llm-blender

/

PairRM

like 192

Follow

LLM Blender 13

Their model architecture is the same. The main difference lies in the training data and context length.

1. For the training data:

• pair-ranker is the ranker trained on llm-blender/mix-instruct dataset, producing the results reported in the llm-blender paper.
• PairRM is trained on openai/summarize_from_feedback, lmsys/chatbot_arena_conversations, etc. which does not contain mix-instruct data. (see in PairRM README)

2. For the context length
   There is a simple table comparing the this. The main difference is that pair-ranker constrains the source and candidate lengths to be shorter than 128, while PairRM can extend that constraint to 1224 and 412 respectively.

Overall, you can consider PairRM a more powerful version of llm-blender/pair-ranker.

Thank you very much for your patient answer.