Add vision only mode#537
Merged
gkumbhat merged 13 commits intoJun 24, 2026
Merged
Conversation
Signed-off-by: Gaurav-Kumbhat <Gaurav.Kumbhat@ibm.com>
Signed-off-by: Gaurav-Kumbhat <Gaurav.Kumbhat@ibm.com>
Signed-off-by: Gaurav-Kumbhat <Gaurav.Kumbhat@ibm.com>
Signed-off-by: Gaurav-Kumbhat <Gaurav.Kumbhat@ibm.com>
Signed-off-by: Gaurav-Kumbhat <Gaurav.Kumbhat@ibm.com>
Signed-off-by: Gaurav-Kumbhat <Gaurav.Kumbhat@ibm.com>
Signed-off-by: gkumbhat <Gaurav.Kumbhat@ibm.com>
Signed-off-by: Gaurav-Kumbhat <Gaurav.Kumbhat@ibm.com>
Signed-off-by: gkumbhat <Gaurav.Kumbhat@ibm.com> Signed-off-by: Gaurav-Kumbhat <Gaurav.Kumbhat@ibm.com>
d14e05d to
2d6eee5
Compare
Signed-off-by: Gaurav-Kumbhat <Gaurav.Kumbhat@ibm.com>
Signed-off-by: Gaurav-Kumbhat <Gaurav.Kumbhat@ibm.com>
Signed-off-by: gkumbhat <Gaurav.Kumbhat@ibm.com>
5 tasks
flaviabeo
approved these changes
Jun 23, 2026
flaviabeo
left a comment
Collaborator
There was a problem hiding this comment.
Tested locally with mistralai/Ministral-3-3B-Instruct-2512, all seems to be working.
Image size: (547, 365)
Embeddings shape: torch.Size([1, 834, 3072])
(batch=1, seq_len=834, hidden_dim=3072)
Let me know if there are any other models of interest that would be nice to test with this mode.
flaviabeo
requested changes
Jun 23, 2026
flaviabeo
left a comment
Collaborator
There was a problem hiding this comment.
So, testing with ibm-granite/granite-vision-3.2-2b I am getting an error - I used the almost the same script from Mistral3 so maybe some issue related specific to granite vision?
| if not isinstance(original_image_size, (list, tuple)): | ||
| original_size = original_image_size.tolist() | ||
|
|
||
| original_height, original_width = original_size |
Collaborator
There was a problem hiding this comment.
As I test with granite-vision, I got and error that traces to this line:
File "/home/flaviabeodebayser/fms/foundation-model-stack/fms/models/llava_next.py", line 270, in select_best_resolution
original_height, original_width = original_size
^^^^^^^^^^^^^
UnboundLocalError: cannot access local variable 'original_size' where it is not associated with a value
INFO:torch._dynamo.eval_frame:TorchDynamo attempted to trace the following frames: [
]
INFO:torch._dynamo.utils:TorchDynamo compilation metrics:
Function, Runtimes (s)
Collaborator
Author
There was a problem hiding this comment.
this seems to be an existing bug which gets uncovered in specific scenario .. I have pushed a fix now.
Signed-off-by: gkumbhat <kumbhat.gaurav@gmail.com>
flaviabeo
approved these changes
Jun 24, 2026
flaviabeo
left a comment
Collaborator
There was a problem hiding this comment.
Now granite-vision also works :)
LGTM!
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.
Changes
This PR adds ability to load vision part of the model from a given multimodal model. This allows ability for sendnn-inference to load the model's different parts differently for efficient processing