diff options
| author | brkirch <brkirch@users.noreply.github.com> | 2023-07-25 07:03:06 +0000 |
|---|---|---|
| committer | brkirch <brkirch@users.noreply.github.com> | 2023-08-13 14:06:25 +0000 |
| commit | 2489252099c299bed49a9d4a39a4ead73b6b6f10 (patch) | |
| tree | 98950f6d50e0a93d08dcb7134db93c49777c460f | |
| parent | 87dd685224b5f7dbbd832fc73cc08e7e470c9f28 (diff) | |
| download | stable-diffusion-webui-gfx803-2489252099c299bed49a9d4a39a4ead73b6b6f10.tar.gz stable-diffusion-webui-gfx803-2489252099c299bed49a9d4a39a4ead73b6b6f10.tar.bz2 stable-diffusion-webui-gfx803-2489252099c299bed49a9d4a39a4ead73b6b6f10.zip | |
`torch.empty` can create issues; use `torch.zeros`
For MPS, using a tensor created with `torch.empty()` can cause `torch.baddbmm()` to include NaNs in the tensor it returns, even though `beta=0`. However, with a tensor of shape [1,1,1], there should be a negligible performance difference between `torch.empty()` and `torch.zeros()` anyway, so it's better to just use `torch.zeros()` for this and avoid unnecessarily creating issues.
| -rw-r--r-- | modules/sub_quadratic_attention.py | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/modules/sub_quadratic_attention.py b/modules/sub_quadratic_attention.py index 497568eb..ae4ee4bb 100644 --- a/modules/sub_quadratic_attention.py +++ b/modules/sub_quadratic_attention.py @@ -58,7 +58,7 @@ def _summarize_chunk( scale: float, ) -> AttnChunk: attn_weights = torch.baddbmm( - torch.empty(1, 1, 1, device=query.device, dtype=query.dtype), + torch.zeros(1, 1, 1, device=query.device, dtype=query.dtype), query, key.transpose(1,2), alpha=scale, @@ -121,7 +121,7 @@ def _get_attention_scores_no_kv_chunking( scale: float, ) -> Tensor: attn_scores = torch.baddbmm( - torch.empty(1, 1, 1, device=query.device, dtype=query.dtype), + torch.zeros(1, 1, 1, device=query.device, dtype=query.dtype), query, key.transpose(1,2), alpha=scale, |