Bug: Critical accuracy bugs for model_type=qwen2: no causal attention and wrong tokenizer #762
+18
−3
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Outdated suggestions cannot be applied.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.
What does this PR do?
What happend:
Qwen2Flash was written to support only
Alibaba-NLP/gte-Qwen2-1.5B-instruct, which was trained and has reference inference code with causal=False (aka use_bidirectional_attention=True). Later on I asked to add a flag in the config to actually read this.https://huggingface.co/Alibaba-NLP/gte-Qwen2-1.5B-instruct/discussions/28
Now, we have also newer models, e.g. jina which are built on qwen2 but with:
The new models are currently loaded, but the output embeddings will be incorrect.
Fixes # (issue)
Before submitting
instasnapshots?Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.