Diff Attention out_proj dimension fix #1636
Open
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Subtle potential issue with
self.out_proj = nn.Linear(embed_dim, embed_dim)
working onattn.reshape(bsz, tgt_len, self.num_heads * 2 * self.head_dim
in the rare case thatembed_dim != self.num_heads * 2 * self.head_dim
Small fix in the init of the base and flash attention implementations.