laion/CoderForge-Preview-v3-10000

laion/CoderForge-Preview-v3-10000数据集是togethercomputer/CoderForge-Preview数据集的一个子集，包含10,000行预处理过的tokenized数据，适用于Qwen3模型。每行数据包括input_ids、attention_mask、labels等多个字段，并已应用了-100掩码。数据集主要用于文本生成任务，支持axolotl框架，序列长度可达32768个token。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（种子=42）生成。

laion/CoderForge-Preview-v3-10000 is a row-subset of the pre-tokenized trajectories in togethercomputer/CoderForge-Preview (trajectories-tokenized_qwencoder subset), containing 10,000 rows. The data is pre-tokenized for Qwen3 (tokenizer shared with Qwen2.5-Coder / Qwen3-Coder / Qwen3-8B). Each row includes columns like input_ids, attention_mask, labels (with -100 masks applied), chat_template_applied, trajectory_id, reward, and source. The dataset is sampled deterministically (seed=42) from a concatenation of 4 source slugs (R2E_Gym, SWE_Rebench, SWE_Smith, filtered_reward1). It is designed for text-generation tasks and supports axolotl framework usage with a sequence length of 32768 tokens.