nebuly
nebuly copied to clipboard
[Chatllama] assert _MODEL_PARALLEL_GROUP is not None, "model parallel group is not initialized"
How to use parallel training in actor. Simple demo is not work. The project is not solid.
Hi @seasidemym, thank you so much for reaching out. Today, we are going to release a more stable version of ChatLLaMA, hopefully you will enjoy it more. Please let us know if you have any other feedback 😄
Any update? I encountered same issue. I simply set environment variables by two command: export MP=1, export WORLD_SIZE=1 . Then start training of actor with "fairscale True" in config.yaml.
Hi @seasidemym @tqjack
This issue should have been fixed in the previous release in which was missing the model parallel initialisation.