Skip to content

Conversation

@xuechendi
Copy link
Collaborator

The reason qwen2.5-vl failed after PR24444 is because that HPU is kept using forward_native for MRotaryEmbedding
After PR24444 merged, it firstly time will go forward_oot instead of forward_native, while forward_oot is not implemented correctly.

Temporary switch to use forward_native in this PR

Signed-off-by: Chendi.Xue <chendi.xue@intel.com>
Signed-off-by: Chendi.Xue <chendi.xue@intel.com>
@PatrykWo
Copy link
Collaborator

/run-gaudi-tests

@kzawora-intel kzawora-intel merged commit bc79342 into vllm-project:main Sep 12, 2025
2 checks passed
@attafosu attafosu mentioned this pull request Sep 12, 2025
xuechendi added a commit that referenced this pull request Sep 16, 2025
- HPU Mrope implementation had a bug which was exposed by
vllm-project/vllm#24444
- Initial workaround was to use the default implementation:
#162
- This PR fixes the bug in the HPU mrope

---------

Signed-off-by: attafosu <thomas.atta-fosu@intel.com>
Co-authored-by: Chendi.Xue <chendi.xue@intel.com>
slokesha pushed a commit to slokesha/vllm-gaudi that referenced this pull request Sep 24, 2025
- HPU Mrope implementation had a bug which was exposed by
vllm-project/vllm#24444
- Initial workaround was to use the default implementation:
vllm-project#162
- This PR fixes the bug in the HPU mrope

---------

Signed-off-by: attafosu <thomas.atta-fosu@intel.com>
Co-authored-by: Chendi.Xue <chendi.xue@intel.com>
Signed-off-by: slokesha <slokeshappa@habana.ai>
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.

3 participants