Skip to content

Conversation

dignfei
Copy link

@dignfei dignfei commented Apr 24, 2025

When the chat_template is not set in the YAML configuration file, the program crashes unexpectedly.

@lewtun
Copy link
Member

lewtun commented Apr 25, 2025

Hello @dignfei can you share which task you're training with (SFT or GRPO) and which model?

@dignfei
Copy link
Author

dignfei commented May 7, 2025

Hello @dignfei can you share which task you're training with (SFT or GRPO) and which model?

This has nothing to do with the model; as long as the chat_template is not set in the YAML file, an error will occur. qwen2.5-14b ,grpo

@lewtun
Copy link
Member

lewtun commented May 7, 2025

I was not able to replicate an error with the following:

# launch server on GPU 0
CUDA_VISIBLE_DEVICES=0 trl vllm-serve --model Qwen/Qwen2.5-0.5B

# run training on GPU 1
CUDA_VISIBLE_DEVICES=1 TRANSFORMERS_VERBOSITY=info ACCELERATE_LOG_LEVEL=info     accelerate launch --config_file recipes/accelerate_configs/zero2.yaml --num_processes 1     src/open_r1/grpo.py --config recipes/DeepSeek-R1-Distill-Qwen-1.5B/grpo/config_demo.yaml --model_name_or_path Qwen/Qwen2.5-0.5B --num_completions_to_print 0 --wandb_log_unique_prompts true

The config is unchanged except for commenting out the chat template.

Can you share your config and arguments to run the server / train?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants