open-thoughts/OpenThoughts-Agent-SFT-1K
收藏资源简介:
OpenThoughts-Agent-SFT-1K 是一个开源数据集,旨在为训练智能体提供高质量数据。它是OpenThoughts-Agent SFT扩展阶梯(包括316、1K、3.16K、10K、31.6K、100K等规模)中的1,000个示例点。该数据集包含(任务,代理轨迹)对,用于微调OpenThinkerAgent-8B-SFT-1K和OpenThinkerAgent-32B-SFT-1K模型。任务来源于四个主要任务源:SWE-Smith、StackExchange-SuperUser、StackExchange-Tezos(通过合成增强以扩展任务多样性)以及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1,000行数据,字段包括对话轨迹、任务描述、任务来源、元数据(代理、模型、模型提供者)以及滚动记录信息(结果、事件、运行ID、试验名称、日期)。
OpenThoughts-Agent-SFT-1K is an open-source dataset curated for training agents. It is the 1,000-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-1K and OpenThinkerAgent-32B-SFT-1K models. Tasks are drawn from the Top-4 task 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 1,000 rows, with fields including conversations (multi-turn agent trajectory), task description, trace source, metadata (agent, model, model_provider), and rollout bookkeeping (result, episode, run_id, trial_name, date).




