Enable multimodal support + qwen2.5-vl #92
Merged
Conversation
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
Contributor
attafosu
commented
Aug 20, 2025
attafosu
commented
- Enables v1 multmodal support
- Enables qwen2.5-vl: Support for MRope
attafosu
requested review from
adobrzyn,
kzawora-intel,
mswiniarsk and
xuechendi
as code owners
Signed-off-by: attafosu <thomas.atta-fosu@intel.com>
Signed-off-by: attafosu <thomas.atta-fosu@intel.com>
* Style formatting Signed-off-by: attafosu <thomas.atta-fosu@intel.com> * Extra mops Signed-off-by: attafosu <thomas.atta-fosu@intel.com> * appease yapf and ruff Signed-off-by: attafosu <thomas.atta-fosu@intel.com> --------- Signed-off-by: attafosu <thomas.atta-fosu@intel.com>
attafosu
force-pushed
the
dev/attafosu/multimodal-qwen2.5-vl
branch
from
5de633f to
d8a23f5
Compare
Contributor
Author
|
/run-gaudi-tests |
Collaborator
|
Only codeowners can request to run Gaudi tests. Contact list: kzawora-intel, xuechendi, mswiniarsk, adobrzyn |
xuechendi
reviewed
Aug 21, 2025
Signed-off-by: attafosu <thomas.atta-fosu@intel.com>
xuechendi
reviewed
Aug 21, 2025
| token_ids = _async_h2d_tensor(token_ids, torch.int32) | ||
| token_positions = _async_h2d_tensor(token_positions, torch.int32) | ||
| if not self.uses_mrope: | ||
| token_positions = _async_h2d_tensor(token_positions, torch.int32) |
Collaborator
There was a problem hiding this comment.
can we not hard_code tensor as HPU in mrope_token_positions = self._align_and_pad_mrope_positions()
So we don't need to add condition for if do H2D for token_positions here?
xuechendi
reviewed
Aug 21, 2025
| token_ids_device = _async_h2d_tensor_copy(token_ids, self.device) | ||
| positions_device = _async_h2d_tensor_copy(positions, self.device) | ||
| positions_device = input_mrope_positions if self.uses_mrope \ | ||
| else _async_h2d_tensor_copy(positions, self.device) |
Collaborator
There was a problem hiding this comment.
same suggestion for input_mrope_positions , let's keep the original logic that firstly done on 'cpu' and use _async_h2d_tensor_copy to convert
Signed-off-by: attafosu <thomas.atta-fosu@intel.com>
xuechendi
reviewed
Aug 21, 2025
| mrope_position_tensor = torch.full(out_shape, | ||
| padding_gen, | ||
| dtype=torch.int32, | ||
| device='hpu') |
Collaborator
There was a problem hiding this comment.
Is this right, I assume we will init as cpu tensor firstly and use async_h2d function to convert
xuechendi
reviewed
Aug 21, 2025
|
|
||
| def _async_h2d_tensor(data, dtype, device='hpu'): | ||
| if isinstance(data, torch.Tensor): | ||
| return data.to(device=device, dtype=dtype, non_blocking=True) |
Contributor
Author
There was a problem hiding this comment.
unnecessary, and removed.
Signed-off-by: attafosu <thomas.atta-fosu@intel.com>
xuechendi
approved these changes
Aug 21, 2025
mswiniarsk
pushed a commit
that referenced
this pull request
Aug 25, 2025
- Enables v1 multmodal support - Enables qwen2.5-vl: Support for MRope --------- Signed-off-by: attafosu <thomas.atta-fosu@intel.com> Signed-off-by: Marcin Swiniarski <marcin.swiniarski@intel.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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.