search

uclaml / SPIN

The official implementation of Self-Play Fine-Tuning (SPIN)
https://uclaml.github.io/SPIN/
Apache License 2.0
1.05k stars 92 forks source link

vllm generation issue. #3

Closed tdolan21 closed 9 months ago

tdolan21 commented 9 months ago

i believe that the argument used in generate_vllm.sh is supposed to be batch_size rather than world_size. When I make that change the script starts working immediately.

yihedeng9 commented 9 months ago

Hello. The arguments for generate_vllm.py and generate_vllm.sh were both world_size. We have unified this to batch_size as in generate.py in our latest commit. Thanks for bringing it up!

yihedeng9 commented 9 months ago

Also, the previous generate_vllm.sh seems to be wrongly calling generate.py. You may want to make sure to be using generate_vllm.py (refer to the latest updates script).