Moonlight556/qwen3.5-0.8b-target-matched-math-240k

这是一个包含239,467行数学推理轨迹的数据集，针对目标模型Qwen/Qwen3.5-0.8B重新生成。用于在la-draftery中训练DFlash speculative-decoding drafters。数据集中的用户提示来自Nemotron v2数学语料库，助手完成是目标模型自身的输出，确保drafters学习预测目标模型将生成的令牌，以满足speculative decoding的高接受长度要求。所有行状态为成功，生成时未启用思考模式。239,467 rows of math-reasoning trajectories regenerated against Qwen/Qwen3.5-0.8B as the target model. Used to train DFlash speculative-decoding drafters in la-draftery. The user prompts come from the Nemotron v2 math corpus, and the assistant completions are the target models own outputs, ensuring drafters learn to predict exactly the tokens the target would emit for high acceptance length in speculative decoding. All rows have status success, with thinking disabled during generation.