DeepSeek-VL deepseek-vl系列的微调支持 (finetune)

deepseek-vl系列的微调支持 (finetune)

Open Jintao-Huang opened this issue 1 year ago • 6 comments

ms-swift大模型训练框架已经支持了deepseek-vl系列模型的推理和微调～

最佳实践可以查看这里: https://github.com/modelscope/swift/blob/main/docs/source/Multi-Modal/deepseek-vl%E6%9C%80%E4%BD%B3%E5%AE%9E%E8%B7%B5.md

Mar 12 '24 09:03 Jintao-Huang

Thank you for supporting DeepSeek-VL!

Mar 12 '24 14:03 RERV

@Jintao-Huang Can you kindly confirm if swift can be used to finetune visual encoder? If so, how? If not, what's the simplest way to support it?

Mar 13 '24 10:03 soloice

还有就是lora微调后怎么部署使用

Mar 13 '24 13:03 xs818818

@soloice Thank you very much for asking

Mar 13 '24 15:03 SinanAkkoyun

LoRA fine-tuning and merge-LoRA have been supported for both the visual encoder and aligner. Full parameter fine-tuning is also supported. 😊

Mar 13 '24 22:03 Jintao-Huang

I am super grateful for your work, thank you a lot!!! ❤️

Mar 13 '24 22:03 SinanAkkoyun