[0.9.1][Fixbug] Fix num_hidden_layers on Qwen2-Audio and Qwen2.5-Omni #2656
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces support for Qwen2-Audio and Qwen2.5-Omni models by fixing how num_hidden_layers is retrieved from the model configuration and adding necessary parameters for audio processing. It also adds new dependencies and tests for audio models. My review identifies a critical bug in the new recursive search function that could fail for certain configuration structures, and a minor but potentially problematic formatting issue in the development requirements file. I've provided suggestions to fix both issues.
zhangxinyuehfad
force-pushed
the
zxy_v0.9.1
branch
from
5e680c7 to
8a68388
Compare
zhangxinyuehfad
force-pushed
the
zxy_v0.9.1
branch
2 times, most recently
from
a7907d2 to
dcfe89d
Compare
github-actions
bot
added
documentation
|
please fix the merge conflic |
|
This pull request has conflicts, please resolve those before we can evaluate the pull request. |
Signed-off-by: hfadzxy <starmoon_zhang@163.com>
zhangxinyuehfad
force-pushed
the
zxy_v0.9.1
branch
from
dcfe89d to
a956298
Compare
done |
|
This pull request has conflicts, please resolve those before we can evaluate the pull request. |
2 participants
What this PR does / why we need it?
audio_feature_lengthsanduse_audio_in_videowhen V0Does this PR introduce any user-facing change?
Users can use Qwen2-Audio and Qwen2.5-Omni in V1 or V0
How was this patch tested?
CI passed with new added test.