Flow-matching model fine-tuning problem

[houzeyu2683]

Flow-matching model fine-tuning question

Thank you for this great research.

I checked the README.md, and it states that flow-matching model training is supported.
I also read the run.sh script in the example folder.

I found the phrase "We only support LLM training for now" in the LLM training section.
Additionally, I noticed the line for model in llm flow hifigan; do.

Is this the correct way to modify run.sh if I want to fine-tune the flow-matching model?
(Or should I use for model in flow; do only?)

Thanks.

[zw76859420]

需要自己配置llm 与 flow 的训练策略，目前二者均支持训练