laion/CoderForge-Preview-v3-1000

laion/CoderForge-Preview-v3-1000是togethercomputer/CoderForge-Preview数据集的一个子集，包含1000行预处理的tokenized数据，专为Qwen3设计。每行数据包含input_ids、attention_mask、labels等多个字段，用于文本生成任务。数据集支持axolotl框架，适用于序列长度较长的场景。数据来源于四个不同的源（R2E_Gym, SWE_Rebench, SWE_Smith, filtered_reward1），并通过确定性采样（seed=42）生成。

laion/CoderForge-Preview-v3-1000 is a row-subset of the pre-tokenized trajectories in togethercomputer/CoderForge-Preview, containing 1,000 rows of native pre-tokenized data for Qwen3. Each row includes columns such as input_ids, attention_mask, labels, and more, designed for text-generation tasks. The dataset supports the axolotl framework and is suitable for scenarios requiring long sequence lengths. The data is sourced from four different slugs (R2E_Gym, SWE_Rebench, SWE_Smith, filtered_reward1) and generated through deterministic sampling (seed=42).