{
  "timestamp_utc": "2026-10-30T03:22:24.245610Z",
  "models": {
    "qwen3b": {
      "repo_id": "mlx-community/Qwen2.5-3B-Instruct-bf16",
      "snapshot_path": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--Qwen2.5-3B-Instruct-bf16/snapshots/bec1ad658e380044bbfe06d4a2268801d3379a11",
      "revision": "bec1ad658e380044bbfe06d4a2268801d3379a11",
      "model_type": "qwen2",
      "num_layers": 36,
      "num_kv_heads": 2,
      "num_attention_heads": 16,
      "head_dim": 128,
      "hidden_size": 2048,
      "vocab_size": 151936,
      "max_position_embeddings": 32768,
      "rope_theta": 1000000.0,
      "rope_scaling": null,
      "sliding_window": 32768,
      "weight_dtype": "bfloat16",
      "weight_quantization": null,
      "tokenizer_hash": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b",
      "chat_template_hash": "d5495a1e5db0611132a97e46a65dbb64a642a499421228b9c8b93229097fa9a4",
      "config_hash": "9f4d0bfcc2df00e16cd130a48c908ab5c5a982d9c656c670f072ecd0d034010f",
      "weight_files_sha256": {
        "model-00001-of-00002.safetensors": "feb3c8c4721f4503d07509e59c1768fe90804340f1d28d9c2aa406215eb6490b",
        "model-00002-of-00002.safetensors": "77b5846eb11fecb6eb27fb70bcc0d0c73a57b8443d01ebcaa558f4066eb34b70"
      },
      "mlx_version": "0.32.4",
      "mlx_lm_version": "0.32.2",
      "bos_token_id": null,
      "eos_token_ids": [
        151645
      ],
      "probe_answer": "ZR-77Q-0931",
      "probe_ok": true,
      "load_ms": 852.6365419384092,
      "context_lengths": {
        "2048": {
          "supported": true,
          "prefill_ms": 528.9326659403741,
          "cache_bytes": 75497472,
          "peak_bytes": 7054786604
        },
        "8192": {
          "supported": true,
          "prefill_ms": 967.8050000220537,
          "cache_bytes": 301989888,
          "peak_bytes": 7323246656
        },
        "16384": {
          "supported": true,
          "prefill_ms": 2285.5356669751927,
          "cache_bytes": 603979776,
          "peak_bytes": 7491051584
        }
      },
      "memory_after_probe": {
        "mlx_active_bytes": 6171877384,
        "mlx_peak_bytes": 7491051584,
        "mlx_cache_bytes": 3520642368,
        "process_rss_bytes": 6603407360
      }
    },
    "llama3b": {
      "repo_id": "mlx-community/Llama-3.2-3B-Instruct-bf16",
      "snapshot_path": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--Llama-3.2-3B-Instruct-bf16/snapshots/6d88ba43024fef71b10e52e101c7cd4598322601",
      "revision": "6d88ba43024fef71b10e52e101c7cd4598322601",
      "model_type": "llama",
      "num_layers": 28,
      "num_kv_heads": 8,
      "num_attention_heads": 24,
      "head_dim": 128,
      "hidden_size": 3072,
      "vocab_size": 128256,
      "max_position_embeddings": 131072,
      "rope_theta": 500000.0,
      "rope_scaling": {
        "factor": 32.0,
        "high_freq_factor": 4.0,
        "low_freq_factor": 1.0,
        "original_max_position_embeddings": 8192,
        "rope_type": "llama3"
      },
      "sliding_window": null,
      "weight_dtype": "bfloat16",
      "weight_quantization": null,
      "tokenizer_hash": "b9d8872cf3a99da99e1fc4b0f71232a451632dc49ca4c6771549b4efa5c1c763",
      "chat_template_hash": "5816fce10444e03c2e9ee1ef8a4a1ea61ae7e69e438613f3b17b69d0426223a4",
      "config_hash": "2b13c35350c57b0c3130063f944cd3a4353c0463fb664e80172d479fb451f923",
      "weight_files_sha256": {
        "model-00001-of-00002.safetensors": "b475af8535933afaf9393b30c6bcb4bfede59586b16ecdd010e7da8a83dfc95b",
        "model-00002-of-00002.safetensors": "9b2903fa2b3728423f6442c0dc667db645906fb113c9d0cd794faeb820cc82d3"
      },
      "mlx_version": "0.32.4",
      "mlx_lm_version": "0.32.2",
      "bos_token_id": 128000,
      "eos_token_ids": [
        128008,
        128001,
        128009
      ],
      "probe_answer": "ZR-77Q-0931",
      "probe_ok": true,
      "load_ms": 2813.813167042099,
      "context_lengths": {
        "2048": {
          "supported": true,
          "prefill_ms": 251.04662496596575,
          "cache_bytes": 234881024,
          "peak_bytes": 7142741028
        },
        "8192": {
          "supported": true,
          "prefill_ms": 1006.7105001071468,
          "cache_bytes": 939524096,
          "peak_bytes": 7851602992
        },
        "16384": {
          "supported": true,
          "prefill_ms": 2449.0344589576125,
          "cache_bytes": 1879048192,
          "peak_bytes": 8778576936
        }
      },
      "memory_after_probe": {
        "mlx_active_bytes": 6425506824,
        "mlx_peak_bytes": 8778576936,
        "mlx_cache_bytes": 8951880172,
        "process_rss_bytes": 6437617664
      }
    },
    "qwen1b5": {
      "repo_id": "mlx-community/Qwen2.5-1.5B-Instruct-bf16",
      "snapshot_path": "/Users/simon-pierreboucher/.cache/huggingface/hub/models--mlx-community--Qwen2.5-1.5B-Instruct-bf16/snapshots/4ae77cb209f06199b8df1c94e21ff341332a3a89",
      "revision": "4ae77cb209f06199b8df1c94e21ff341332a3a89",
      "model_type": "qwen2",
      "num_layers": 28,
      "num_kv_heads": 2,
      "num_attention_heads": 12,
      "head_dim": 128,
      "hidden_size": 1536,
      "vocab_size": 151936,
      "max_position_embeddings": 32768,
      "rope_theta": 1000000.0,
      "rope_scaling": null,
      "sliding_window": 32768,
      "weight_dtype": "bfloat16",
      "weight_quantization": null,
      "tokenizer_hash": "5c5a439ba2677fe28e1dea9b22371b270ecd5e34acbc849bc619689605f9547b",
      "chat_template_hash": "d5495a1e5db0611132a97e46a65dbb64a642a499421228b9c8b93229097fa9a4",
      "config_hash": "a1cd2dc97c13a2dce95741f912fa212cfa321260c42627202a122b80c68a52c9",
      "weight_files_sha256": {
        "model.safetensors": "073400b311d049b6379b8f3f7ad5a0258268031dd009b99389cfd4030144850a"
      },
      "mlx_version": "0.32.4",
      "mlx_lm_version": "0.32.2",
      "bos_token_id": null,
      "eos_token_ids": [
        151645
      ],
      "probe_answer": "ZR-77Q-0931",
      "probe_ok": true,
      "load_ms": 500.1337080029771,
      "context_lengths": {
        "2048": {
          "supported": true,
          "prefill_ms": 103.63012494053692,
          "cache_bytes": 58720256,
          "peak_bytes": 3775302700
        },
        "8192": {
          "supported": true,
          "prefill_ms": 493.9766660099849,
          "cache_bytes": 234881024,
          "peak_bytes": 3993431104
        },
        "16384": {
          "supported": true,
          "prefill_ms": 1201.0092079872265,
          "cache_bytes": 469762048,
          "peak_bytes": 4261899328
        }
      },
      "memory_after_probe": {
        "mlx_active_bytes": 3087428616,
        "mlx_peak_bytes": 4261899328,
        "mlx_cache_bytes": 2816668640,
        "process_rss_bytes": 3474096128
      }
    }
  }
}