https://www.deeplearning.ai/short-courses/building-agentic-rag-with-llamaindex/ > 主要就是教 Function Calling + RAG tools，也就是 [[Data Agents]] ## Router Query Engine ![[Pasted image 20240513202450.png]] * 安裝 nest_asyncio 是為了與 notebook 相容支援非同步 * 在 [[Multi-Vector Retriever]] 內有介紹 Summary Index * 兩個工具的描述是 * "Useful for summarization questions related to MetaGPT" * "Useful for retrieving specific context from the MetaGPT paper." ## Tool Calling 改用 function calling 做 ![[Pasted image 20240513222713.png]] ![[Pasted image 20240513224200.png]] ## Building an Agent Reasoning Loop ![[Pasted image 20240513234831.png]] > demo 似乎讓 function calling 就是 single function call 了，而不是 parallel function calling > 查了 allow_parallel_tool_calls 參數，預設是 False，哎 ![[Pasted image 20240513235237.png]] 在教 llamaindex low-level agent API 了 ## Build a Multi-Document Agent ![[Pasted image 20240514000230.png]] 假設定三份文件: 把每個文件變成一個 2 個 tools，三份文件共 6 個 tools ![[Pasted image 20240514000826.png]] ![[Pasted image 20240514001929.png]] ### 來做 11篇的情境，這時 tools 太多了.... 解法: 針對 tools 選擇做 RAG: 根據問題，找出最相似的三個工具 ![[Pasted image 20240514002135.png]] * 範例問題: 有摘要、也有比較 * "Tell me about the evaluation dataset used in MetaGPT and compare it against SWE-Bench" * "Compare and contrast the LoRA papers (LongLoRA, LoftQ). Analyze the approach in each paper first. " ## Conclusion ![[Pasted image 20240514002834.png]]