I have this strange error with GRPO Trainer

[MohamedAliRashad]

ValueError: The global train batch size (1 x 1) must be evenly divisible by the number of generations per prompt (4). Given the current train batch size, the valid values for the number of generations are: [].

This is the output of trl env

INFO 02-16 21:48:30 __init__.py:190] Automatically detected platform cuda.

Copy-paste the following information when reporting an issue:

- Platform: Linux-5.15.0-122-generic-x86_64-with-glibc2.35
- Python version: 3.11.10
- PyTorch version: 2.5.1
- CUDA device(s): NVIDIA RTX A6000
- Transformers version: 4.48.3
- Accelerate version: 1.3.0
- Accelerate config: not found
- Datasets version: 3.3.0
- HF Hub version: 0.28.1
- TRL version: 0.15.0
- bitsandbytes version: 0.45.2
- DeepSpeed version: not installed
- Diffusers version: 0.32.2
- Liger-Kernel version: not installed
- LLM-Blender version: not installed
- OpenAI version: 1.63.0
- PEFT version: 0.14.0

This error didn't happen with me couple of days ago and i didn't change anything in my code.

[leonardtang]

There was a new version of the GRPOTrainer introduced recently that changed the meaning of per_device_batch_size.

"Instead of defining per_device_batch_size as the number of prompts per device, it now represents the number of generations per device." -- See PR

[MohamedAliRashad]

@leonardtang The problem is when i made per_device_batch_size equal to 4 so it can be divisiable to the number of generations it gave me another shape error

[Superskyyy]

@leonardtang The problem is when i made per_device_batch_size equal to 4 so it can be divisiable to the number of generations it gave me another shape error

Please post the another shape error. Thanks

[wojtess]

I have same problem, I dont have per_device_batch_size in GRPOConfig.init()
TypeError: GRPOConfig.__init__() got an unexpected keyword argument 'per_device_batch_size'
Edit: After reading error message one more time I changed per_device_train_batch_size to same value as num_generations and this fixed problem