ning423/nemotron-nano-hermes-traces

Nemotron Nano Hermes Agent Reasoning Traces是一个精心策划的数据集，包含用于训练本地AI编排代理的推理轨迹。该数据集专为Nemotron 3 Nano Omni的监督微调（SFT）和强化学习（RL）训练而设计，旨在成为最佳的本地Hermes代理模型。数据集包含28,000个SFT行和28,000个RL提示，格式为ShareGPT（对话列）。目标模型为nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16，训练框架为Unsloth Studio。数据来源包括claude-sonnet、lambda-kimi、lambda-glm51和mimo-generated等AI模型。数据集涵盖多个类别，如代理工具、存储库任务、多工具工作流、浏览器自动化、文件操作、调度、规划与组织、终端与编码、对话等，以及心理学和数学等多个领域。

Nemotron Nano Hermes Agent Reasoning Traces is a curated dataset of reasoning traces for training local AI orchestrator agents. Designed for SFT and RL training of Nemotron 3 Nano Omni to be the best local Hermes Agent model. The dataset includes 28,000 SFT rows and 28,000 RL prompts, formatted in ShareGPT (conversations column). The target model is nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16, and the training framework is Unsloth Studio. Sources include AI models like claude-sonnet, lambda-kimi, lambda-glm51, and mimo-generated. Categories cover a wide range of topics from Agent Tools, Repository Tasks, Multi-Tool Workflows, Browser Automation, File Operations, Scheduling, Planning & Organization, Terminal & Coding, Conversational, to psychology and mathematics.