spb/localvm-research Public License
Running LLMs larger than memory on a consumer Mac — falsification-driven research: margin-gated deferred refinement, out-of-core verification on Apple Silicon. TR-01 published.
Python 63.2%
JavaScript 23.5%
CSS 11.8%
Shell 0.9%
Makefile 0.5%
1loading mlx-community/Qwen3-1.7B-bf16 …2
Fetching 9 files: 0%| | 0/9 [00:00<?, ?it/s]
Fetching 9 files: 100%|██████████| 9/9 [00:00<00:00, 5729.92it/s]348 trajectories in 49s4precomputing 4-bit weights …5all-4-bit floor:6 all_4bit: agree=0.8753 KL=0.19827degrade-one (rest bf16):8 degrade_g0_L0-3: agree=0.9473 KL=0.03449 degrade_g1_L4-7: agree=0.9631 KL=0.023210 degrade_g2_L8-11: agree=0.9505 KL=0.032811 degrade_g3_L12-15: agree=0.9460 KL=0.039912 degrade_g4_L16-19: agree=0.9440 KL=0.042213 degrade_g5_L20-23: agree=0.9601 KL=0.017614 degrade_g6_L24-27: agree=0.9604 KL=0.016015repair-one (rest 4-bit):16 repair_g0_L0-3: agree=0.8880 KL=0.176517 repair_g1_L4-7: agree=0.8864 KL=0.173818 repair_g2_L8-11: agree=0.8815 KL=0.171519 repair_g3_L12-15: agree=0.8905 KL=0.155120 repair_g4_L16-19: agree=0.8892 KL=0.154621 repair_g5_L20-23: agree=0.8838 KL=0.182422 repair_g6_L24-27: agree=0.8815 KL=0.187923repair-top-k (best groups bf16):24 repair_top2_groups_[3, 4]: agree=0.9053 KL=0.112625 repair_top3_groups_[0, 3, 4]: agree=0.9160 KL=0.08652627wrote /Users/simon-pierreboucher/Desktop/localvm-research/results/expF_error_accumulation/20260812T043945Z/results.json28