Update app.py

app.py

@@ -27,10 +27,10 @@ else:
 # -------------------------------
 MODEL_MAP = {
     "Auto": None,
-    "Gemma-2B": "google/gemma-2b",
-    "BTLM-3B-8K": "tiiuae/btlm-3b-8k-base",
-    "DistilGPT2": "distilgpt2",
-    "BART-Base": "facebook/bart-base"
+    "Gemma-2B": "google/gemma-2b",             # gated，需要 Access repository
+    "BTLM-3B-8K": "cerebras/btlm-3b-8k-base",  # 正確 repo
+    "DistilGPT2": "distilgpt2",                # 小模型
+    "BART-Base": "facebook/bart-base"          # 小模型
 }
 
 # -------------------------------
@@ -68,9 +68,9 @@ def test_models():
                 tokenizer=local_dir,
                 device_map="cpu"
             )
-            print(f"✅ 模型 {name} ({local_dir}) 可用")
+            print(f"✅ 模型 {name} 可用")
         except Exception as e:
-            print(f"❌ 模型 {name} ({local_dir}) 無法載入: {e}")
+            print(f"❌ 模型 {name} 無法載入: {e}")
 
 test_models()
 
@@ -112,7 +112,7 @@ def get_pipeline(model_name):
         local_path = LOCAL_MODEL_DIRS.get(model_name)
         if not local_path:
             raise ValueError(f"❌ 模型 {model_name} 尚未下載")
-        print(f"🔄 正在載入本地模型 {model_name} from {local_path}")
+        print(f"🔄 正在載入模型 {model_name} from {local_path}")
         generator = pipeline(
             "text-generation",
             model=local_path,
@@ -135,11 +135,13 @@ def call_local_inference(model_name, prompt, max_new_tokens=512):
 # -------------------------------
 def pick_model_auto(segments):
     if segments <= 3:
-        return "Gemma-2B"
+        return "DistilGPT2"   # 短文用最小模型，快
     elif segments <= 6:
-        return "BTLM-3B-8K"
+        return "Gemma-2B"     # 中篇用 Gemma-2B
+    elif segments <= 8:
+        return "BTLM-3B-8K"   # 長文用 BTLM
     else:
-        return "BART-Base"
+        return "BART-Base"    # 超長用 Bart-base
 
 def generate_article_progress(query, model_name, segments=5):
     docx_file = "/tmp/generated_article.docx"