nvidia/Nemotron-RL-agent-calendar_scheduling

Calendar-Scheduling-Dataset是一个多轮对话数据集，能够理解自然语言调度约束、跨多消息遵循指令、推断调度冲突并同时满足多个约束。每个事件都有关于持续时间（例如45分钟）和时间（例如应在下午3点后安排）的约束。用户以自然对话格式随机顺序提及事件和相关约束，助手需要以满足对话中所有约束的方式安排/重新排序事件。数据集生成过程包括使用种子角色生成事件名称、为每个事件生成持续时间和约束、使用LLM生成合成用户消息等步骤。数据集可用于模型安排用户日历中的多个事件，并作为NVIDIA NeMo Gym的一部分发布。

The Calendar-Scheduling-Dataset is a multi-turn conversation dataset that can understand natural language scheduling constraints, follow instructions across multiple messages, infer scheduling conflicts and satisfy multiple constraints simultaneously. Each event has constraints around duration (e.g. 45 min) and timing (e.g. should be scheduled after 3pm). The user mentions the events and associated constraints in a random order in a natural conversational format. The assistant is expected to schedule/reorder events in a way that satisfies all the constraints mentioned in the conversation. The dataset is generated using a seed persona to generate event names, sampling durations and constraints for each event, and prompting an LLM to generate synthetic user messages. This dataset can be used by a model to schedule multiple events in a users calendar and is released as part of NVIDIA NeMo Gym.