[ydshieh]

What does this PR do?

• tests/models/qwen2/test_modeling_qwen2.py::Qwen2IntegrationTest::test_model_450m_logits:

  • failed due to 6217adc

• tests/models/smolvlm/test_modeling_smolvlm.py::SmolVLMForConditionalGenerationIntegrationTest::test_integration_test_video:

  • passed in 42fe769 (SmolVLM test fixes #40275)
  • failed in d7fe311 (Fix idefics3 vision embeddings indices dtype #40360), but remain the same error
  • error changed since 8843333 (🚨🚨 Refactor Image Processors to support different backends #43514)): the input pixel_values may be somehow large tensor(0.3672, device='cuda:0', dtype=torch.bfloat16)
  • simply update the expected outputs seems to fine and reasonable (considering the changes in the above 2 PRs and the difference of the outputs). No, should a fix, see Fix SmolVLM video processor resize using wrong interpolation after backend refactor #45258

• a

  • c

• b

• b

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[ArthurZucker]

Closing: this PR has no changed files left (0 additions, 0 deletions).