My stack:
- llm: gemma2-27B-instruct-fp16
- embed: Alibaba-NLP/gte-Qwen2-1.5B-instruct
- agent: ReActAgent
- vector store: chroma
- text splitter: SentenceSplitter(chunk_size=1024, chunk_overlap=150)
When I do common tasks with ReActAgent, it performs well. When it comes to RAG, the results get significantly worse. Are there ways to improve the result? Currently, the Query Engine Tool returns the wrong result from the document to the user's request