修改gemma3 默认模型为4b

src/podcast_transcribe/llm/llm_gemma_transfomers.py

@@ -9,7 +9,7 @@ class GemmaTransformersChatCompletion(TransformersBaseChatCompletion):
     
     def __init__(
         self,
-        model_name: str = "google/gemma-3-12b-it",
+        model_name: str = "google/gemma-3-4b-it",
         use_4bit_quantization: bool = False,
         device_map: Optional[str] = "auto",
         device: Optional[str] = None,
@@ -36,7 +36,7 @@ class GemmaTransformersChatCompletion(TransformersBaseChatCompletion):
 
 # 为了保持向后兼容性，也可以提供一个简化的工厂函数
 def create_gemma_transformers_client(
-    model_name: str = "google/gemma-3-12b-it",
+    model_name: str = "google/gemma-3-4b-it",
     use_4bit_quantization: bool = False,
     device: Optional[str] = None,
     **kwargs

src/podcast_transcribe/llm/llm_router.py

@@ -36,7 +36,7 @@ class LLMRouter:
             "gemma-transformers": {
                 "module_path": "llm_gemma_transfomers",
                 "class_name": "GemmaTransformersChatCompletion",
-                "default_model": "google/gemma-3-12b-it",
+                "default_model": "google/gemma-3-4b-it",
                 "supported_params": [
                     "model_name", "use_4bit_quantization", "device_map", 
                     "device", "trust_remote_code"
@@ -416,7 +416,7 @@ def chat_completion(
         response = chat_completion(
             messages=[{"role": "user", "content": "你好"}],
             provider="gemma-transformers",
-            model="google/gemma-3-12b-it",
+            model="google/gemma-3-4b-it",
             device="cuda",
             use_4bit_quantization=True
         )

src/podcast_transcribe/transcriber.py

@@ -61,13 +61,13 @@ class CombinedTranscriber:
             elif torch.cuda.is_available():
                 device = "cuda"
                 if not llm_model_name:
-                    llm_model_name = "google/gemma-3-12b-it"
+                    llm_model_name = "google/gemma-3-4b-it"
                 if not llm_provider:
                     llm_provider = "gemma-transformers"
             else:
                 device = "cpu"
                 if not llm_model_name:
-                    llm_model_name = "google/gemma-3-12b-it"
+                    llm_model_name = "google/gemma-3-4b-it"
                 if not llm_provider:
                     llm_provider = "gemma-transformers"