user_in_github issues

Repositories
Issues
Comments

Results 1 issues of


                                            user_in_github

[Qwen2.5 + GRPO] Cannot Run on V100: float16 crashes, float32 fails during validation

Hi, I'm trying to run **GRPO training with VERL** on **NVIDIA V100 GPUs** using the **Qwen2.5-0.5B** model. ### Problem Summary: - When using `float16` (`actor_rollout_ref.rollout.dtype=float16`), training fails almost immediately with...