laion/Sera-4.5A-Full-T1-v3-10000

该数据集是allenai/Sera-4.5A-Full-T1数据集的子集，包含10,000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messages字段（作为JSON字符串）、instance_id、rollout_patch、func_name、func_path、problem_statement、target_patch、docker_image等字段，并新增了指向父数据集的source字段。每个助手消息携带原生tool_calls数组（OpenAI工具调用格式）和train标志（用于逐消息损失掩码），这些内容未扁平化为shareGPT格式。数据集设计用于直接通过axolotl使用，配置为type: chat_template、chat_template: chatml、message_field_training: train。采样方法为确定性随机，种子为42，按行索引从完整数据集中抽取。

This dataset is a subset of allenai/Sera-4.5A-Full-T1, containing 10,000 rows (full dataset: 72,118 rows). The format is raw JSONL with OpenAI-native messages layout. It preserves the original messages field (as JSON string), instance_id, rollout_patch, func_name, func_path, problem_statement, target_patch, docker_image, and adds a source field pointing back to the parent dataset. Each assistant message carries a native tool_calls array (OpenAI tool-calling format) and a train flag for per-message loss masking — these are not flattened into shareGPT. Intended for direct consumption by axolotl with type: chat_template, chat_template: chatml, message_field_training: train. Sampling is deterministic random with seed=42, row-indexed into the full dataset.