R1_Qwen3_8B_dpo

Checkpoint from the RoleVerse project — a multilingual benchmark for social reasoning through same-universe role-playing across 5 languages (EN, ZH, JA, FR, ES).

This checkpoint was produced by the RoleVerse training pipeline (SFT and/or DPO). See the project repo and dataset for details on how it was trained.

• Dataset: https://huggingface.co/datasets/tinyrolls/RoleVerse
• Repo ID: tinyrolls/R1_Qwen3_8B_dpo

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "tinyrolls/R1_Qwen3_8B_dpo"
tok = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype="auto", device_map="auto")