Skip to content

Conversation

mohiso22
Copy link
Contributor

@mohiso22 mohiso22 commented Sep 8, 2025

No description provided.

@mohiso22 mohiso22 marked this pull request as ready for review September 8, 2025 16:39
**kwargs,
) -> Tuple[torch.Tensor, Optional[Tuple[torch.Tensor, torch.Tensor]]]:
if not self.config.norm_after:
if self._activation_checkpoint_fn is not None:
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Remove all the training related code throughout this file.

Copy link
Contributor

@quic-amitraj quic-amitraj left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Remove all the training related code throughout modelling file.

@mohiso22 mohiso22 force-pushed the molmo branch 2 times, most recently from 4156fd1 to 2c9e1b4 Compare September 26, 2025 08:40
Mohit Soni added 2 commits October 6, 2025 06:25
Signed-off-by: Mohit Soni <mohisoni@qti.qualcomm.com>
Signed-off-by: Mohit Soni <mohisoni@qti.qualcomm.com>

img_size = 588
img_tile = 576
num_images = 5

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Could you make num_images a parameter instead of hardcoding it? The value of num_images here is essentially max_crops (which is one of the image processor kwargs) plus 1. That way, in VLLM, we can apply the same preprocessing logic as we do for llama4

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