open-thoughts/OpenThoughts-Agent-SFT-10K
收藏资源简介:
OpenThoughts-Agent-SFT-10K是一个开源数据集,旨在为训练代理模型提供高质量数据。它是OpenThoughts-Agent SFT扩展阶梯中的一个10,000示例点(规模包括316、1K、3.16K、10K、31.6K、100K)。数据集包含(任务,代理轨迹)对,用于微调OpenThinkerAgent-8B-SFT-10K和OpenThinkerAgent-32B-SFT-10K模型。任务来自四个主要来源:SWE-Smith、StackExchange-SuperUser、StackExchange-Tezos(通过合成增强以扩展任务多样性)和IssueTasks。代理轨迹由GLM-4.7-AWQ作为教师模型在terminus-2框架中生成,并过滤出至少包含5个模型轮次的轨迹。数据集包含10,000行数据,字段包括对话(多轮代理轨迹)、任务描述、轨迹来源、代理元数据(如模型和提供者)以及滚动簿记信息(如结果和日期)。
OpenThoughts-Agent-SFT-10K is an open-source dataset curated for training agents. It is the 10,000-example point of the OpenThoughts-Agent SFT scaling ladder (sizes 316 / 1K / 3.16K / 10K / 31.6K / 100K). It contains (task, agent-trajectory) pairs used to fine-tune OpenThinkerAgent-8B-SFT-10K and OpenThinkerAgent-32B-SFT-10K models. Tasks are drawn from the Top-4 sources: SWE-Smith, StackExchange-SuperUser, StackExchange-Tezos (synthetically augmented to expand task diversity), and IssueTasks. Agentic trajectories are generated by GLM-4.7-AWQ acting as the teacher in the terminus-2 harness, then filtered to traces with at least 5 model turns. The dataset consists of 10,000 rows with fields including conversations (multi-turn agent trajectory), task description, trace source, agent metadata (e.g., model and provider), and rollout bookkeeping (e.g., result and date).




