LlamaIndex+本地部署InternLM实践RAG的时候，结果不理想，想知道怎么优化呢？

我按照文档《LlamaIndex+本地部署InternLM实践》中的示例代码，使用一个pdf文件作为知识库，模型我有尝试替换成qwen7b和chatglm6b的，embed模型也换成了bge-large-zh，但是基于知识库的问答，结果比较差，想知道问题出在哪里呢，怎么优化呢？

Start with the example code from the “LlamaIndex+本地部署InternLM实践” documentation and reproduce the PDF-based knowledge-base question answering described in the issue. Compare the reported behavior using the original model, qwen7b, chatglm6b, and bge-large-zh, recording the questions and results. Done would require identifying the cause of the poor answers and documenting a validated optimization.

由索引模型根据 Issue 内容生成。