瀏覽代碼

Merge pull request #13604 from Acters/dev

feat: Improve Default Code Interpreter Prompt
Tim Jaeryang Baek 5 月之前
父節點
當前提交
e60fa7927d
共有 1 個文件被更改,包括 2 次插入1 次删除
  1. 2 1
      backend/open_webui/config.py

+ 2 - 1
backend/open_webui/config.py

@@ -1718,7 +1718,8 @@ DEFAULT_CODE_INTERPRETER_PROMPT = """
 1. **Code Interpreter**: `<code_interpreter type="code" lang="python"></code_interpreter>`
    - You have access to a Python shell that runs directly in the user's browser, enabling fast execution of code for analysis, calculations, or problem-solving.  Use it in this response.
    - The Python code you write can incorporate a wide array of libraries, handle data manipulation or visualization, perform API calls for web-related tasks, or tackle virtually any computational challenge. Use this flexibility to **think outside the box, craft elegant solutions, and harness Python's full potential**.
-   - To use it, **you must enclose your code within `<code_interpreter type="code" lang="python">` XML tags** and stop right away. If you don't, the code won't execute. Do NOT use triple backticks.
+   - To use it, **you must enclose your code within `<code_interpreter type="code" lang="python">` XML tags** and stop right away. If you don't, the code won't execute. 
+   - When writing code in the code_interpreter XML tag, Do NOT use the triple backticks code block for markdown formatting, example: ```py # python code ``` will cause an error because it is markdown formatting, it is not python code.
    - When coding, **always aim to print meaningful outputs** (e.g., results, tables, summaries, or visuals) to better interpret and verify the findings. Avoid relying on implicit outputs; prioritize explicit and clear print statements so the results are effectively communicated to the user.  
    - After obtaining the printed output, **always provide a concise analysis, interpretation, or next steps to help the user understand the findings or refine the outcome further.**  
    - If the results are unclear, unexpected, or require validation, refine the code and execute it again as needed. Always aim to deliver meaningful insights from the results, iterating if necessary.