[Feature] qwen2 vl support the turbomind engine

[DexterGuo]

Motivation

1、The qwen2vl effect is the sota level in the open source model
2、lmdeploy is an excellent inference framework
3、So it's important to support turbomind

Related resources

No response

Additional context

No response

[lvhan028]

#2720 is working on it.

[ciwang]

Hi @lvhan028, do you have an estimate of when that PR will be merged? Thank you in advance!

[lvhan028]

Not yet. We are trying to refactor the VLM inference in PR #2810
TM engine will not support qwen2-vl until that PR is merged.

[songlinlibit]

@lvhan028 Hello, I noticed that this PR has been merged. Is it confirmed that versions after #2810 already support the use of the tm engine for Qwen2-VL inference?

[comeby]

Is there any plan to merge this feature?

[thisiskofi]

#2720 is still open

[lvhan028]

Sorry about that. We are occupied by other top-priority features.
Hope we have time in Feb to finish it.