Safetensors
GGUF
llama
Inference Endpoints
conversational

Issue with Padding

#1
by akashD22 - opened

Hii, Wanted to ask if you got any padding issue while fine-tuning with GRPO. And if yes, how did you tackle the same.

Also did u use a custom GRPO code or the HF library?

Hi! I didn't experience any issues with padding, as I set it directly in tokenizer_config.json and the model's config.json to 128004. I use GRPOTrainer from the Trl library.

Sign up or log in to comment