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

Upgrade hydra-node fork to latest transformers #2

Merged
merged 9 commits into from
Sep 10, 2024
Merged

Conversation

dmarx
Copy link
Collaborator

@dmarx dmarx commented Sep 5, 2024

This is a dependency for Mistral-7B and any other new models. https://coreweave.slack.com/archives/C03L6UD9EJ1/p1725489273247449?thread_ts=1725391234.309029&cid=C03L6UD9EJ1

  • sync coreweave/transformers
  • validate baseline tests/generation/test_streamers.py
    • my plan is to do this TDD style, so I want to make sure all of these tests pass before I start changing anything
    • 5 passed, 10 warnings in 6.15s
  • validate baseline pytest tests/generation
    • 1 failed, 114 passed, 89 skipped, 85 warnings in 50.52s
    • FAILED tests/generation/test_configuration_utils.py::GenerationConfigTest::test_initialize_new_kwargs - AttributeError: 'GenerationConfig' object has no attribute 'get_text_config'
  • Add target test cases
    • TestOutputIteratorStreamer
      • 23 failed, 5 passed, 50 warnings in 14.42s
    • uh... anything else? Test the transformer_patch maybe?
    • Now that we have these failing test cases, the goal is to get this thing to pass
  • Add new Streamer stuff
    • generation/streamers.py
      • 20 failed, 8 passed, 33 warnings in 13.99s
    • generation/utils.py
      • add _prepare_output()
      • integrate _prepare_output() with all samplers (main output)
        • ._contrastive()
          • 21 failed, 117 passed, 89 skipped, 108 warnings in 59.51s
        • ._sample()
          • 21 failed, 117 passed, 89 skipped, 108 warnings in 54.48s
      • integrate _prepare_output() with all samplers streaming
        • .generate(streamer)
          • 21 failed, 117 passed, 89 skipped, 108 warnings in 54.48s
        • ._sample(streamer)
          • 13 failed, 125 passed, 89 skipped, 108 warnings in 65.16s (0:01:05)
        • ._contrastive(streamer)
          • 9 failed, 129 passed, 89 skipped, 108 warnings in 50.25s
  • Resolve failing tests when output_scores=True
    • Probably need to track down something that has changed since I first implemented this
    • We only need to be able to stream token_ids, scores, and logits. Disabled tests for streaming other attributes.

@dmarx dmarx changed the base branch from main to hydra-node September 5, 2024 01:13
@dmarx dmarx merged commit 853b9cc into hydra-node Sep 10, 2024
4 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.

1 participant