May 1 11:22 Delete _patch_diagnostics.py 11:20 Fix: backward() inside @torch .no_grad() — use torch.enable_grad() for dense gradient computation 11:18 Fix: compute_relaxer_diagnostics called backward inside no_grad context" 07:59 Upload exp5_mechanism.py with huggingface_hub 07:31 Upload exp4_relaxation.py with huggingface_hub 06:09 Upload results/exp3.json with huggingface_hub 06:09 Upload results/exp2.json with huggingface_hub 06:09 Upload results/exp1.json with huggingface_hub 06:09 Upload ablations_lite.py with huggingface_hub 06:08 Major revision: add phantom momentum ablation, compute-matched baselines, multi-seed predictor accuracy Apr 30 23:01 Upload ablations.py with huggingface_hub 23:00 Add definitive ablation suite addressing all critique gaps 22:36 Add sparse transformer v19 with Triton-backed KNN scheduler and various backward modes. Includes utilities for synthetic data generation and model training. Implements chunked sparse updates and integrates with existing sparse linear layers. 19:59 Upload triton_v2.py with huggingface_hub 19:53 Add complete results with all measured numbers 19:52 Upload e2e_bench.py with huggingface_hub 19:51 Upload e2e_full.py with huggingface_hub 19:51 Upload triton_sparse.py with huggingface_hub 19:51 Add all experiment code 19:49 Upload README.md 19:38 initial commit