elichen-skymizer/llm-ground-truth-general

Qwen3-4B指令数据集，包含多个版本，每个版本有不同的训练配置。数据集中的每个样本包含问题、来源、类别、输入ID列表、输入令牌长度、生成的文本、生成的令牌长度、预填充令牌数、种子和标签。数据集适用于文本生成任务，分为多个训练集，每个训练集包含200个样本（除vllm-trial版本为198个样本）。

Qwen3-4B Instruct dataset with multiple versions, each with different training configurations. Each sample in the dataset contains a question, source, category, list of input IDs, length of input tokens, generated text, length of generated tokens, number of pre-filled tokens, seed, and labels. The dataset is suitable for text generation tasks and is divided into several training sets, each containing 200 samples (except for the vllm-trial versions which contain 198 samples).