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

ci: update Transformers to v4.49.0 #1282

Draft
wants to merge 1 commit into
base: main
Choose a base branch
from

Conversation

dvrogozh
Copy link
Contributor

@dvrogozh dvrogozh commented Jan 13, 2025

Changes:

  • Benchmarking scripts are pruned from Transformers by v4.49.0 due to deprecation. So we don't need to test them anymore.
  • Some cuda specific tests were generalized to cover non-cuda devices which uncovered some issues.
  • Some new tests were added which fail for both cuda and xpu.
  • Few regressions due to changes on Transformers side

Fixed tests:

Commits which added new tests (or enabled previously skipped tests) which fail:

  • huggingface/transformers@23d782e
    • tests/pipelines/test_pipelines_text_generation.py::TextGenerationPipelineTests::test_return_dict_in_generate
    • tests/test_pipeline_mixin.py::TextGenerationPipelineTests::test_return_dict_in_generate
  • huggingface/transformers@2fa876d
    • test_cpu_offload (some of)
    • test_disk_offload_bin (some of)
    • test_disk_offload_safetensors (some of)
    • tests/pipelines/test_pipelines_text_generation.py::TextGenerationPipelineTests::test_small_model_pt_bloom_accelerate
  • huggingface/transformers@be2ac09
    • tests/models/paligemma/test_modeling_paligemma.py::PaliGemmaForConditionalGenerationModelTest::test_generate_compilation_all_outputs
    • tests/models/paligemma2/test_modeling_paligemma2.py::PaliGemma2ForConditionalGenerationModelTest::test_generate_compilation_all_outputs
  • huggingface/transformers@f19135a
    • tests/pipelines/test_pipelines_audio_classification.py::AudioClassificationPipelineTests::test_small_model_pt_fp16
  • huggingface/transformers@1fae54c
    • tests/trainer/test_trainer.py::TrainerIntegrationPrerunTest::test_gradient_accumulation_loss_alignment_with_model_loss
  • huggingface/transformers@15ec971
    • tests/models/qwen2_5_vl/test_processor_qwen2_5_vl.py::Qwen2_5_VLProcessorTest::test_chat_template_video_custom_sampling
    • tests/models/qwen2_5_vl/test_processor_qwen2_5_vl.py::Qwen2_5_VLProcessorTest::test_chat_template_video_special_processing

Regressions:

Signed-off-by: Dmitry Rogozhkin [email protected]

@dvrogozh dvrogozh marked this pull request as ready for review January 13, 2025 22:00
@dvrogozh dvrogozh marked this pull request as draft January 14, 2025 01:02
@dvrogozh dvrogozh force-pushed the transformers branch 2 times, most recently from 62ed9a5 to fef2644 Compare January 14, 2025 17:34
@dvrogozh
Copy link
Contributor Author

dvrogozh commented Jan 14, 2025

Transformers v4.48.0 has a regression in tests:

# TRANSFORMERS_TEST_DEVICE_SPEC=spec.py python3 -m pytest tests/models/marian/test_modeling_marian.py -k backbone

E   ModuleNotFoundError: No module named 'transformers.models.marian.convert_marian_to_pytorch'

I don't see this issue in main at c23a1c193. It might be reasonable to wait for v4.49.0 or bring this issue to HF if it will persist.

@dvrogozh
Copy link
Contributor Author

Transformers v4.48.0 has a regression in tests:

Same issue observed on v4.49.0. I reported it to Transformers project and will exclude the test manually for now:

@dvrogozh dvrogozh changed the title ci: update Transformers to v4.48.0 ci: update Transformers to v4.49.0 Feb 18, 2025
@dvrogozh dvrogozh changed the title ci: update Transformers to v4.49.0 ci: update Transformers to v4.48.3 Feb 19, 2025
@dvrogozh dvrogozh changed the title ci: update Transformers to v4.48.3 ci: update Transformers to v4.49.0 Feb 20, 2025
Changes:
* Benchmarking scripts are pruned from Transformers by v4.49.0 due to
  deprecation. So we don't need to test them anymore.
* Some cuda specific tests were generalized to cover non-cuda devices
  which uncovered some issues.
* Some new tests were added which fail for both cuda and xpu.
* Few regressions due to changes on Transformers side

Fixed tests:
* huggingface/transformers@b912f5e
  * `tests/models/git/test_modeling_git.py::GitModelTest::test_inputs_embeds_matches_input_ids`
* huggingface/transformers@b5aaf87
  * `tests/pipelines/test_pipelines_video_classification.py::VideoClassificationPipelineTests::test_small_model_pt`
  * `tests/test_pipeline_mixin.py::VideoClassificationPipelineTests::test_small_model_pt`
* huggingface/transformers@42c8ccf
  * `tests/generation/test_utils.py::GenerationIntegrationTests::test_generated_length_assisted_generation`
* huggingface/transformers@9fd123a
  * `test_model_parallelization`
  * `test_model_parallel_equal_results`

Commits which added new tests (or enabled previously skipped tests) which fail:
* huggingface/transformers@23d782e
  * `tests/pipelines/test_pipelines_text_generation.py::TextGenerationPipelineTests::test_return_dict_in_generate`
  * `tests/test_pipeline_mixin.py::TextGenerationPipelineTests::test_return_dict_in_generate`
* huggingface/transformers@2fa876d
  * `test_cpu_offload` (some of)
  * `test_disk_offload_bin` (some of)
  * `test_disk_offload_safetensors` (some of)
  * `tests/pipelines/test_pipelines_text_generation.py::TextGenerationPipelineTests::test_small_model_pt_bloom_accelerate`
* huggingface/transformers@be2ac09
  * `tests/models/paligemma/test_modeling_paligemma.py::PaliGemmaForConditionalGenerationModelTest::test_generate_compilation_all_outputs`
  * `tests/models/paligemma2/test_modeling_paligemma2.py::PaliGemma2ForConditionalGenerationModelTest::test_generate_compilation_all_outputs`
* huggingface/transformers@f19135a
  * `tests/pipelines/test_pipelines_audio_classification.py::AudioClassificationPipelineTests::test_small_model_pt_fp16`
* huggingface/transformers@1fae54c
  * `tests/trainer/test_trainer.py::TrainerIntegrationPrerunTest::test_gradient_accumulation_loss_alignment_with_model_loss`
* huggingface/transformers@15ec971
  * `tests/models/qwen2_5_vl/test_processor_qwen2_5_vl.py::Qwen2_5_VLProcessorTest::test_chat_template_video_custom_sampling`
  * `tests/models/qwen2_5_vl/test_processor_qwen2_5_vl.py::Qwen2_5_VLProcessorTest::test_chat_template_video_special_processing`

Regressions:
* huggingface/transformers@365fecb
  * `tests/generation/test_utils.py::GenerationIntegrationTests::test_encoder_decoder_generate_attention_mask`
* huggingface/transformers@da334bc
  * `tests/generation/test_utils.py::GenerationIntegrationTests::test_generate_input_features_as_encoder_kwarg`
* huggingface/transformers@bcfc9d7
  * `tests/models/llava/test_modeling_llava.py::LlavaForConditionalGenerationModelTest::test_config`
* huggingface/transformers#36267
  * `tests/utils/test_import_utils.py`
* huggingface/transformers#36267
  * `tests/models/marian/test_modeling_marian.py`

Signed-off-by: Dmitry Rogozhkin <[email protected]>
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.

1 participant