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%

1.4 KB · 24 lines text

Raw Blame History

1loading reference 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, 5306.26it/s]3  generated code4  generated math5  generated chat6  generated french7  generated english8  generated reasoning9reference generation done in 49s (48 trajectories)10quantizing to 2-bit (group 64) …11
Fetching 9 files:   0%|          | 0/9 [00:00<?, ?it/s]
Fetching 9 files: 100%|██████████| 9/9 [00:00<00:00, 4840.82it/s]12  2-bit: agree=0.0296 AUROC=0.605 esc@99%=0.9897460937513quantizing to 3-bit (group 64) …14
Fetching 9 files:   0%|          | 0/9 [00:00<?, ?it/s]
Fetching 9 files: 100%|██████████| 9/9 [00:00<00:00, 6333.68it/s]15  3-bit: agree=0.7310 AUROC=0.852 esc@99%=0.601725260416666616quantizing to 4-bit (group 64) …17
Fetching 9 files:   0%|          | 0/9 [00:00<?, ?it/s]
Fetching 9 files: 100%|██████████| 9/9 [00:00<00:00, 4795.32it/s]18  4-bit: agree=0.8740 AUROC=0.898 esc@99%=0.365885416666666719quantizing to 8-bit (group 64) …20
Fetching 9 files:   0%|          | 0/9 [00:00<?, ?it/s]
Fetching 9 files: 100%|██████████| 9/9 [00:00<00:00, 6190.35it/s]21  8-bit: agree=0.9803 AUROC=0.975 esc@99%=0.0511067708333333362223wrote /Users/simon-pierreboucher/Desktop/localvm-research/results/expG_decision_stability/20260812T042744Z/results.json24