[Feature] will support qwen3.5-mtp training and qwen3.5-eagle3 training?

[aidenchn]

Checklist

• 1. If the issue you raised is not a feature but a question, please raise a discussion at https://github.com/sgl-project/SpecForge/discussions/new/choose Otherwise, it will be closed.
• 2. Please use English, otherwise it will be closed.

Motivation

Frist of all, thanks for your work,it's really fantastic.
And will you have a plan for supporting qwen3.5-mtp(actually eagle1) training and qwen3.5-eagle3? I have tested qwen3.5-mtp and it has a good performance, so I think maybe qwen3.5-eagle3 would be better.

Sincerely waiting for your replay.

Related resources

No response

[Swipe4057]

#496

[jiapingW]

We also support qwen3.5-Dflash training in this PR. We are looking for a suitable way to integrate MTP for fine-tuning. If you have any good ideas, please feel free to share them.

[gme-hong]

@jiapingW 为什么我在训练Qwen3.5-Eagle3的时候，GPU的利用率总是卡在30%上不去，而且使用离线训练的时候，我发现即使使用is-vlm依旧调用的是extend，走的是文本路线，pixel信息被丢弃了，请问这正确吗？