Update app.py

app.py

@@ -16,7 +16,7 @@ import operator
 from langchain_core.messages import BaseMessage, HumanMessage, AIMessage, ToolMessage, SystemMessage
 from langchain_core.tools import tool
 from langgraph.graph import StateGraph, END
-from langgraph.prebuilt import ToolNode          # <-- 改用 ToolNode
+from langgraph.prebuilt import ToolNode
 from langchain_core.utils.function_calling import convert_to_openai_function
 
 # 其他工具依赖
@@ -27,14 +27,16 @@ from youtube_transcript_api import YouTubeTranscriptApi
 # 配置常量
 # =============================================================================
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
-AGICTO_BASE_URL = os.getenv("AGICTO_BASE_URL", "https://agicto.com/model")
+# AGICTO_BASE_URL 请设置为 https://api.agicto.cn （不含 /v1）
+AGICTO_BASE_URL = os.getenv("AGICTO_BASE_URL", "https://api.agicto.cn")
 AGICTO_API_KEY = os.getenv("AGICTO_API_KEY", "")
 QWEN_MODEL = "qwen3.5-35b-a3b"
 
 # =============================================================================
-# 进度监控器（仅用于 UI，不参与评分）
+# 进度监控器
 # =============================================================================
 class ProgressMonitor:
+    # ... 保持不变 ...
     def __init__(self):
         self.current = 0
         self.total = 0
@@ -78,13 +80,17 @@ class ProgressMonitor:
             return html
 
 # =============================================================================
-# Qwen LLM 封装（通过 agicto.com API）
+# Qwen LLM 封装（修正 URL 拼接）
 # =============================================================================
 class QwenLLM:
     def __init__(self, model=QWEN_MODEL):
         self.model = model
         self.api_key = AGICTO_API_KEY
-        self.base_url = AGICTO_BASE_URL
+        # 规范化 base_url，确保末尾没有多余斜杠，并去掉可能存在的 /v1
+        base = AGICTO_BASE_URL.rstrip('/')
+        if base.endswith('/v1'):
+            base = base[:-3]  # 去掉 /v1
+        self.base_url = base
         if not self.api_key:
             print("⚠️ 未设置 AGICTO_API_KEY，请检查环境变量")
 
@@ -102,8 +108,11 @@ class QwenLLM:
         if functions:
             body["tools"] = [{"type": "function", "function": f} for f in functions]
             body["tool_choice"] = "auto"
+
+        # 统一使用 /v1/chat/completions 路径
+        url = f"{self.base_url}/v1/chat/completions"
         try:
-            resp = requests.post(f"{self.base_url}/v1/chat/completions", headers=headers, json=body, timeout=60)
+            resp = requests.post(url, headers=headers, json=body, timeout=60)
             resp.raise_for_status()
             return resp.json()
         except Exception as e:
@@ -181,13 +190,19 @@ class QwenLLM:
         return formatted
 
 # =============================================================================
-# 工具定义
+# 工具定义（analyze_image / transcribe_audio 也需要使用修正后的 URL）
 # =============================================================================
 api_url_tasks = DEFAULT_API_URL
 
+def _get_api_base():
+    base = AGICTO_BASE_URL.rstrip('/')
+    if base.endswith('/v1'):
+        base = base[:-3]
+    return base
+
 @tool
 def web_search(query: str) -> str:
-    """搜索互联网信息"""
+    # ... 不变 ...
     try:
         url = "https://api.duckduckgo.com/"
         params = {"q": query, "format": "json", "no_html": 1}
@@ -205,7 +220,7 @@ def web_search(query: str) -> str:
 
 @tool
 def web_scraper(url: str) -> str:
-    """抓取网页文本内容"""
+    # ... 不变 ...
     try:
         headers = {"User-Agent": "Mozilla/5.0"}
         resp = requests.get(url, headers=headers, timeout=15)
@@ -220,7 +235,7 @@ def web_scraper(url: str) -> str:
 
 @tool
 def calculator(expression: str) -> str:
-    """计算数学表达式"""
+    # ... 不变 ...
     try:
         import math
         allowed = {k: v for k, v in math.__dict__.items() if not k.startswith("__")}
@@ -244,7 +259,10 @@ def analyze_image(image_data: str) -> str:
             ]}],
             "max_tokens": 800
         }
-        resp = requests.post(f"{AGICTO_BASE_URL}/v1/chat/completions", headers=headers, json=body, timeout=30)
+        # 使用修正后的基地址
+        base = _get_api_base()
+        url = f"{base}/v1/chat/completions"
+        resp = requests.post(url, headers=headers, json=body, timeout=30)
         if resp.status_code == 200:
             return resp.json()["choices"][0]["message"]["content"]
         return f"图片分析失败: {resp.status_code}"
@@ -263,7 +281,9 @@ def transcribe_audio(audio_path: str) -> str:
         else:
             audio_data = open(audio_path, "rb")
         files = {"file": audio_data, "model": (None, "whisper-1")}
-        resp = requests.post(f"{AGICTO_BASE_URL}/v1/audio/transcriptions", headers=headers, files=files, timeout=60)
+        base = _get_api_base()
+        url = f"{base}/v1/audio/transcriptions"
+        resp = requests.post(url, headers=headers, files=files, timeout=60)
         if resp.status_code == 200:
             return resp.json()["text"]
         return f"转录失败: {resp.status_code}"
@@ -272,7 +292,7 @@ def transcribe_audio(audio_path: str) -> str:
 
 @tool
 def get_youtube_transcript(video_url: str) -> str:
-    """获取 YouTube 视频字幕"""
+    # ... 不变 ...
     try:
         if "watch?v=" in video_url:
             vid = video_url.split("v=")[1].split("&")[0]
@@ -287,7 +307,7 @@ def get_youtube_transcript(video_url: str) -> str:
 
 @tool
 def download_file_for_task(task_id: str) -> str:
-    """下载 GAIA 任务关联的文件（图片、音频等）并返回内容或描述"""
+    # ... 不变 ...
     try:
         url = f"{api_url_tasks}/files/{task_id}"
         resp = requests.get(url, timeout=20)
@@ -310,25 +330,15 @@ def download_file_for_task(task_id: str) -> str:
         return f"文件下载失败: {e}"
 
 # =============================================================================
-# LangGraph Agent 状态与图构建
+# LangGraph Agent
 # =============================================================================
 class AgentState(TypedDict):
     messages: Annotated[Sequence[BaseMessage], operator.add]
     final_answer: str
     task_id: str
 
-tools = [
-    web_search,
-    web_scraper,
-    calculator,
-    analyze_image,
-    transcribe_audio,
-    get_youtube_transcript,
-    download_file_for_task
-]
-
-# 创建 ToolNode（直接基于工具列表）
-tool_node = ToolNode(tools)   # 新版 LangGraph 使用 ToolNode
+tools = [web_search, web_scraper, calculator, analyze_image, transcribe_audio, get_youtube_transcript, download_file_for_task]
+tool_node = ToolNode(tools)
 
 llm = QwenLLM()
 functions = [convert_to_openai_function(t) for t in tools]
@@ -363,28 +373,19 @@ def build_graph():
     workflow.add_node("agent", agent_node)
     workflow.add_node("tools", tool_node)
     workflow.add_node("finish", finish_node)
-
     workflow.set_entry_point("agent")
     workflow.add_conditional_edges("agent", should_continue, {"tools": "tools", "finish": "finish"})
     workflow.add_edge("tools", "agent")
     workflow.add_edge("finish", END)
-
     return workflow.compile()
 
-# =============================================================================
-# Agent 类
-# =============================================================================
 class LangGraphAgent:
     def __init__(self):
         self.graph = build_graph()
         print("LangGraphAgent 初始化完成，使用模型:", QWEN_MODEL)
 
     def __call__(self, question: str, task_id: str = "") -> str:
-        state = {
-            "messages": [HumanMessage(content=question)],
-            "final_answer": "",
-            "task_id": task_id
-        }
+        state = {"messages": [HumanMessage(content=question)], "final_answer": "", "task_id": task_id}
         try:
             final_state = self.graph.invoke(state)
             return final_state["final_answer"]
@@ -393,8 +394,10 @@ class LangGraphAgent:
             return f"Error: {e}"
 
 # =============================================================================
-# 主运行函数（生成器，实时进度）
+# 主运行函数
 # =============================================================================
+import pandas as pd
+
 def run_and_submit_all(profile: gr.OAuthProfile | None) -> Generator:
     space_id = os.getenv("SPACE_ID")
     if not profile:
@@ -412,7 +415,6 @@ def run_and_submit_all(profile: gr.OAuthProfile | None) -> Generator:
         yield f"<div>Agent 初始化失败: {e}</div>", f"Agent 初始化失败: {e}", pd.DataFrame()
         return
 
-    # 获取题目
     try:
         resp = requests.get(f"{api_url}/questions", timeout=15)
         resp.raise_for_status()
@@ -448,11 +450,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None) -> Generator:
         yield monitor.get_html(), "没有答案可提交", pd.DataFrame(results_log)
         return
 
-    submission = {
-        "username": username.strip(),
-        "agent_code": agent_code,
-        "answers": answers_payload
-    }
+    submission = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
     try:
         resp = requests.post(f"{api_url}/submit", json=submission, timeout=60)
         resp.raise_for_status()
@@ -472,28 +470,27 @@ def run_and_submit_all(profile: gr.OAuthProfile | None) -> Generator:
 # =============================================================================
 # Gradio 界面
 # =============================================================================
-import pandas as pd
-
 with gr.Blocks(title="GAIA Agent") as demo:
     gr.Markdown("""
     # 🤖 GAIA Level 1 Agent (LangGraph + Qwen)
     **模型:** Qwen3.5-35B-A3B | **API:** agicto.com  
     点击按钮获取题目，Agent 自动调用工具并回答，最后提交评分。
     """)
-
     gr.LoginButton()
-
     run_btn = gr.Button("🚀 运行评测并提交", variant="primary")
-
     progress_html = gr.HTML(label="实时进度")
     status_output = gr.Textbox(label="提交结果 / 总分", lines=5, interactive=False)
     results_table = gr.DataFrame(label="题目与 Agent 答案", wrap=True)
-
     run_btn.click(
         fn=run_and_submit_all,
         outputs=[progress_html, status_output, results_table]
     )
 
 if __name__ == "__main__":
+    # 检查必要环境变量
+    if not AGICTO_API_KEY:
+        print("❌ 错误：AGICTO_API_KEY 未设置！请在 Space 的 Settings -> Repository Secrets 中添加。")
+    if "v1" in AGICTO_BASE_URL:
+        print("⚠️ 提示：AGICTO_BASE_URL 不应包含 /v1，已自动去除。请考虑设置为 https://api.agicto.cn")
     print("启动 Gradio App...")
     demo.queue().launch(server_name="0.0.0.0", server_port=7860)