改成小模型

大模型跑不動

app.py

@@ -61,15 +61,17 @@ def call_hf_inference(model_name, prompt, max_new_tokens=512):
         "parameters": {"max_new_tokens": max_new_tokens}
     }
     try:
-        response = requests.post(api_url, headers=HEADERS, json=payload, timeout=60)
+        response = requests.post(api_url, headers=HEADERS, json=payload, timeout=180)  # timeout 拉長
         response.raise_for_status()
         data = response.json()
         if isinstance(data, list) and "generated_text" in data[0]:
             return data[0]["generated_text"]
         elif isinstance(data, dict) and "error" in data:
-            return f"（生成失敗：{data['error']}）"
+            return f"（生成失敗：{data['error']}，請嘗試換一個模型）"
         else:
             return str(data)
+    except requests.exceptions.ReadTimeout:
+        return "（生成失敗：等待超時，請嘗試換小一點的模型或增加 timeout 秒數）"
     except Exception as e:
         return f"（生成失敗：{e}）"
 
@@ -120,11 +122,11 @@ with gr.Blocks() as demo:
     model_dropdown = gr.Dropdown(
         choices=[
             "gpt2",
-            "EleutherAI/gpt-neo-2.7B",
-            "EleutherAI/gpt-j-6B",
-            "facebook/bart-large-cnn"
+            "facebook/bart-large-cnn",
+            "bigscience/bloom-560m",
+            "bigscience/bloomz-560m"
         ],
-        value="EleutherAI/gpt-neo-2.7B",
+        value="bigscience/bloomz-560m",   # 預設比較聽得懂指令
         label="選擇生成模型"
     )
     segments_input = gr.Slider(minimum=1, maximum=10, step=1, value=5, label="段落數")