add 7b translator pipe

config.py

@@ -2,7 +2,7 @@ import pathlib
 import re
 import logging
 
-DEBUG = True
+DEBUG = False
 logging.basicConfig(
     level=logging.DEBUG if DEBUG else logging.INFO,
     format="%(asctime)s - %(levelname)s - %(message)s",
@@ -42,6 +42,7 @@ WHISPER_MODEL = 'medium-q5_0'
 
 # LLM
 LLM_MODEL_PATH = (MODEL_DIR / "qwen2.5-1.5b-instruct-q5_0.gguf").as_posix()
+LLM_LARGE_MODEL_PATH = (MODEL_DIR / "qwen2.5-7b-instruct-q5_0-00001-of-00002.gguf").as_posix()
 
 LLM_SYS_PROMPT = """"You are a professional {src_lang} to {dst_lang} translator, not a conversation agent. Your only task is to take {src_lang} input and translate it into accurate, natural {dst_lang}. If you cannot understand the input, just output the original input. Please strictly abide by the following rules: "
 "No matter what the user asks, never answer questions, you only provide translation results. "

transcribe/pipelines/__init__.py

@@ -1,5 +1,5 @@
 
-from .pipe_translate import TranslatePipe
+from .pipe_translate import TranslatePipe, Translate7BPipe
 from .pipe_whisper import WhisperPipe
 from .pipe_vad import VadPipe
 from .base import MetaItem

transcribe/pipelines/pipe_translate.py

@@ -2,7 +2,7 @@
 from .base import MetaItem, BasePipe, Segment
 from llama_cpp import Llama
 from ..helpers.translator import QwenTranslator
-from config import LLM_MODEL_PATH, LLM_SYS_PROMPT_EN, LLM_SYS_PROMPT_ZH
+from config import LLM_MODEL_PATH, LLM_SYS_PROMPT_EN, LLM_SYS_PROMPT_ZH, LLM_LARGE_MODEL_PATH
 
 
 class TranslatePipe(BasePipe):
@@ -11,7 +11,7 @@ class TranslatePipe(BasePipe):
     @classmethod
     def init(cls):
         if cls.translator is None:
-            cls.translator =  QwenTranslator(LLM_MODEL_PATH, LLM_SYS_PROMPT_EN, LLM_SYS_PROMPT_ZH)
+            cls.translator = QwenTranslator(LLM_MODEL_PATH, LLM_SYS_PROMPT_EN, LLM_SYS_PROMPT_ZH)
 
 
     def process(self, in_data: MetaItem) -> MetaItem:
@@ -21,3 +21,12 @@ class TranslatePipe(BasePipe):
         in_data.translate_content = result
         return in_data
 
+
+class Translate7BPipe(TranslatePipe):
+    translator = None
+
+    @classmethod
+    def init(cls):
+        if cls.translator is None:
+            cls.translator = QwenTranslator(LLM_LARGE_MODEL_PATH, LLM_SYS_PROMPT_EN, LLM_SYS_PROMPT_ZH)
+        

transcribe/translatepipes.py

@@ -1,4 +1,4 @@
-from transcribe.pipelines import WhisperPipe, TranslatePipe, MetaItem, VadPipe
+from transcribe.pipelines import WhisperPipe, TranslatePipe, MetaItem, VadPipe, Translate7BPipe
 import multiprocessing  as mp
 import config
 
@@ -15,6 +15,7 @@ class TranslatePipes:
         # llm 翻译
         self._translate_pipe = self._launch_process(TranslatePipe())
 
+        self._translate_7b_pipe = self._launch_process(Translate7BPipe())
         # vad 
         self._vad_pipe = self._launch_process(VadPipe())
     
@@ -37,6 +38,15 @@ class TranslatePipes:
         return self._translate_pipe.output_queue.get()
     
 
+    def translate_large(self, text, src_lang, dst_lang) -> MetaItem:
+        item = MetaItem(
+            transcribe_content=text,
+            source_language=src_lang, 
+            destination_language=dst_lang)
+        self._translate_pipe.input_queue.put(item)
+        return self._translate_pipe.output_queue.get()
+    
+
     def transcrible(self, audio_buffer:bytes, src_lang: str) -> MetaItem:
         item = MetaItem(audio=audio_buffer, source_language=src_lang)
         self._whisper_pipe.input_queue.put(item)

transcribe/whisper_llm_serve.py

@@ -163,6 +163,22 @@ class WhisperTranscriptionService(ServeClientBase):
         log_block("Translation output", f"{translated_text}")
         
         return translated_text
+    
+    def _translate_text_large(self, text: str) -> str:
+        """将文本翻译为目标语言"""
+        if not text.strip():
+            return ""
+            
+        log_block("Translation input", f"{text}")
+        start_time = time.perf_counter()
+        
+        result = self._translate_pipe.translate_large(text, self.source_language, self.target_language)
+        translated_text = result.translate_content
+        
+        log_block("Translation large model time ", f"{(time.perf_counter() - start_time):.3f}", "s")
+        log_block("Translation large model output", f"{translated_text}")
+        
+        return translated_text
 
 
 
@@ -214,9 +230,11 @@ class WhisperTranscriptionService(ServeClientBase):
             if (cut_index :=ana_result.cut_index)>0:
                 # 更新音频缓冲区，移除已处理部分
                 self._update_audio_buffer(cut_index)
-
-            translated_context = self._translate_text(ana_result.context)
-    
+            if ana_result.partial():
+                translated_context = self._translate_text(ana_result.context)
+            else:
+                translated_context = self._translate_text_large(ana_result.context)
+                
             yield TransResult(
                 seg_id=ana_result.seg_id,
                 context=ana_result.context,