Skip to content

GitLab

"scripts/wan22/run_wan22_moe_flf2v.sh" did not exist on "7fde70631ce0b7f67cb2476b44934cde93a2944d"
Switch branch/tag
  1. 08 Jul, 2023 1 commit
    • 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
  3. 04 Jul, 2023 1 commit
  5. 03 Jul, 2023 2 commits
  7. 02 Jul, 2023 1 commit
  9. 02 Jun, 2023 1 commit
  11. 30 May, 2023 2 commits
  13. 27 May, 2023 1 commit
  15. 19 May, 2023 1 commit
  17. 06 May, 2023 2 commits
  19. 05 May, 2023 1 commit
  21. 21 Apr, 2023 4 commits
  23. 19 Apr, 2023 1 commit
  25. 18 Apr, 2023 2 commits
  27. 14 Apr, 2023 1 commit
  29. 13 Apr, 2023 4 commits
  31. 12 Apr, 2023 1 commit
  33. 31 Mar, 2023 2 commits
  35. 29 Mar, 2023 2 commits
  37. 22 Mar, 2023 2 commits
  39. 15 Mar, 2023 2 commits
  41. 14 Mar, 2023 1 commit
  43. 23 Jan, 2023 1 commit
  45. 19 Jan, 2023 1 commit
  47. 18 Jan, 2023 2 commits
  49. 16 Jan, 2023 1 commit