More Configurable options about Performance

[CLY009]

Hi team,

As I delve deeper into optimizing this project, I'm specifically looking for a comprehensive understanding of all user-configurable parameters that influence its behavior and performance.

While I've noted the presence of .env files and settings.py for general configurations, I'm particularly interested in parameters related to core functionalities, such as:

• Retrieval configurations: e.g., top_k (number of documents to retrieve), similarity_threshold.
• Text processing/chunking: e.g., chunk_size, chunk_overlap.
• Model-specific parameters: e.g., temperature, max_new_tokens for LLMs.
• Any other performance-critical knobs that can be adjusted.

Any insights or pointers to specific files/sections would be incredibly helpful!

Thank you for your time and assistance.

[dartpain]

Hi,

top_k of sorts is configurable via Settings -> general -> chunks dropdown.

Also a decent config summary:
https://docs.docsgpt.cloud/Deploying/DocsGPT-Settings

Rest of the config:

1. Retrieval Parameters
   VECTOR_STORE: Selects the vector database backend (faiss, elasticsearch, qdrant, milvus, lancedb).
   RETRIEVERS_ENABLED: List of enabled retriever types (classic_rag, duckduck_search, brave_search).
   top_k / Chunks per query: Determines how many relevant text chunks are retrieved per query (sometimes called “chunks,” “top_k,” or similar).
   token_limit: Maximum tokens to be used in retrieval or chunking.

application/retriever/classic_rag.py

2. Chunking/Text Processing
   chunking_strategy: The algorithm used to split documents (e.g., classic_chunk).
   max_tokens / min_tokens: Control the maximum/minimum size of each chunk in tokens.
   chunk_overlap: Not directly found in the code sample, but commonly used in chunkers.
   duplicate_headers: Whether to duplicate headers in chunks.
   Relevant Files:

application/parser/chunking.py
See Chunker.init() for parameters

3. Agent
   Agents can have their own meta, source, and prompt settings, such as:
   Chunks per query
   Prompt (system instructions)
   Source (knowledge base selection)
   Relevant Files:

Agent basics docs

If you have any suggestion, or ideas about how we can improve settings - dont hesitate to provide suggestions!

[CLY009]

Thank you so much for this incredibly detailed and helpful response! This is exactly the kind of comprehensive overview I was hoping for to understand the various configuration points beyond just .env and settings.py.

I'll dive into the provided documentation and code next. Your offer for suggestions on improving settings is also very encouraging, and I'll definitely keep that in mind as I explore further.

Thanks again for your excellent guidance!

Hi,  你好

top_k of sorts is configurable via Settings -> general -> chunks dropdown.top_k 可以通过 Settings -> general -> chunks 下拉菜单进行配置。

Also a decent config summary:也是一个不错的配置摘要： https://docs.docsgpt.cloud/Deploying/DocsGPT-Settings

Rest of the config:其余配置：

1. Retrieval Parameters  检索参数
   VECTOR_STORE: Selects the vector database backend (faiss, elasticsearch, qdrant, milvus, lancedb).VECTOR_STORE：选择向量数据库后端（faiss、elasticsearch、qdrant、milvus、lancedb）。
   RETRIEVERS_ENABLED: List of enabled retriever types (classic_rag, duckduck_search, brave_search).RETRIEVERS_ENABLED：已启用的检索器类型（classic_rag、duckduck_search brave_search）的列表。
   top_k / Chunks per query: Determines how many relevant text chunks are retrieved per query (sometimes called “chunks,” “top_k,” or similar).top_k / 每个查询的块数：确定每个查询检索多少个相关文本块（有时称为“块”、“top_k”或类似内容）。
   token_limit: Maximum tokens to be used in retrieval or chunking.token_limit：检索或分块中使用的最大令牌数。

application/retriever/classic_rag.py应用程序/检索器/classic_rag.py

2. Chunking/Text Processing  分块/文本处理
   chunking_strategy: The algorithm used to split documents (e.g., classic_chunk).chunking_strategy：用于拆分文档的算法（例如，classic_chunk）。
   max_tokens / min_tokens: Control the maximum/minimum size of each chunk in tokens.max_tokens / min_tokens：控制 tokens 中每个 chunk 的最大/最小大小。
   chunk_overlap: Not directly found in the code sample, but commonly used in chunkers.chunk_overlap：不直接在代码示例中找到，但通常用于 chunker。
   duplicate_headers: Whether to duplicate headers in chunks.duplicate_headers：是否以 chunk 为单位复制 Header。
   Relevant Files:  相关文件：

application/parser/chunking.py应用程序/解析器/chunking.py See Chunker.init() for parameters请参阅 Chunker。init（） 参数

3. Agent  代理
   Agents can have their own meta, source, and prompt settings, such as:代理可以有自己的元、源和提示设置，例如：
   Chunks per query  每个查询的块数
   Prompt (system instructions)提示符（系统说明）
   Source (knowledge base selection)源（知识库选择）
   Relevant Files:  相关文件：

Agent basics docs  代理基础知识文档

If you have any suggestion, or ideas about how we can improve settings - dont hesitate to provide suggestions!如果您有任何建议或关于我们如何改进设置的想法 - 请随时提供建议！