Skip to content

[Model] Dynamic image size support for LLaVA-NeXT#5279

Closed
DarkLight1337 wants to merge 18 commits intovllm-project:mainfrom
DarkLight1337:mm-image-tokenizer-llava
Closed

[Model] Dynamic image size support for LLaVA-NeXT#5279
DarkLight1337 wants to merge 18 commits intovllm-project:mainfrom
DarkLight1337:mm-image-tokenizer-llava

Conversation

@DarkLight1337
Copy link
Copy Markdown
Member

@DarkLight1337 DarkLight1337 commented Jun 5, 2024

A proof-of-concept showcasing the application of input processor for inserting a dynamic number of image tokens. This is required to fully take advantage of the multi-resolution feature in LLaVA-NeXT.

Related contributions

This PR is essentially #5276 applied on top of #4199. It will leave draft status when both dependences are merged.

Update: Since #4199 has been merged, #5276 now supersedes this PR.

@DarkLight1337
Copy link
Copy Markdown
Member Author

DarkLight1337 commented Jun 11, 2024

Closing, since LLaVA-NeXT has already been added to vLLM. #5276 should be able to take care of this now.

For reference, I'm deferring the deletion of the branch for this PR until #5276 is merged.

@DarkLight1337
Copy link
Copy Markdown
Member Author

#5276 is in a pretty good spot now so I'll delete this branch.

@DarkLight1337 DarkLight1337 deleted the mm-image-tokenizer-llava branch June 27, 2024 09:26
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.

1 participant