OpenLLMAI / OpenRLHF

An Easy-to-use, Scalable and High-performance RLHF Framework (70B+ PPO Full Tuning & Iterative DPO & LoRA & Mixtral)
https://openrlhf.readthedocs.io/
Apache License 2.0
1.72k stars 161 forks source link

How to get score for a single response from a trained RM #260

Closed UbeCc closed 2 months ago

UbeCc commented 2 months ago

Hello!

I'm newly working on evaluation of RM. I ran the pipeline of RM traning and got one. I want to use it directly to get score for a single response, but I did not find appropriate API for it.

So I turned to use it as AutoModelForSequenceClassification, but I got

Some weights of LlamaForSequenceClassification were not initialized from the model checkpoint 

at /workspace/OpenRLHF/ckpt/vallina_arena_7b_llama_1epoch and are newly initialized: ['score.weight']

What can I do?

Sincerely thanks for your help!

UbeCc commented 2 months ago

I found it in the src, thks!