laion/Sera-4.5A-Full-T1-v3-3160

该数据集是allenai/Sera-4.5A-Full-T1数据集的子集，包含3,160行数据（完整数据集有72,118行）。数据格式为原始JSONL，采用OpenAI原生消息布局，保留了原始字段如messages、instance_id等，并添加了source字段指向父数据集。每条助手消息包含一个原生的tool_calls数组和一个train布尔标志，用于每条消息的损失掩码。数据集适用于直接通过axolotl使用，配置为type: chat_template和chat_template: chatml。采样方法是确定性随机，种子为42，按行索引到完整数据集中。

This dataset is a subset of allenai/Sera-4.5A-Full-T1, containing 3,160 rows (full dataset: 72,118 rows). The format is raw JSONL, with OpenAI-native messages layout, preserving original fields like messages, instance_id, etc., and adding a source field pointing back to the parent dataset. Each assistant message carries a native tool_calls array and a train boolean flag for per-message loss masking. The dataset is intended for direct consumption by axolotl with type: chat_template and chat_template: chatml configuration. Sampling is deterministic random, seed=42, row-indexed into the full dataset.