Support for fine-tuning more multimodal embedding models (beyond GME)

[groklab]

Describe the feature
Hi ms-swift Team - could you please add support for fine-tuning the other multimodal embedding models? Not just GME-Qwen2-VL-2B/7B?

Paste any useful information
The other models on https://huggingface.co/spaces/TIGER-Lab/MMEB-Leaderboard and https://huggingface.co/spaces/vidore/vidore-leaderboard

Thank you!

[tastelikefeet]

This is already supported.
Just add:
--task_type embedding
You can use other mm models as embedding models for training.
deployment is on the way.

[tastelikefeet]

https://huggingface.co/spaces/TIGER-Lab/MMEB-Leaderboard
And for these embedding models on the leaderboard, we will choose important ones to support.