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

[dignfei]

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

[lewtun]

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

[dignfei]

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]

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?