Skip to content

Added changes to enable dynamo export and compile of VLMs using Qwen2.5 and 3 VLs as example. - #1322

Draft
quic-dhirajku wants to merge 4 commits into
quic:quic-enable-weightfree-for-vlmfrom
quic-dhirajku:vlm_dynamo
Draft

quic-dhirajku wants to merge 4 commits into
quic:quic-enable-weightfree-for-vlmfrom
quic-dhirajku:vlm_dynamo

Conversation

@quic-dhirajku

Copy link
Copy Markdown
Contributor

These changes would work for non disagg setups(not verified for disagg setup).

….5 and 3 VLs as example.

These changes would work for non disagg setups(not verified for disagg setup).

Signed-off-by: Dhiraj Kumar Sah <dhirajku@qti.qualcomm.com>
…nges. This PR now independently works on dynamo export of VLMs.

Edited qwen 2.5 vl modeling for a minor bug that caused subfunctions to be different for each layer.

Signed-off-by: Dhiraj Kumar Sah <dhirajku@qti.qualcomm.com>
@quic-amitraj
quic-amitraj changed the base branch from main to quic-enable-weightfree-for-vlm September 16, 2026 08:31
@quic-amitraj

Copy link
Copy Markdown
Contributor

Please check the unit tests @quic-dhirajku

… Qwen3.5MoE(Referred Sharvaris initial code) models.

The ONNX transform to handle subfunctions renaming is a WIP fix, we'll replace it with a torch patch down the line.

Signed-off-by: Dhiraj Kumar Sah <dhirajku@qti.qualcomm.com>
Signed-off-by: Dhiraj Kumar Sah <dhirajku@qti.qualcomm.com>

This branch has not been deployed

No deployments
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.

2 participants