Fix CohereASR training loss double shift#46958
Open
w-Jessamine wants to merge 1 commit into
Open
Conversation
Contributor
|
[For maintainers] Suggested jobs to run (before merge) run-slow: cohere_asr |
eustlb
reviewed
Jul 6, 2026
eustlb
left a comment
eustlb
left a comment
Collaborator
There was a problem hiding this comment.
#47090 is the correct fix IMO, see #47090 (review) for explaination
cc @vasqu @SunMarc if you agree
5 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Fixes #46894.
Supersedes #46947, which was closed after the fork branch was renamed.
CohereASR right-shifts
labelsintodecoder_input_ids, then usedself.loss_function(ForCausalLMLoss) to compute the training loss. That loss shifts labels internally again, so training targets were effectivelylabels[..., 1:].This switches CohereASR to a plain
CrossEntropyLoss, matching the Moonshine fix and seq2seq models that already align decoder inputs from labels before computing loss.Tests
PYTHONPATH=src /tmp/transformers-pr-venv/bin/python -m pytest tests/models/cohere_asr/test_modeling_cohere_asr.py::CohereAsrModelTest::test_training_loss_no_double_shift -qCI recap
Dashboard: View test results in Grafana
Latest run: 28370058252
Result:
success| Grafana metrics are not available yet.