Merge pull request #7 from jihe520/newdev

feat: local code interpreter

Author: jihe520

README.md

@@ -41,7 +41,7 @@
 - [ ] 更多测试案例
 - [ ] docker 部署
 - [ ] 引入用户的交互（选择模型，重写等等）
-- [ ] codeinterpreter 接入云端 如 e2b 等供应商..
+- [x] codeinterpreter 接入云端 如 e2b 等供应商..
 - [ ] 多语言: R 语言, matlab
 - [ ] 绘图 napki,draw.io
 
@@ -64,16 +64,14 @@
 
 ```bash
 ENV=dev
-#兼容 OpenAI 格式都行，具体看官方文档
+# 兼容 OpenAI 格式都行，具体看官方文档
 DEEPSEEK_API_KEY=
 DEEPSEEK_MODEL=
 DEEPSEEK_BASE_URL=
 # 模型最大问答次数
 MAX_CHAT_TURNS=60
 # 思考反思次数
 MAX_RETRIES=5
-# https://e2b.dev/ 
-E2B_API_KEY=
 
 LOG_LEVEL=DEBUG
 DEBUG=true

backend/.env.dev.example

@@ -7,8 +7,9 @@ DEEPSEEK_BASE_URL=
 MAX_CHAT_TURNS=60
 # 思考反思次数
 MAX_RETRIES=5
-# https://e2b.dev/ 
-E2B_API_KEY=
+
+# 不需要填，默认调用本地 Python
+# E2B_API_KEY=
 
 LOG_LEVEL=DEBUG
 DEBUG=true

backend/app/config/setting.py

@@ -1,7 +1,7 @@
 from pydantic import AnyUrl, BeforeValidator, computed_field, field_validator, Field
 from pydantic_settings import BaseSettings, SettingsConfigDict
 import os
-from typing import Annotated
+from typing import Annotated, Optional
 
 
 def parse_cors(value: str) -> list[str]:
@@ -22,14 +22,18 @@ class Settings(BaseSettings):
     DEEPSEEK_BASE_URL: str
     MAX_CHAT_TURNS: int
     MAX_RETRIES: int
-    E2B_API_KEY: str
+    E2B_API_KEY: Optional[str] = None
     LOG_LEVEL: str
     DEBUG: bool
     REDIS_URL: str
     REDIS_MAX_CONNECTIONS: int
     CORS_ALLOW_ORIGINS: Annotated[list[str] | str, BeforeValidator(parse_cors)]
 
-    model_config = SettingsConfigDict(env_file=".env.dev", env_file_encoding="utf-8")
+    model_config = SettingsConfigDict(
+        env_file=".env.dev",
+        env_file_encoding="utf-8",
+        extra="allow",
+    )
 
     def get_deepseek_config(self) -> dict:
         return {

backend/app/core/agents.py

@@ -13,10 +13,10 @@
 from app.utils.enums import CompTemplate, FormatOutPut
 from app.utils.log_util import logger
 from app.config.setting import settings
-from app.tools.code_interpreter import E2BCodeInterpreter
 from app.utils.common_utils import get_current_files
 from app.utils.redis_manager import redis_manager
 from app.schemas.response import SystemMessage
+from app.tools.base_interpreter import BaseCodeInterpreter
 
 
 class Agent:
@@ -93,7 +93,7 @@ def __init__(
         work_dir: str,  # 工作目录
         max_chat_turns: int = settings.MAX_CHAT_TURNS,  # 最大聊天次数
         max_retries: int = settings.MAX_RETRIES,  # 最大反思次数
-        code_interpreter: E2BCodeInterpreter = None,
+        code_interpreter: BaseCodeInterpreter = None,
     ) -> None:
         super().__init__(task_id, model, max_chat_turns)
         self.work_dir = work_dir

backend/app/core/functions.py

@@ -21,6 +21,8 @@
     },
 ]
 
+# have installed: numpy scipy pandas matplotlib seaborn scikit-learn xgboost
+
 # TODO: pip install python
 
 # TODO: read files

backend/app/core/prompts.py

@@ -1,5 +1,7 @@
 from app.utils.enums import FormatOutPut
 
+# TODO: 设计成一个类？
+
 MODELER_PROMPT = """
 role：你是一名数学建模经验丰富的建模手，负责建模部分。
 task：你需要根据用户要求和数据建立数学模型求解问题。
@@ -9,6 +11,8 @@
 **不需要建立复杂的模型,简单规划需要步骤**
 """
 
+# TODO : 对于特大 csv 读取
+
 CODER_PROMPT = """You are an AI code interpreter.
 Your goal is to help users do a variety of jobs by executing Python code.
 
@@ -72,12 +76,12 @@ def get_writer_prompt(
     return f"""
         role：你是一名数学建模经验丰富的写作手，负责写作部分。
         task: 根据问题和如下的模板写出解答,
-        skill：熟练掌握{format_output}排版,
-        output：你需要按照要求的格式排版,只输出{format_output}排版的内容
+        skill：熟练掌握{format_output}排版,如图片、**公式**、表格、列表等
+        output：你需要按照要求的格式排版,只输出正确的{format_output}排版的内容
         
         1. 当你输入图像引用时候，使用![image_name](image_name.png)
         2. 你不需要输出markdown的这个```markdown格式，只需要输出markdown的内容，
-        3. Latex公式使用$$ $$包裹 
+        3. LaTex: 行内公式（Inline Formula） 和 块级公式（Block Formula
         4. 严格按照参考用户输入的格式模板以及**正确的编号顺序**
         5. 不需要询问用户 
         6. 当提到图片时，请使用提供的图片列表中的文件名

backend/app/core/workflow.py

@@ -7,11 +7,12 @@
 from app.utils.common_utils import create_work_dir, simple_chat, get_config_template
 from app.models.user_output import UserOutput
 from app.config.setting import settings
+from app.tools.interpreter_factory import create_interpreter
 from app.core.llm import DeepSeekModel
-from app.tools.code_interpreter import E2BCodeInterpreter
 import json
 from app.utils.redis_manager import redis_manager
 from app.utils.notebook_serializer import NotebookSerializer
+from app.tools.base_interpreter import BaseCodeInterpreter
 
 
 class WorkFlow:
@@ -58,9 +59,10 @@ async def execute(self, problem: Problem):
             SystemMessage(content="正在创建代码沙盒环境"),
         )
 
-        e2b_code_interpreter = await E2BCodeInterpreter.create(
-            workd_dir=self.work_dir,
+        code_interpreter = await create_interpreter(
+            kind="local",
             task_id=self.task_id,
+            work_dir=self.work_dir,
             notebook_serializer=notebook_serializer,
             timeout=3000,
         )
@@ -81,7 +83,7 @@ async def execute(self, problem: Problem):
             work_dir=self.work_dir,
             max_chat_turns=settings.MAX_CHAT_TURNS,
             max_retries=settings.MAX_RETRIES,
-            code_interpreter=e2b_code_interpreter,
+            code_interpreter=code_interpreter,
         )
 
         ################################################ solution steps
@@ -106,7 +108,7 @@ async def execute(self, problem: Problem):
 
             # TODO: 是否可以不需要coder_response
             writer_prompt = self.get_writer_prompt(
-                key, coder_response, e2b_code_interpreter, config_template
+                key, coder_response, code_interpreter, config_template
             )
 
             await redis_manager.publish_message(
@@ -125,7 +127,7 @@ async def execute(self, problem: Problem):
             ## TODO: 图片引用错误
             writer_response = await writer_agent.run(
                 writer_prompt,
-                available_images=await e2b_code_interpreter.get_created_images(key),
+                available_images=await code_interpreter.get_created_images(key),
                 sub_title=key,
             )
 
@@ -138,7 +140,7 @@ async def execute(self, problem: Problem):
 
         # 关闭沙盒
 
-        await e2b_code_interpreter.shutdown_sandbox()
+        await code_interpreter.cleanup()
         logger.info(user_output.get_res())
 
         ################################################ write steps
@@ -223,7 +225,7 @@ def get_writer_prompt(
         self,
         key: str,
         coder_response: str,
-        code_interpreter: E2BCodeInterpreter,
+        code_interpreter: BaseCodeInterpreter,
         config_template: dict,
     ) -> str:
         """根据不同的key生成对应的writer_prompt

backend/app/routers/modeling.py

@@ -31,7 +31,7 @@ async def config():
     }
 
 
-@router.post("/modeling/")
+@router.post("/modeling")
 async def modeling(
     background_tasks: BackgroundTasks,
     ques_all: str = Form(...),  # 从表单获取

backend/app/tools/base_interpreter.py

@@ -0,0 +1,87 @@
+# base_interpreter.py
+import abc
+import re
+from app.utils.notebook_serializer import NotebookSerializer
+from app.utils.redis_manager import redis_manager
+from app.utils.common_utils import get_current_files
+from app.utils.log_util import logger
+from app.schemas.response import (
+    CoderMessage,
+    OutputItem,
+    AgentType,
+)
+
+
+class BaseCodeInterpreter(abc.ABC):
+    def __init__(
+        self,
+        task_id: str,
+        work_dir: str,
+        notebook_serializer: NotebookSerializer,
+    ):
+        self.task_id = task_id
+        self.work_dir = work_dir
+        self.notebook_serializer = notebook_serializer
+        self.section_output: dict[str, dict[str, list[str]]] = {}
+        self.created_images: list[str] = []
+
+    @abc.abstractmethod
+    async def initialize(self):
+        """初始化解释器，必要时上传文件、启动内核等"""
+        ...
+
+    @abc.abstractmethod
+    async def _pre_execute_code(self):
+        """执行初始化代码"""
+        ...
+
+    @abc.abstractmethod
+    async def execute_code(self, code: str) -> tuple[str, bool, str]:
+        """执行一段代码，返回 (输出文本, 是否出错, 错误信息)"""
+        ...
+
+    @abc.abstractmethod
+    async def cleanup(self):
+        """清理资源，比如关闭沙箱或内核"""
+        ...
+
+    async def _push_to_websocket(self, content_to_display: list[OutputItem] | None):
+        logger.info("执行结果已推送到WebSocket")
+
+        agent_msg = CoderMessage(
+            agent_type=AgentType.CODER,
+            code_results=content_to_display,
+            files=get_current_files(self.work_dir, "all"),
+        )
+        logger.debug(f"发送消息: {agent_msg.model_dump_json()}")
+        await redis_manager.publish_message(
+            self.task_id,
+            agent_msg,
+        )
+
+    def add_section(self, section_name: str) -> None:
+        """确保添加的section结构正确"""
+
+        if section_name not in self.section_output:
+            self.section_output[section_name] = {"content": [], "images": []}
+
+    def add_content(self, section: str, text: str) -> None:
+        """向指定section添加文本内容"""
+        self.add_section(section)
+        self.section_output[section]["content"].append(text)
+
+    def get_code_output(self, section: str) -> str:
+        """获取指定section的代码输出"""
+        return "\n".join(self.section_output[section]["content"])
+
+    def delete_color_control_char(self, string):
+        ansi_escape = re.compile(r"(\x9B|\x1B\[)[0-?]*[ -\/]*[@-~]")
+        return ansi_escape.sub("", string)
+
+    def _truncate_text(self, text: str, max_length: int = 1000) -> str:
+        """截断文本，保留开头和结尾的重要信息"""
+        if len(text) <= max_length:
+            return text
+
+        half_length = max_length // 2
+        return text[:half_length] + "\n... (内容已截断) ...\n" + text[-half_length:]