Fix ROPE embeddings for LLama#29138
Closed
zucchini-nlp wants to merge 2 commits intohuggingface:mainfrom
Closed
Conversation
5 tasks
Collaborator
ArthurZucker
left a comment
There was a problem hiding this comment.
Thanks for opening the PR 🤗 sorry maybe too early to review 😉
| def forward(self, x, position_ids, seq_len=None): | ||
| # x: [bs, num_attention_heads, seq_len, head_size] | ||
| freqs = (self.inv_freq[:, None].float().expand(-1, position_ids.shape[0]) @ (position_ids.float())).t() | ||
| freqs = torch.einsum("i,bl->bli", self.inv_freq.float(), position_ids.float()) |
Collaborator
There was a problem hiding this comment.
No einsum should not be used, for 1 we like to have explicit computation, and secondly, this will fail with different dtypes. Bfloat float etc.
Member
Author
There was a problem hiding this comment.
I just saw it is fixed on main. Never mind, I'll close the PR
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What does this PR do?
This test failed on my PR and I checked to see the reason. I found that the changes introduced to make llama compile compatible are causing the issue.
The fixes here are tested with fullgraph compile, compilation is still working without graph breaks. Additionally, the failing test was run 500 times. I found that aside from rope embeddings, the cause of test failure was in SDPA attention. I cannot say what is the reason exactly, but running the test 500 times gives 95% success in SDPA and 100% success in eager, using the fixes introduced in this PR. Prior to these fixes, the tests were running with 90% success for both attentions.
Before submitting
Pull Request section?
to it if that's the case.
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.
@gante