": 1, "stall_s": 0.002606875030323863, "remote": {"remote_pairs": 5185, "local_pairs": 5311, "layers_remote": 1050, "remote_wait_s": 2.274140348425135, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 5282, "exec_wall_s": 11.692408705712296, "bytes_sent": 27887767, "bytes_recv": 107469402}}}}
	54	+{"backend": "titan-distributed", "capacity": 64, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5988340126350522, 0.534667051397264, 0.5446670111268759, 0.547999981790781, 0.5813329480588436, 0.5235420539975166, 0.634583062492311, 0.5202499451115727, 0.6310830358415842, 0.5090000340715051]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 64 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "general", "prompt": "What should a visitor know before hiking in Gatineau Park in", "ts": "2026-11-01T01:57:57", "prompt_tokens": 16, "gen_tokens": 64, "ttft_ms": 287.85133291967213, "prompt_tps": 55.84103800491083, "gen_tps": 16.437401480807562, "decode_tps_wall": 16.180715476516337, "itl_ms_p50": 62.47966596856713, "itl_ms_p95": 69.74233302753419, "itl_ms_p99": 72.38837506156415, "itl_ms_mean": 61.80196428590173, "total_s": 4.181375082931481, "peak_mem_gb": 3.929057676, "text": "\n\n1. Temperature and Weather Conditions: The temperature in Gatineau Park can drop to as low as -20\u00b0C (-4\u00b0F) during winter. Visitors should dress warmly, including layers of clothing, a hat, glove", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5195, "local_pairs": 5045, "layers_remote": 1056, "remote_wait_s": 2.3621108037186787, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 6338, "exec_wall_s": 14.095058789243922, "bytes_sent": 33266207, "bytes_recv": 128852601}}}}
	55	+{"backend": "titan-distributed", "capacity": 64, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5988340126350522, 0.534667051397264, 0.5446670111268759, 0.547999981790781, 0.5813329480588436, 0.5235420539975166, 0.634583062492311, 0.5202499451115727, 0.6310830358415842, 0.5090000340715051]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 64 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Write a Go function that reads lines from stdin and prints t", "ts": "2026-11-01T01:58:01", "prompt_tokens": 21, "gen_tokens": 64, "ttft_ms": 328.16816703416407, "prompt_tps": 64.29895954419706, "gen_tps": 16.441890008158396, "decode_tps_wall": 16.18513256726093, "itl_ms_p50": 61.68937508482486, "itl_ms_p95": 70.47816691920161, "itl_ms_p99": 72.9299170197919, "itl_ms_mean": 61.78509788809433, "total_s": 4.220629333984107, "peak_mem_gb": 3.929732608, "text": "package main\n\nimport (\n\t\"fmt\"\n\t\"strings\"\n)\n\nfunc main() {\n\tfmt.Println(\"Enter your lines (press Ctrl+D to finish):\")\n\tvar line string\n\tfmt.Scan(&line)\n\n\tif len(line) >", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5303, "local_pairs": 5577, "layers_remote": 1056, "remote_wait_s": 2.3896616504061967, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 7394, "exec_wall_s": 16.524280771496706, "bytes_sent": 38973417, "bytes_recv": 150678220}}}}
	56	+{"backend": "titan-distributed", "capacity": 64, "placement": "odd", "workers": ["w0"], "rtt_ms": {"w0": [0.5988340126350522, 0.534667051397264, 0.5446670111268759, 0.547999981790781, 0.5813329480588436, 0.5235420539975166, 0.634583062492311, 0.5202499451115727, 0.6310830358415842, 0.5090000340715051]}, "n_remote_experts": 512, "predictor": "none", "m": 8, "max_tokens": 64, "split": "heldout", "tag": "", "model": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b", "cmd": "benchmarks/scripts/bench_distributed.py --model /Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--OLMoE-1B-7B-0125-Instruct-4bit/snapshots/e23844197887b031e7ddddbb0b8959c5a6853a7b --workers w0=192.168.2.85:9411 --placement odd --capacity 64 --repeats 2 --out benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl", "env": {"host": "MacStudnpierre2", "chip": "Apple M3 Ultra", "ram_gb": 96, "macos": "26.6", "python": "3.12.13", "mlx": "0.32.4", "mlx_lm": "0.32.2", "numpy": "2.5.4", "git": ""}, "seed": 0, "dlogit_vs_mlx_lm": 0.0, "warmup": true, "repeat": -1, "family": "code", "prompt": "Implement Dijkstra's shortest path in Python using a binary ", "ts": "2026-11-01T01:58:05", "prompt_tokens": 20, "gen_tokens": 64, "ttft_ms": 318.29787499736995, "prompt_tps": 63.094882802830774, "gen_tps": 16.187676379146705, "decode_tps_wall": 15.934791127721224, "itl_ms_p50": 62.68637499306351, "itl_ms_p95": 69.6389579679817, "itl_ms_p99": 81.47662505507469, "itl_ms_mean": 62.75576453966399, "total_s": 4.271911040996201, "peak_mem_gb": 3.928974552, "text": "import heapq\n\ndef dijkstra(graph, start):\n # Create a priority queue (binary heap)\n pq = []\n for node, distance in graph.items():\n heapq.heappush(pq, (distance, node))\n\n # Initializ", "cache_hit_rate": 1.0, "cache_misses": 0, "stall_s": 0.0, "remote": {"remote_pairs": 5276, "local_pairs": 5476, "layers_remote": 1053, "remote_wait_s": 2.444692069548182, "failures": 0, "workers": {"w0": {"worker": "w0", "exec_calls": 8447, "exec_wall_s": 19.00972998223733, "bytes_sent": 44602299, "bytes_recv": 172392932}}}}

added figures/capacity_sweep.pdf +0 −0

Binary file not shown.

added figures/sim_hit_vs_capacity.pdf +0 −0

Binary file not shown.

added figures/two_node.pdf +0 −0

Binary file not shown.

added paper/figures/capacity_sweep.pdf +0 −0

Binary file not shown.

added paper/figures/sim_hit_vs_capacity.pdf +0 −0

Binary file not shown.

added paper/figures/two_node.pdf +0 −0

Binary file not shown.

modified paper/main.pdf +0 −0

Binary file not shown.

modified paper/sections/results.tex +38 −2

		`@@ -52,6 +52,14 @@ log-probabilities only 0.16 with these few traces. Table~\ref{tab:sim} is genera`
52	52
53	53	\input{tables/sim}
54	54
	55	+\begin{figure}[t]
	56	+\centering
	57	+\includegraphics[width=\textwidth]{sim_hit_vs_capacity.pdf}
	58	+\caption{Trace-driven simulation on the v2 held-out prompts: per-layer hit rate (left) and modelled stall per token (right)
	59	+by slot capacity and predictor. Generated by \code{scripts/make\_figures.py} from \code{benchmarks/processed/gateC/}.}
	60	+\label{fig:sim}
	61	+\end{figure}
	62	+
55	63	\subsection{Prefetching (H1)}
56	64	\label{sec:results-prefetch}
57	65	With the one-layer horizon the online prefetcher raised the hit rate from 0.48 to 0.63 (general) and from 0.67 to 0.73
		`@@ -63,6 +71,15 @@ Depth-$d$ and token-horizon prefetching on the cold-NVMe tier are reported in Ta`
63	71
64	72	\input{tables/cold}
65	73
	74	+\begin{figure}[t]
	75	+\centering
	76	+\includegraphics[width=\textwidth]{capacity_sweep.pdf}
	77	+\caption{Decode throughput of every single-node configuration measured on the laptop (median over repeats, bar = range):
	78	+mlx-lm, \titan{} all-resident with and without the synchronisation-free path, bounded residency on the page-cache tier, and the
	79	+cold-NVMe tier with and without prefetch. Source files are named in brackets.}
	80	+\label{fig:capacity}
	81	+\end{figure}
	82	+
66	83	\subsection{Medium model (Qwen3-30B-A3B)}
67	84	On M3U96a (M3 Ultra, 96~GB) the stacked-layout registry reads byte-equal slices, logits are bit-identical to mlx-lm at 128, 64,
68	85	32 and 16 slots per layer and greedy decoding is token-identical at 32 slots
		`@@ -71,5 +88,24 @@ On M3U96a (M3 Ultra, 96~GB) the stacked-layout registry reads byte-equal slices,`
71	88
72	89	\input{tables/qwen3}
73	90
74		−\subsection{Distributed execution (H2, H3), Metal kernels (H4), energy frontier (H5)}
75		−\textsc{pending}: Gates D--F have protocols (Section~\ref{sec:algorithms}, \code{research/protocol.md}) and no logs in this version.
	91	+\subsection{Two-node execution over gigabit Ethernet (H2, H3: first evidence)}
	92	+\label{sec:results-twonode}
	93	+With M3U96a as coordinator and M3U96b as worker (OLMoE, 32 slots, page-cached checkpoint on both, held-out general and code
	94	+prompts, \code{benchmarks/raw/gateE/two\_node\_m3u96a\_m3u96b.jsonl}), executing the odd-numbered experts of every layer on the
	95	+worker keeps the logits bit-identical to mlx-lm but reduces decode throughput from 70 / 87~\tps{} (local control) to
	96	+15 / 15~\tps{} (Figure~\ref{fig:twonode}). The remote wait is 2.5~s per 64 tokens, i.e.\ about 2.5~ms per layer: a 0.5--0.65~ms TCP
	97	+round trip plus the worker's array creation, kernel dispatch, evaluation and serialisation, paid 16 times per token. This is
	98	+larger than a cold-NVMe demand miss at 16 slots (about 1.2~ms per layer) and twenty times the local compute per layer. On this
	99	+topology the remote tier is therefore a capacity mechanism, not a latency one, exactly as Section~\ref{sec:prefetch-condition}
	100	+anticipates; H2 can only be tested in the regime where a node cannot hold the experts it needs locally.
	101	+
	102	+\begin{figure}[t]
	103	+\centering
	104	+\includegraphics[width=0.6\textwidth]{two_node.pdf}
	105	+\caption{Two Mac Studios over 1~GbE: local bounded residency vs remote execution of half the experts (medians).}
	106	+\label{fig:twonode}
	107	+\end{figure}
	108	+
	109	+\subsection{Metal kernels (H4), energy frontier (H5)}
	110	+\textsc{pending}: Gates D and the energy study have protocols (Section~\ref{sec:algorithms}, \code{research/protocol.md}) and no
	111	+logs in this version.

modified research/LOG.md +8 −0

		`@@ -103,3 +103,11 @@ Chronological, append-only. Times are local (America/Toronto). Every entry names`
103	103	- Cold sweep v3 (in progress, 16 slots): none ≈ 36 tok/s; transition2 31.7–34.1 (stall 0.82–0.92 s vs ≈ 1.2 s), transition3 26.8–28.0,
104	104	prev_rank (m = 16) 18–19 tok/s → deeper horizons cut the stall by ~30 % but the host-side cost of prefetch (installs, GIL contention of
105	105	reader threads, eviction churn) cancels the gain on this small model; to be attributed with per-token timers after the sweep.
	106	+- Two-node result (Gate E, H2/H3 first evidence) — coordinator M3U96a, worker M3U96b, OLMoE, 32 slots, page-cached, 64-token
	107	+ greedy, held-out general/code (`benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl`): local control 70.1 / 87.0 tok/s (median,
	108	+ hit 0.74 / 0.84); odd experts executed remotely: 15.0 / 14.9 tok/s, logits still bit-identical (Δ = 0.0), remote wait
	109	+ 2.5 s per 64 tokens = ≈ 2.5 ms per layer round trip (TCP RTT 0.5–0.65 ms + worker-side array creation, gather_qmm dispatch,
	110	+ eval and serialisation) × 16 layers ≈ 40 ms per token. On this cluster remote execution over 1 GbE therefore costs more per layer
	111	+ than a cold-NVMe demand miss (≈ 1.2 ms per layer at 16 slots) and far more than local compute (0.13 ms): as the formulation
	112	+ predicts, the remote tier cannot be a latency win for a model whose experts fit on local NVMe; it can only add capacity.
	113	+ Worker-side cost breakdown to be measured once the sweep finishes.

added scripts/make_figures.py +79 −0

		`@@ -0,0 +1,79 @@`
	1	+#!/usr/bin/env python3
	2	+"""Generate every figure of the paper from logged measurements (benchmarks/raw, benchmarks/processed). PDF outputs in
	3	+paper/figures/ and figures/. Nothing here is hand-drawn from numbers typed in."""
	4	+import glob, json, os, statistics, sys
	5	+import matplotlib
	6	+matplotlib.use("Agg")
	7	+import matplotlib.pyplot as plt
	8	+ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
	9	+FIG = os.path.join(ROOT, "paper", "figures"); FIG2 = os.path.join(ROOT, "figures")
	10	+os.makedirs(FIG, exist_ok=True); os.makedirs(FIG2, exist_ok=True)
	11	+plt.rcParams.update({"font.size": 9, "axes.spines.top": False, "axes.spines.right": False, "figure.dpi": 150})
	12	+PAL = ["#2E5A88", "#C8553D", "#4F8A5B", "#8E6C8A", "#D9A441", "#5B5B5B", "#7FB2D6"]
	13	+
	14	+def save(fig, name):
	15	+ for d in (FIG, FIG2):
	16	+ fig.savefig(os.path.join(d, name + ".pdf"), bbox_inches="tight")
	17	+ plt.close(fig); print("wrote", name)
	18	+
	19	+def fig_sim_hit_vs_capacity():
	20	+ p = os.path.join(ROOT, "benchmarks/processed/gateC/sim_olmoe_v2_ssd043_m8.json")
	21	+ if not os.path.exists(p): return
	22	+ d = json.load(open(p)); rows = [r for r in d["results"] if r["split"] == "heldout" and r["policy"] == "lru"]
	23	+ preds = []
	24	+ for r in rows:
	25	+ if r["predictor"] not in preds: preds.append(r["predictor"])
	26	+ fig, axes = plt.subplots(1, 2, figsize=(7.2, 2.8))
	27	+ for i, pr in enumerate(preds):
	28	+ rs = sorted([r for r in rows if r["predictor"] == pr], key=lambda r: r["capacity"])
	29	+ axes[0].plot([r["capacity"] for r in rs], [r["hit_rate"] for r in rs], marker="o", ms=3, color=PAL[i % len(PAL)], label=pr)
	30	+ axes[1].plot([r["capacity"] for r in rs], [r["stall_ms_per_token"] for r in rs], marker="o", ms=3, color=PAL[i % len(PAL)], label=pr)
	31	+ axes[0].set_xlabel("slots per layer $C$"); axes[0].set_ylabel("expert hit rate (held-out)"); axes[0].set_ylim(0.4, 1.0)
	32	+ axes[1].set_xlabel("slots per layer $C$"); axes[1].set_ylabel("modelled stall (ms / token)")
	33	+ axes[1].legend(frameon=False, fontsize=7, ncol=2)
	34	+ fig.suptitle(f"OLMoE-1B-7B, trace-driven simulation (t_read {d['t_read_ms']} ms, {d['loaders']} loader)", fontsize=9)
	35	+ save(fig, "sim_hit_vs_capacity")
	36	+
	37	+def fig_capacity_sweep():
	38	+ paths = [os.path.join(ROOT, "benchmarks/raw/gateB/capacity_sweep_laptop/sweep.jsonl"),
	39	+ os.path.join(ROOT, "benchmarks/raw/gateC/prefetch_online_laptop/smoke_lazyinstall.jsonl"),
	40	+ os.path.join(ROOT, "benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl")]
	41	+ recs = []
	42	+ for p in paths:
	43	+ if os.path.exists(p):
	44	+ recs += [dict(json.loads(l), src=os.path.basename(os.path.dirname(p))) for l in open(p) if l.strip()]
	45	+ recs = [r for r in recs if not r.get("warmup")]
	46	+ if not recs: return
	47	+ groups = {}
	48	+ for r in recs:
	49	+ key = (r["src"], r["backend"], r.get("capacity"), r.get("predictor") or "none", r.get("m"), bool(r.get("nocache")), r.get("tag", ""))
	50	+ groups.setdefault(key, []).append(r["gen_tps"])
	51	+ labels, vals, errs, cols = [], [], [], []
	52	+ for k, v in sorted(groups.items(), key=lambda kv: (kv[0][5], -(kv[0][2] or 999), kv[0][3])):
	53	+ src, be, cap, pr, m, cold, tag = k
	54	+ if be == "mlx_lm": lab = "mlx-lm (all resident)"
	55	+ else: lab = f"C={cap} {pr}{' m='+str(m) if pr!='none' else ''}{' cold' if cold else ' page-cache'} [{src.split('_')[0]}{(' '+tag) if tag else ''}]"
	56	+ labels.append(lab); vals.append(statistics.median(v)); errs.append((max(v) - min(v)) / 2); cols.append(PAL[1] if cold else PAL[0])
	57	+ fig, ax = plt.subplots(figsize=(7.2, 0.22 * len(labels) + 1.2))
	58	+ y = range(len(labels)); ax.barh(y, vals, xerr=errs, color=cols, height=0.7, error_kw={"lw": 0.8})
	59	+ ax.set_yticks(list(y)); ax.set_yticklabels(labels, fontsize=7); ax.invert_yaxis(); ax.set_xlabel("decode tok/s (median; bar = range)")
	60	+ ax.set_title("OLMoE-1B-7B 4-bit on M5 Max: residency, tier and prefetch", fontsize=9)
	61	+ save(fig, "capacity_sweep")
	62	+
	63	+def fig_two_node():
	64	+ p = os.path.join(ROOT, "benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl")
	65	+ if not os.path.exists(p): return
	66	+ recs = [json.loads(l) for l in open(p) if l.strip()]; recs = [r for r in recs if not r["warmup"]]
	67	+ groups = {}
	68	+ for r in recs: groups.setdefault((r["capacity"], r["placement"]), []).append(r)
	69	+ fig, ax = plt.subplots(figsize=(4.2, 2.6))
	70	+ labels = []; i = 0
	71	+ for (cap, pl), v in sorted(groups.items()):
	72	+ tps = [x["gen_tps"] for x in v]; rw = [x["remote"]["remote_wait_s"] / max(1, x["gen_tokens"]) * 1e3 if x.get("remote") else 0 for x in v]
	73	+ ax.bar(i, statistics.median(tps), color=PAL[0 if pl == "none" else 1]); labels.append(f"C={cap}\n{pl}"); i += 1
	74	+ ax.set_xticks(range(i)); ax.set_xticklabels(labels, fontsize=7); ax.set_ylabel("decode tok/s (median)")
	75	+ ax.set_title("OLMoE: M3U96a local vs odd experts on M3U96b (1 GbE)", fontsize=8)
	76	+ save(fig, "two_node")
	77	+
	78	+if __name__ == "__main__":
	79	+ fig_sim_hit_vs_capacity(); fig_capacity_sweep(); fig_two_node()
	80

spb/wp33_uqo

Figures (sim, capacity sweep, two-node), paper results v0.1 with two-node evidence, LOG

Theme