TIDE-dllm/distill_llada2_sft

该数据集是一个预处理的SFT（监督微调）混合数据集，用于LLaDA2-teacher蒸馏训练。数据集包含多个来源的数据，如allenai/tulu-3-sft-mixture、HuggingFaceTB/smoltalk等，并且已经预标记化，适用于Qwen3-0.6B-BD3LM学生模型和LLaDA2.0-mini教师模型。数据集的结构包括输入ID、标签、教师输入ID、教师标签等字段，并支持流式加载和本地持久化使用。This dataset is a preprocessed SFT (Supervised Fine-Tuning) mixture used for LLaDA2-teacher distillation training. The dataset includes data from multiple sources such as allenai/tulu-3-sft-mixture, HuggingFaceTB/smoltalk, etc., and is pre-tokenized for both the Qwen3-0.6B-BD3LM student model and the LLaDA2.0-mini teacher model. The dataset structure includes fields like input IDs, labels, teacher input IDs, teacher labels, etc., and supports streaming loading and local persistence usage.