CodeRM LLM Judge Trajectories (Multi-Model)




	
	Dataset Description

This dataset contains selection trajectories from multiple model solutions evaluated by an LLM judge.

	Available Splits




	

Split Trajectories

qwen2_5_32b 1,048

Total: 1,048 trajectories across 1 models

from datasets import load_dataset

ds = load_dataset("t2ance/bilevel-grpo-4b-step110-bestofn-qwen2_5_32b")

ds =… See the full description on the dataset page: https://huggingface.co/datasets/t2ance/bilevel-grpo-4b-step110-bestofn-qwen2_5_32b.