Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Test HF's implementation of Qwen 2 model #1406

Merged
merged 8 commits into from
Nov 13, 2024
Merged

Test HF's implementation of Qwen 2 model #1406

merged 8 commits into from
Nov 13, 2024

Conversation

IvanYashchuk
Copy link
Collaborator

This PR adds a test to verify that the computed cross-entropy loss matches between HF's implementation and application of Thunder on this model.

Currently backward doesn't work because it hits a bug in nvFuser (NVIDIA/Fuser#871 (comment)).

Ref #1286.

@IvanYashchuk
Copy link
Collaborator Author

Interesting:

        # https://github.com/NVIDIA/Fuser/issues/871#issuecomment-2461562917
>       with pytest.raises(RuntimeError, match="Found two different const extents in the same set"):
E       Failed: DID NOT RAISE <class 'RuntimeError'>

thunder/tests/test_networks.py:443: Failed

thunder/tests/test_networks.py Outdated Show resolved Hide resolved
@riccardofelluga
Copy link
Collaborator

It seems like the backward issue is not there anymore. As of time of writing, both CI and the latest container do not seem to present the issue.

Copy link
Collaborator

@t-vi t-vi left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@t-vi t-vi enabled auto-merge (squash) November 13, 2024 10:09
@t-vi t-vi merged commit 2b8284d into main Nov 13, 2024
41 checks passed
@t-vi t-vi deleted the test-hf-qwen2 branch November 13, 2024 10:09
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
huggingface For supporting HF models
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants