Refactor response generation logic in `generator.py`

- Adjust sentence length evaluation for end mark conditions.
- Update chunk filtering to handle `<think>` and newline tags consistently.
- Fix inconsistency in sentence preprocessing with remaining content handling.
- Clean up unnecessary whitespace in method definitions.

src/voice_dialogue/services/text/generator.py

@@ -97,13 +97,12 @@ class LLMResponseGenerator(BaseThread):
         # 普通句子的判断逻辑
         if is_chinese_sentence:
             sentence_words = len(sentence)
+            return sentence_words > 4
         else:
             sentence_words = len(sentence.split())
+            return sentence_words > 4 or (sentence_words > 2 and sentence_end_mark in {'.', '?', '!', })
 
-        return sentence_words > 4
-
-    def _send_sentence_to_queue(self, voice_task: VoiceTask, sentence: str,
-                                answer_index: int) -> None:
+    def _send_sentence_to_queue(self, voice_task: VoiceTask, sentence: str, answer_index: int) -> None:
         """将句子发送到队列"""
         voice_task.answer_index = answer_index
         voice_task.answer_sentence = sentence.strip()
@@ -149,9 +148,9 @@ class LLMResponseGenerator(BaseThread):
 
         try:
             for chunk in pipeline.stream(input={'input': user_question}, config=config):
-                if not chunk.content.strip():
+                if not chunk.content:
                     continue
-                elif chunk.content.strip() in {'<think>', '</think>'}:
+                elif chunk.content in {'<think>', '\n\n', '</think>'}:
                     continue
 
                 chunk_content = f'{chunk.content}'
@@ -161,6 +160,7 @@ class LLMResponseGenerator(BaseThread):
 
                 sentence = preprocess_sentence_text(chunks)
                 if not sentence:
+                    chunks.append(remain_content)
                     continue
 
                 # 检查是否应该结束当前句子
@@ -192,9 +192,9 @@ class LLMResponseGenerator(BaseThread):
 
     def run(self):
         model_path = paths.LLM_MODELS_PATH / 'qwen' / 'Qwen3-8B-Q6_K.gguf'
-        
+
         model_params = get_llm_model_params()
-        
+
         # 打印芯片信息和优化配置
         chip_summary = get_apple_silicon_summary()
         print(f"检测到芯片: {chip_summary['chip_name']}")
@@ -202,7 +202,7 @@ class LLMResponseGenerator(BaseThread):
         print(f"使用线程数: {chip_summary['optimal_n_threads']} (仅使用性能核心)")
         print(f"上下文窗口: {chip_summary['optimal_n_ctx']}")
         print(f"配置说明: {chip_summary['config_note']}")
-        
+
         self.model_instance = create_langchain_chat_llamacpp_instance(
             local_model_path=model_path, model_params=model_params
         )