Trajectories for the BehR-WM Ricardo-H/BehR-WorldModel-Textworld-Llama3.1-8B evaluated with Qwen/Qwen3-8B as the agent.
World model: Ricardo-H/BehR-WorldModel-Textworld-Llama3.1-8B
World model serving: vLLM, tensor parallel size 1, max_model_len=32768, max_num_seqs=32
Agent: Qwen/Qwen3-8B
Agent serving: vLLM, tensor parallel size 1, max_model_len=32768, temperature 0, thinking disabled… See the full description on the dataset page:
https://huggingface.co/datasets/Ricardo-H/tw-behr-llama-3.1-8b-textworld-wm-w2r-qwen3-8b.