Update app.py

app.py

@@ -1,10 +1,89 @@
+import os
 import gradio as gr
+import spaces
+import fitz
+import docx
+from huggingface_hub import InferenceClient
 
-with gr.Blocks(fill_height=True) as demo:
-    with gr.Sidebar():
-        gr.Markdown("# Inference Provider")
-        gr.Markdown("This Space showcases the meta-llama/Meta-Llama-3-70B-Instruct model, served by the novita API. Sign in with your Hugging Face account to use this API.")
-        button = gr.LoginButton("Sign in")
-    gr.load("models/meta-llama/Meta-Llama-3-70B-Instruct", accept_token=button, provider="novita")
+# 1. 引擎配置：Llama-3-70B-Instruct (全球开源标杆)
+model_id = "meta-llama/Meta-Llama-3-70B-Instruct"
+hf_token = os.getenv("HF_TOKEN")
+
+client = InferenceClient(
+    model=model_id, 
+    token=hf_token,
+    headers={"Authorization": f"Bearer {hf_token}"} if hf_token else None
+)
+
+def extract_text(file_path):
+    if not file_path: return ""
+    try:
+        ext = str(file_path).lower()
+        if ext.endswith(".pdf"):
+            doc = fitz.open(file_path)
+            text = "".join([page.get_text() for page in doc])
+            doc.close()
+            return text[:15000]
+        elif ext.endswith(".docx"):
+            doc = docx.Document(file_path)
+            return "\n".join([p.text for p in doc.paragraphs])[:15000]
+        elif ext.endswith(".txt"):
+            with open(file_path, 'r', encoding='utf-8', errors='ignore') as f:
+                return f.read()[:15000]
+        return ""
+    except:
+        return ""
+
+@spaces.GPU(duration=60)
+def chat_fn(message, history):
+    clean_messages = []
+    # 系统提示词：设定为国际化法律与技术专家
+    clean_messages.append({"role": "system", "content": "You are a professional global intellectual property and technical expert. Answer accurately and professionally."})
+    
+    for turn in history:
+        clean_messages.append({"role": turn["role"], "content": str(turn["content"])})
     
-demo.launch()
+    prompt_text = message.get("text", "")
+    files = message.get("files", [])
+    
+    if files:
+        f = files[0]
+        file_path = f.get("path") if isinstance(f, dict) else f
+        context = extract_text(file_path)
+        if context:
+            prompt_text = f"【Reference Document】:\n{context}\n\n【Instruction】: {prompt_text}"
+        
+    clean_messages.append({"role": "user", "content": prompt_text})
+
+    response = ""
+    try:
+        stream = client.chat_completion(clean_messages, max_tokens=4096, stream=True, temperature=0.5)
+        for msg in stream:
+            if hasattr(msg, 'choices') and len(msg.choices) > 0:
+                delta_content = msg.choices[0].delta.content
+                if delta_content:
+                    # Llama 的 API 响应通常比较标准，但我们依然保留清洗逻辑
+                    response += str(delta_content)
+                    yield response
+        
+        if response:
+            yield response + "\n\n---\n> **OpenIPOS MegaNode | Meta-Llama-3 全球通用节点**"
+            
+    except Exception as e:
+        yield f"⚠️ 算力调度提示: {str(e)}。由于 Llama-3 节点访问量巨大，请刷新重试。"
+
+# 3. 统一品牌装修
+with gr.Blocks(fill_height=True) as demo:
+    gr.HTML("""
+        <div style="display: flex; align-items: center; gap: 20px; padding: 10px;">
+            <img src="https://s3.bmp.ovh/2026/03/30/rIAGKg0O.png" style="height: 60px; width: auto;">
+            <div>
+                <h1 style="margin: 0; font-size: 24px;">OpenIPOS MegaNode</h1>
+                <p style="margin: 0; color: #666;">全球通用人工智能算力出口 (Meta-Llama-3-70B)</p>
+            </div>
+        </div>
+    """)
+    gr.ChatInterface(chat_fn, multimodal=True)
+    gr.Markdown("© 2026 OpenIPOS Global Limited. 算力驱动：Together AI / Novita Relay")
+
+demo.launch(theme=gr.themes.Soft())