OptimalScale / LMFlow

An Extensible Toolkit for Finetuning and Inference of Large Foundation Models. Large Models for All.
https://optimalscale.github.io/LMFlow/
Apache License 2.0
8.23k stars 822 forks source link

Does it support llama3? #785

Open orderer0001 opened 5 months ago

orderer0001 commented 5 months ago

Does it support multi-gpu raining llama3 with lisa ?

research4pan commented 5 months ago

Thanks for your interest in LMFlow! We currently support single-gpu training of LLaMA-3-7b with LISA. We are working on the multi-gpu model-parallelism version, hopefully in 1-2 weeks, please stay tuned. Thanks for your understanding 😄

orderer0001 commented 5 months ago

LLaMA-3-7b

Do you support llama 2-7b? Or llama 3-8B? llama3-7b does not exist.

research4pan commented 5 months ago

Yes we support both models as long as the access is granted from huggingface repo for your huggingface account.

orderer0001 commented 5 months ago

Yes we support both models as long as the access is granted from huggingface repo for your huggingface account.

Which parameter is used to configure authorized account related information?

wheresmyhair commented 5 months ago

Yes we support both models as long as the access is granted from huggingface repo for your huggingface account.

Which parameter is used to configure authorized account related information?

You may first switch to lmflow conda env using

conda activate YOUR_ENV_NAME

then use

huggingface-cli login

and follow the instructions.