Skip to content

Disable gemma2 and qwen2_vl tests #288

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

Merged
merged 4 commits into from
Oct 1, 2024

Conversation

shimizust
Copy link
Collaborator

@shimizust shimizust commented Oct 1, 2024

Summary

  • Gemma2 convergence tests were erroneously passing before due to all tensors having NaN values. Using attn_implementation="eager" fixes the NaNs, but results don't pass convergence criteria. Will need to investigate further, but skipping these for now.
  • The discrepancy was revealed after transformers 4.44.2 -> 4.45.1 update which seems to have fixed to fall back on eager attn implementation
  • Qwen2_VL convergence tests are failing and also require access to internet (HF Hub), so having a hard time debugging. Skipping this for now.

Testing Done

  • Hardware Type: A100
  • run make test to ensure correctness
  • run make checkstyle to ensure code style
  • run make test-convergence to ensure convergence

ByronHsu
ByronHsu previously approved these changes Oct 1, 2024
@shimizust shimizust changed the title Disable gemma2 tests Disable gemma2 and qwen2_vl tests Oct 1, 2024
@@ -225,6 +225,9 @@ def run_mini_model_multimodal(
return {"loss": loss_list, "logits": output.logits, "model": model}


@pytest.mark.skip(
reason="This test needs to be fixed and work without access to HF Hub"
)
Copy link
Collaborator

@tyler-romero tyler-romero Oct 1, 2024

Choose a reason for hiding this comment

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

@ByronHsu ByronHsu merged commit e62fc98 into linkedin:main Oct 1, 2024
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants