Add Code Interpreter demo

api/routes/chat.py

@@ -93,7 +93,7 @@ async def create_chat_completion(request: ChatCompletionRequest, raw_request: Re
             res, function_call = parse_response(content["text"])
             content["text"] = res
 
-        if isinstance(function_call, dict):
+        if isinstance(function_call, dict) and "arguments" in function_call:
             finish_reason = "function_call"
             function_call = FunctionCallResponse(**function_call)
 
@@ -126,6 +126,7 @@ async def chat_completion_stream_generator(
     https://developer.mozilla.org/en-US/docs/Web/API/Server-sent_events/Using_server-sent_events#event_stream_format
     """
     _id = f"chatcmpl-{secrets.token_hex(12)}"
+    use_tool = bool(gen_params["functions"] is not None)
     for i in range(n):
         # First chunk with role
         choice_data = ChatCompletionResponseStreamChoice(
@@ -160,14 +161,14 @@ async def chat_completion_stream_generator(
             function_call = None
             if finish_reason == "function_call" and "chatglm3" in config.MODEL_NAME.lower():
                 try:
-                    function_call = process_response_v3(decoded_unicode, use_tool=True)
+                    function_call = process_response_v3(decoded_unicode, use_tool=use_tool)
                 except:
                     logger.warning("Failed to parse tool call")
 
             elif finish_reason == "function_call" and "qwen" in config.MODEL_NAME.lower():
                 _, function_call = parse_response(decoded_unicode)
 
-            if isinstance(function_call, dict):
+            if isinstance(function_call, dict) and "arguments" in function_call:
                 function_call = FunctionCallResponse(**function_call)
 
             delta = DeltaMessage(

examples/chatglm3/tool_using.py

@@ -65,6 +65,7 @@ def run_conversation(query: str, stream=False, functions=None, max_retry=5):
                     params["messages"].append(
                         {
                             "role": "assistant",
+                            "function_call": function_call,
                             "content": output
                         }
                     )

streamlit-demo/.env

@@ -4,4 +4,6 @@ TOOL_CHAT_API_BASE = "http://192.168.20.59:7891/v1"  # 调用工具模型接口
 EMBEDDING_API_BASE = "http://192.168.0.53:7891/v1"  # 嵌入模型接口地址（可选）
 API_KEY = "xxx"  # 默认不需要配置
 EMBEDDING_NAME = ""  # 使用本地嵌入模型的路径（可选，EMBEDDING_API_BASE 和 EMBEDDING_NAME 两种方式选一种即可）
-SERPAPI_API_KEY = ""  # 搜索功能需要
+SERPAPI_API_KEY = ""  # 搜索功能需要
+IPYKERNEL = "llm"  # python解释器名称
+INTERPRETER_CHAT_API_BASE = "http://192.168.20.59:7891/v1"  # 代码解释器模型接口地址（可选）

streamlit-demo/README.md

@@ -8,3 +8,9 @@ streamlit run streamlit_app.py
 
 **环境变量配置见 [.env](.env)**
 
+## 代码解释器（基于 ChatGLM3 模型）【测试版本】
+
+```shell
+ipython kernel install --name llm --user
+```
+

streamlit-demo/streamlit_app.py

@@ -7,7 +7,7 @@
 
 def main():
     from streamlit_gallery.apps import gallery
-    from streamlit_gallery.components import chat, doc_chat, sql_chat, search_chat, tool_chat
+    from streamlit_gallery.components import chat, doc_chat, sql_chat, search_chat, tool_chat, code_interpreter
 
     page = page_group("p")
 
@@ -32,6 +32,9 @@ def main():
             if os.getenv("TOOL_CHAT_API_BASE", ""):
                 page.item("Tool Chat", tool_chat)
 
+            if os.getenv("INTERPRETER_CHAT_API_BASE", ""):
+                page.item("Code Interpreter", code_interpreter)
+
         with st.expander("🐧 PARAMTERS", False):
             max_tokens = st.slider("MaxTokens", 20, 4096, 1024)
             temperature = st.slider("Temperature", 0.0, 1.0, 0.9)

streamlit-demo/streamlit_gallery/components/__init__.py

@@ -3,3 +3,4 @@
 from .sql_chat.streamlit_app import main as sql_chat
 from .search_chat.streamlit_app import main as search_chat
 from .tool_chat.streamlit_app import main as tool_chat
+from .code_interpreter.streamlit_app import main as code_interpreter

streamlit-demo/streamlit_gallery/components/code_interpreter/streamlit_app.py

@@ -0,0 +1,121 @@
+import os
+
+import openai
+import streamlit as st
+
+from .utils import CodeKernel, extract_code, execute, postprocess_text
+
+
+@st.cache_resource
+def get_kernel():
+    return CodeKernel()
+
+
+SYSTEM_MESSAGE = [
+    {
+        "role": "system",
+        "content": "你是一位智能AI助手，你叫ChatGLM，你连接着一台电脑，但请注意不能联网。在使用Python解决任务时，你可以运行代码并得到结果，如果运行结果有错误，你需要尽可能对代码进行改进。你可以处理用户上传到电脑上的文件，文件默认存储路径是/mnt/data/。"
+    }
+]
+
+
+def chat_once(message_placeholder):
+    params = dict(
+        model="chatglm3",
+        messages=SYSTEM_MESSAGE + st.session_state.messages,
+        stream=True,
+        max_tokens=st.session_state.get("max_tokens", 512),
+        temperature=st.session_state.get("temperature", 0.9),
+    )
+    response = openai.ChatCompletion.create(**params)
+
+    display = ""
+    for _ in range(5):
+        full_response = ""
+        for chunk in response:
+            content = chunk.choices[0].delta.get("content", "")
+            full_response += content
+            display += content
+            message_placeholder.markdown(postprocess_text(display) + "▌")
+
+            if chunk.choices[0].finish_reason == "stop":
+                message_placeholder.markdown(postprocess_text(display) + "▌")
+                st.session_state.messages.append(
+                    {
+                        "role": "assistant",
+                        "content": full_response
+                    }
+                )
+                return
+
+            elif chunk.choices[0].finish_reason == "function_call":
+                try:
+                    code = extract_code(full_response)
+                except:
+                    continue
+
+                with message_placeholder:
+                    with st.spinner("Executing code..."):
+                        try:
+                            res_type, res = execute(code, get_kernel())
+                        except Exception as e:
+                            st.error(f"Error when executing code: {e}")
+                            return
+
+                if res_type == "text":
+                    res = postprocess_text(res)
+                    display += "\n" + res
+                    message_placeholder.markdown(postprocess_text(display) + "▌")
+                elif res_type == "image":
+                    st.image(res)
+
+                st.session_state.messages.append(
+                    {
+                        "role": "assistant",
+                        "content": full_response,
+                        "function_call": {"name": "interpreter", "arguments": ""},
+                    }
+                )
+                st.session_state.messages.append(
+                    {
+                        "role": "function",
+                        "content": "[Image]" if res_type == "image" else res,  # 调用函数返回结果
+                    }
+                )
+
+                break
+
+        params["messages"] = st.session_state.messages
+        response = openai.ChatCompletion.create(**params)
+
+
+def main():
+    st.title("💬 Code Interpreter")
+
+    openai.api_base = os.getenv("INTERPRETER_CHAT_API_BASE", "http://192.168.20.59:7891/v1")
+    openai.api_key = os.getenv("API_KEY", "xxx")
+
+    if "messages" not in st.session_state:
+        st.session_state.messages = []
+
+    for message in st.session_state.messages:
+        role = message["role"]
+        if role in ["user", "function"]:
+            with st.chat_message("user"):
+                st.markdown(message["content"])
+        else:
+            with st.chat_message("assistant"):
+                st.markdown(postprocess_text(message["content"]))
+
+    if prompt := st.chat_input("What is up?"):
+        st.session_state.messages.append({"role": "user", "content": prompt})
+        with st.chat_message("user"):
+            st.markdown(prompt)
+
+        with st.chat_message("assistant"):
+            message_placeholder = st.empty()
+            chat_once(message_placeholder)
+
+
+if __name__ == "__main__":
+    main()