Skip to content

Karen/vlm fix#311

Merged
HenryNdubuaku merged 3 commits intocactus-compute:mainfrom
BruinAI:karen/vlm-fix
Feb 1, 2026
Merged

Karen/vlm fix#311
HenryNdubuaku merged 3 commits intocactus-compute:mainfrom
BruinAI:karen/vlm-fix

Conversation

@kar-m
Copy link
Copy Markdown
Collaborator

@kar-m kar-m commented Feb 1, 2026

No description provided.

kar-m added 2 commits January 31, 2026 18:04
Signed-off-by: Karen Mosoyan <[email protected]>
Copilot AI review requested due to automatic review settings February 1, 2026 02:07
Copy link
Copy Markdown
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

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

Pull request overview

This pull request fixes a bug in entropy calculation for Vision-Language Models (VLM) where logits buffers with multiple sequence dimensions were not being handled correctly.

Changes:

  • Fixed entropy calculation to correctly extract the last token's logits from multi-dimensional logits buffers
  • Applied the fix consistently across both the VLM model (model_lfm2vl.cpp) and the base engine model (engine_model.cpp)

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File Description
cactus/models/model_lfm2vl.cpp Added logic to handle multi-dimensional logits buffers by computing row offset to extract the last sequence position; removed unused blank line; added <numeric> header (unused)
cactus/engine/engine_model.cpp Added identical multi-dimensional logits handling logic for consistency with VLM model

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

@HenryNdubuaku HenryNdubuaku merged commit 6c1d1e4 into cactus-compute:main Feb 1, 2026
0 of 2 checks passed
Goldokpa pushed a commit to Goldokpa/cactus that referenced this pull request Feb 11, 2026
* fixed entropy calculation for vlm

Signed-off-by: Karen Mosoyan <[email protected]>

* reverted run.sh change

Signed-off-by: Karen Mosoyan <[email protected]>

* Update cactus/models/model_lfm2vl.cpp

Co-authored-by: Copilot <[email protected]>

---------

Signed-off-by: Karen Mosoyan <[email protected]>
Co-authored-by: Copilot <[email protected]>
ncylich pushed a commit that referenced this pull request Feb 24, 2026
* fixed entropy calculation for vlm

Signed-off-by: Karen Mosoyan <[email protected]>

* reverted run.sh change

Signed-off-by: Karen Mosoyan <[email protected]>

* Update cactus/models/model_lfm2vl.cpp

Co-authored-by: Copilot <[email protected]>

---------

Signed-off-by: Karen Mosoyan <[email protected]>
Co-authored-by: Copilot <[email protected]>
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