ViperEkura 6f49738991 feat: auto-select best available attention backend ...

- Default backend resolves to highest-priority available: flash -> cuda -> torch
- attention() falls back per-call for training/fp32/unsupported head_dim
- Re-apply index_copy_ for CUDA KV cache writes (index_put_ race mitigation)

2026-08-07 13:48:59 +08:00

..

conftest.py

fix: broadcast ref/old model state_dict for FSDP

2026-07-31 08:32:22 +08:00

test_backend_equivalence.py

feat: auto-select best available attention backend

2026-08-07 13:48:59 +08:00

test_backend.py

feat: auto-select best available attention backend

2026-08-07 13:48:59 +08:00

test_kernel_mask.py

fix: broadcast ref/old model state_dict for FSDP

2026-07-31 08:32:22 +08:00