Skip to content

GitLab

Switch branch/tag
  1. 01 Aug, 2023 2 commits
  3. 29 Jul, 2023 1 commit
  5. 28 Jul, 2023 3 commits
  7. 27 Jul, 2023 1 commit
  9. 26 Jul, 2023 4 commits
  11. 23 Jul, 2023 10 commits
  13. 22 Jul, 2023 1 commit
  15. 21 Jul, 2023 2 commits
  17. 20 Jul, 2023 2 commits
  19. 19 Jul, 2023 4 commits
  21. 18 Jul, 2023 2 commits
  23. 17 Jul, 2023 2 commits
  25. 16 Jul, 2023 1 commit
  27. 15 Jul, 2023 3 commits
  29. 08 Jul, 2023 2 commits
    • Tri Dao's avatar
      Merge pull request #299 from proger/rotary-inference-mode · 72ad03ea
      Tri Dao authored 
      rotary: update cos/sin cache when switching from inference mode
      72ad03ea
    • Volodymyr Kyrylov's avatar
      rotary: update cos/sin cache when switching from inference mode · 70ab266a
      Volodymyr Kyrylov authored 
      This resolves RuntimeErrors after running evaluation in inference mode:

```
  File "/home/proger/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/proger/.local/lib/python3.10/site-packages/flash_attn/modules/mha.py", line 492, in forward
    qkv = self.rotary_emb(qkv)
  File "/home/proger/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/proger/.local/lib/python3.10/site-packages/flash_attn/layers/rotary.py", line 229, in forward
    return apply_rotary_emb_qkv_(
  File "/home/proger/.local/lib/python3.10/site-packages/torch/autograd/function.py", line 506, in apply
    return super().apply(*args, **kwargs)  # type: ignore[misc]
RuntimeError: Inference tensors cannot be saved for backward. To work around you can make a clone to get a normal tensor and use it in autograd.
```
      70ab266a