celsowm/

valdoria-dpo-qwen35-dataset

Dataset de preferências conversacional para uso direto com trl.DPOTrainer.

train.jsonl: 1.889 pares validation.jsonl: 236 pares test.jsonl: 237 pares

Cada linha contém: { "id": "...", "prompt": [{"role": "system", "content": "..."}, {"role": "user", "content": "..."}], "chosen": [{"role": "assistant", "content": "..."}], "rejected": [{"role": "assistant", "content": "..."}], "metadata": {"task_type": "..."… See the full description on the dataset page: https://huggingface.co/datasets/celsowm/valdoria-dpo-qwen35-dataset.