Skip to content

Add Qwen3.5 VLM CI#1814

Open
zhuzilin wants to merge 6 commits intomainfrom
ci
Open

Add Qwen3.5 VLM CI#1814
zhuzilin wants to merge 6 commits intomainfrom
ci

Conversation

@zhuzilin
Copy link
Copy Markdown
Contributor

@zhuzilin zhuzilin commented Apr 7, 2026

Register the Geo3K Qwen3.5-4B VLM smoke test in the permanent PR workflows and document the new VLM coverage.\n\nCo-authored-by: Copilot 223556219+Copilot@users.noreply.github.com

Register the Geo3K Qwen3.5-4B VLM smoke test in the permanent PR workflows and document the new VLM coverage.\n\nCo-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@coding-famer
Copy link
Copy Markdown
Contributor

I think the Qwen3.5 VLM bridge mode requires upgrading the Megatron‑Bridge. The example uses a different branch:

# TODO: Remove this once Megatron & Megatron Bridge are upgraded upstream.
pip install git+https://github.com/coding-famer/Megatron-Bridge-slime.git@qwen35 --no-build-isolation

@zhuzilin
Copy link
Copy Markdown
Contributor Author

zhuzilin commented Apr 7, 2026

I think the Qwen3.5 VLM bridge mode requires upgrading the Megatron‑Bridge. The example uses a different branch:

get~ hmm... I'll try to convert the ckpt to torch dist with mbridge for this ci. the current code should be able to run qwen3.5 at the moment.

@coding-famer
Copy link
Copy Markdown
Contributor

coding-famer commented Apr 7, 2026

I think the Qwen3.5 VLM bridge mode requires upgrading the Megatron‑Bridge. The example uses a different branch:

get~ hmm... I'll try to convert the ckpt to torch dist with mbridge for this ci. the current code should be able to run qwen3.5 at the moment.

VLMs have customized Megatron model classes, which depend on Megatron-Bridge, as well as the weight updating👀

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants