.

meta_ui.py

@@ -1,14 +1,15 @@
 import gradio as gr
 import requests
 
-API_URL = "http://localhost:30000/v1/completions"
+API_URL = "http://localhost:30000/generate"  # ✅ 使用原生 generate 接口
 API_KEY = "token-abc123"
 MODEL_NAME = "Qwen3-14b-base"
 
-# 🚫 不再格式化拼接历史，只取最后一条用户输入
+# 🚫 不再拼接上下文，只保留用户当前输入
 def build_prompt(history, user_message):
-    return user_message  # 原样作为 prompt 送入模型
+    return user_message
 
+# 主对话函数
 def chat(user_message, history, max_tokens, temperature):
     prompt = build_prompt(history, user_message)
 
@@ -26,12 +27,13 @@ def chat(user_message, history, max_tokens, temperature):
     try:
         response = requests.post(API_URL, headers=headers, json=payload, timeout=30)
         result = response.json()
-        reply = result["choices"][0]["text"].strip()
+        reply = result["text"].strip()  # ✅ /generate 接口返回字段是 text
     except Exception as e:
         reply = f"[请求失败] {e}"
 
     return reply
 
+# 测试 API 连通性
 def test_api_connection(max_tokens, temperature):
     headers = {
         "Authorization": f"Bearer {API_KEY}",
@@ -46,11 +48,12 @@ def test_api_connection(max_tokens, temperature):
 
     try:
         resp = requests.post(API_URL, headers=headers, json=payload, timeout=10)
-        out = resp.json()["choices"][0]["text"].strip()
+        out = resp.json()["text"].strip()  # ✅ 修改这里
         return f"✅ API 可用，响应: {out}"
     except Exception as e:
         return f"❌ API 请求失败: {e}"
 
+# Gradio 界面
 with gr.Blocks(title="Base 模型测试 UI") as demo:
     gr.Markdown("# 💬 Base 模型对话界面")