by NJUxlj · Agent Tool · ★ 69
Last updated: · Indexed by AgentSkillsHub · Auto-synced every 8h
基于Qwen2.5+LoRA微调+RLHF+RAG的旅游路径规划智能体 注： 如果各位运行项目出现问题无法解决的话，可以直接给我提issue，我会抽空fix的。 最好附上报错截图。 近期修复： 2026-04-15 代码全面审查与修复：共修复26个问题 训练器修复：ppo/dpo/grpo/sft初始化、参数误用、语法错误、混合精度 RAG系统：correctiverag实现、rag()方法、导入路径、数据库初始化逻辑 安全修复：移除toolexecutor中的eval安全风险 代码质量：环境变量配置、循环导入、代码重复、梯度检查点 导入路径：修复main.py、rag.py、ragwebdemo.py等处的所有导入错误 项目的组成部分 本项目希望借助轻量级大模型帮助用户在本地更好地规划旅行路径，提高旅行体验。本项目是基于Qwen2.5+SFT微调+RLHF+RAG的旅游路径规划智能体。 项目的整体框架如下： RAG系统 辅助工具： 工具调用系统 (Google Search + Weather API + Hotel Booking API + Plane Ticket API + Shortest Path API) 自定义的Prompt模板, 继承了 query+context+文档库匹配段落+工具列表+工具格式。它可以让大模型返回用户query命中的工具的函数API字符串. ToolDispatcher: 工具调度器，它可以解析用户query命中的工具的函数API字符串，随后使用ToolExecutor对象调用对应的工具函数。 ToolExecutor: 实际调用各种工具API的执行器。 ChatPDF 传统RAG： 基于传统文档匹配+BM25实现的RAG 完全基于Langchain实现的RAG MemWalker [一种RAG方法，可以自己去搜paper]: 记忆树构建过程 将长文本分割成适合LLM上下文窗口的小块 使用LLM为每个段落生成摘要节点 (summary node / leaf node) 递归地将这些摘要进一步汇总形成高层次摘要节点 (parent node) 最终构建完整的树形记忆结构 交互式导航机制 从树的根节点开始导航 LLM检查不同文本部分，确定与查询相关的路径, 最后LLM会通过相似度比较，一直走到某个叶节点。 最后通过自我纠错能力，判断该叶节点是不是需要的答案，如果不是，则回溯到父节点，继续搜寻下一个叶节点。 在遍历过程中维护工作记忆 能够从早期导航步骤中的错误中恢复, 通过让模型在叶节点输出 action=-1 / -2 / 0 这样的标识，来决定是否回溯
| Stars | 69 |
| Forks | 6 |
| Language | Python |
| Category | Agent Tool |
| Quality Score | 35.75/100 |
| Open Issues | 5 |
| Last Updated | 2026-04-16 |
| Created | 2024-04-19 |
| Platforms | python |
| Est. Tokens | ~9840k |
These tools work well together with Travel-Agent-based-on-Qwen2-RLHF for enhanced workflows:
Looking for a Travel-Agent-based-on-Qwen2-RLHF alternative? If you're comparing Travel-Agent-based-on-Qwen2-RLHF with other agent tool tools, these 6 projects are the closest alternatives on Agent Skills Hub — ranked by topic overlap, star count, and community traction.
Zero-friction LLM fine-tuning skill for Claude Code, Gemini CLI & any ACP agent. Unsloth on NVIDIA · TRL+MPS/M
A framework for agentic tool use training with reinforcement learning
A systematic AI Agent development tutorial covering LLM agents, RAG, tool use, memory systems, multi-agent sys
An introduction to the world of AI Agents
Web app for teams of 20+ members. In-built connections to major LLMs via API. Share chats, prompts, and agents
A ReAct-Based Highly Robust Autonomous Agent Framework.
Explore other popular agent tool tools:
Travel-Agent-based-on-Qwen2-RLHF is A travel agent based on Qwen2.5, fine-tuned by SFT + DPO/PPO/GRPO using traveling question-answer dataset, a mindmap can be output using the response. A RAG system is build upon the tuned qwen2, using. It is categorized as a Agent Tool with 69 GitHub stars.
Travel-Agent-based-on-Qwen2-RLHF is primarily written in Python. It covers topics such as agent, dpo, grpo.
You can find installation instructions and usage details in the Travel-Agent-based-on-Qwen2-RLHF GitHub repository at github.com/NJUxlj/Travel-Agent-based-on-Qwen2-RLHF. The project has 69 stars and 6 forks, indicating an active community.
The top alternatives to Travel-Agent-based-on-Qwen2-RLHF on Agent Skills Hub include unsloth-buddy, ToolBrain, agent_learning. Each offers a different approach to the same problem space — compare them side-by-side by stars, quality score, and community activity.