1.88 MB

Ctrl+K

1 contributor

History: 21 commits

theapemachine

Delete _patch_diagnostics.py

746fa5b verified 6 days ago

experiments
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. 7 days ago
paper
Major revision: add phantom momentum ablation, compute-matched baselines, multi-seed predictor accuracy 6 days ago
results
Upload results/exp3.json with huggingface_hub 6 days ago
.gitattributes

1.52 kB
initial commit 7 days ago
README.md

1.19 kB
Upload README.md 7 days ago
RESULTS.md

5.67 kB
Add complete results with all measured numbers 7 days ago
ablations.py

36.8 kB
Upload ablations.py with huggingface_hub 7 days ago
ablations_lite.py

15.9 kB
Upload ablations_lite.py with huggingface_hub 6 days ago
exp4_relaxation.py

18.1 kB
Upload exp4_relaxation.py with huggingface_hub 6 days ago
exp5_mechanism.py

22.2 kB
Fix: backward() inside @torch.no_grad() — use torch.enable_grad() for dense gradient computation 6 days ago
sparse_transformer_v18_fast_knn.py

19.8 kB
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. 7 days ago
sparse_transformer_v18_fast_knn_triton.py

35.7 kB
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. 7 days ago