jescy525/archon-sft-v1-orchestration
收藏资源简介:
该数据集是AETHER家族SFT数据集的一部分,属于orchestration组,用于工具使用和函数调用的监督微调,针对ARCHON编排角色。数据格式为JSONL ChatML消息,应用了MinHash去重(阈值0.85)。每条记录包含消息(角色为system、user或assistant)、任务类型(如function_calling、code、reasoning_cot等)、来源数据集ID、语言(实际为100%英语,尽管标签包含法语)和系统来源(archon_default或overridden_from_source)。任务类型分布以tool_use_multi(77%)和math_symbolic(21%)为主,其他包括function_calling(1.3%)和protobuf_grpc(0.3%)。数据来源主要为NousResearch的hermes-3-dataset(98.6%)和hermes-function-calling-v1(1.4%)。数据集由prepare_sft.py管道生成(2026-05-25),规模在10万到100万条之间。
This dataset is part of the AETHER family SFT datasets, belonging to the orchestration group, designed for supervised fine-tuning of tool use and function calling for the ARCHON orchestration role. The format is JSONL ChatML messages with MinHash deduplication applied (threshold 0.85). Each record includes messages (with roles such as system, user, or assistant), task_type (e.g., function_calling, code, reasoning_cot), source dataset ID, language (actually 100% English despite tags including French), and system_source (archon_default or overridden_from_source). The task_type distribution is dominated by tool_use_multi (77%) and math_symbolic (21%), with others including function_calling (1.3%) and protobuf_grpc (0.3%). The data sources are primarily NousResearchs hermes-3-dataset (98.6%) and hermes-function-calling-v1 (1.4%). The dataset was generated by the prepare_sft.py pipeline (2026-05-25) and has a size between 100K and 1M samples.




