1) replace gpt4.1 to gpt5.4, 2) accelerate loading speed

translator/app.py

@@ -3,56 +3,69 @@
 # author: deng
 # date  : 20250604
 
+import threading
+
 import streamlit as st
-from langchain.prompts import ChatPromptTemplate
-from langchain_ollama import ChatOllama
-from langchain_openai import ChatOpenAI
 from utils import parse_config
 
 
 class TranslatorApp:
     """Streamlit App for Language Translation"""
 
-    def __init__(self, config_path: str = 'config.yaml'):
+    def __init__(self, config_path: str = 'assets/config.toml'):
         self._config = parse_config(config_path)
         self._chain = None
 
-    def _prepare_chain(self) -> None:
-        """Prepare the chain for translation"""
+        # Start pre-warming imports and LLM client in a background thread
+        self._lock = threading.Lock()
+        threading.Thread(target=self.prepare_chain, daemon=True).start()
+
+    def prepare_chain(self) -> None:
+        """Prepare the chain for translation.
+        Thread-safe: safe to call from both background thread and main thread.
+        """
         if self._chain is not None:
             return
 
-        system_template = (
-            '你是專業的翻譯人員，請判斷接下來句子的語言是否為{source_lang}，若是的話則請將該句翻譯成{target_lang}，'
-            '並且符合{description}（僅回傳翻譯結果即可），若非的話則請回傳一模一樣的句子。'
-        )
-        user_template = '{input_text}'
+        with self._lock:
+            if self._chain is not None:
+                return
 
-        if self._config['app']['llm_mode'] == 'ollama':
-            llm = ChatOllama(
-                base_url=self._config['app']['ollama']['url'],
-                model=self._config['app']['ollama']['model_name'],
-                temperature=self._config['app']['ollama']['temperature'],
-                max_tokens=self._config['app']['ollama']['max_tokens'],
-                top_p=self._config['app']['ollama']['top_p'],
-                keep_alive=self._config['app']['ollama']['keep_alive'],
-                stop=None
-            )
-        elif self._config['app']['llm_mode'] == 'openai':
-            llm = ChatOpenAI(
-                model=self._config['app']['openai']['model_name'],
-                temperature=self._config['app']['openai']['temperature'],
-                max_tokens=self._config['app']['openai']['max_tokens'],
-                top_p=self._config['app']['openai']['top_p']
-            )
-        else:
-            raise ValueError(f'Unsupported llm model: {self._config['app']['llm_mode']}')
+            from langchain.prompts import ChatPromptTemplate
 
-        prompt = ChatPromptTemplate.from_messages([
-            ('system', system_template),
-            ('human', user_template)
-        ])
-        self._chain = prompt | llm
+            system_template = (
+                '你是專業的翻譯人員，請判斷接下來句子的語言是否為{source_lang}，若是的話則請將該句翻譯成{target_lang}，'
+                '並且符合{description}（僅回傳翻譯結果即可），若非的話則請回傳一模一樣的句子。'
+            )
+            user_template = '{input_text}'
+
+            if self._config['app']['llm_mode'] == 'ollama':
+                from langchain_ollama import ChatOllama
+                llm = ChatOllama(
+                    base_url=self._config['app']['ollama']['url'],
+                    model=self._config['app']['ollama']['model_name'],
+                    temperature=self._config['app']['ollama']['temperature'],
+                    max_tokens=self._config['app']['ollama']['max_tokens'],
+                    top_p=self._config['app']['ollama']['top_p'],
+                    keep_alive=self._config['app']['ollama']['keep_alive'],
+                    stop=None
+                )
+            elif self._config['app']['llm_mode'] == 'openai':
+                from langchain_openai import ChatOpenAI
+                llm = ChatOpenAI(
+                    model=self._config['app']['openai']['model_name'],
+                    temperature=self._config['app']['openai']['temperature'],
+                    max_tokens=self._config['app']['openai']['max_tokens'],
+                    top_p=self._config['app']['openai']['top_p']
+                )
+            else:
+                raise ValueError(f"Unsupported llm model: {self._config['app']['llm_mode']}")
+
+            prompt = ChatPromptTemplate.from_messages([
+                ('system', system_template),
+                ('human', user_template)
+            ])
+            self._chain = prompt | llm
 
     def run(self) -> None:
         """ Run the Streamlit app """
@@ -79,14 +92,12 @@ class TranslatorApp:
         translate_button = st.button('翻譯')
         output_container = st.empty()
 
-        # Prepare chain
-        self._prepare_chain()
-        
         # Action
         if input_text or translate_button:
             if not input_text.strip():
-                st.warning("請輸入要翻譯的文字")
+                st.warning('請輸入要翻譯的文字')
             else:
+                self.prepare_chain()
                 with st.spinner('翻譯中...'):
                     result = self._chain.stream({
                         'input_text': input_text,
@@ -99,6 +110,11 @@ class TranslatorApp:
                 )
 
 
+@st.cache_resource(show_spinner=False, max_entries=1)
+def get_app_instance() -> TranslatorApp:
+    return TranslatorApp()
+
+
 if __name__ == '__main__':
-    app = TranslatorApp()
+    app = get_app_instance()
     app.run()