search

OpenBMB / MiniCPM-V

MiniCPM-Llama3-V 2.5: A GPT-4V Level Multimodal LLM on Your Phone
Apache License 2.0
7.98k stars 558 forks source link

Finetuning Question #193

Closed babla9 closed 1 month ago

babla9 commented 1 month ago

Thanks for your work and the support for peft tuning!

I have the following questions about finetuning:

  1. Does fine-tuning support text-only based samples? (ie conversations without any images)
  2. The finetuning docs say it will work on 2xV100s, curious if that's been tested with DeepSpeed and other deps installed?
  3. Lastly, I wanted to ask if SWIFT-finetuned models can be deployed using MiniCPM provided VLLM code?

Thanks!

qyc-98 commented 1 month ago

1.yes, we support it 2.yes,you can directly run our fine-tuning script with support of deepspeed. And with zero3, the minnum memory is about 18GiB