Update translator.py

translator.py

@@ -33,13 +33,13 @@ MAX_LENGTH = 512
 
 class Translator:
     def __init__(self):
-        self._pipeline  = None
-        self._tokenizer = None
-        self._model     = None
-        # ✅ FIX: _init_nllb() was never called before — NLLB always silently
-        #         fell through to Google Translate. Now it loads properly.
-        self._init_nllb()
-        print("[Translator] Ready")
+        self._pipeline    = None
+        self._tokenizer   = None
+        self._model       = None
+        self._nllb_loaded = False
+        # ✅ LAZY LOAD: Don't load 2.5GB model on startup
+        # Loads automatically on first translation request instead
+        print("[Translator] Ready (NLLB loads on first use)")
 
     # ── Public ──────────────────────────────────────────────────────
     def translate(self, text: str, src_lang: str, tgt_lang: str):
@@ -53,6 +53,10 @@ class Translator:
         if src_lang == tgt_lang:
             return text, "skipped (same language)"
 
+        if not self._nllb_loaded:
+            self._init_nllb()
+            self._nllb_loaded = True
+
         if self._pipeline is not None or self._model is not None:
             try:
                 return self._translate_nllb(text, src_lang, tgt_lang)
@@ -148,4 +152,4 @@ class Translator:
             return translated, "Google Translate (fallback)"
         except Exception as e:
             logger.error(f"[Translator] Google fallback also failed: {e}")
-            return f"[Translation failed: {str(e)}]", "error"
+            return f"[Translation failed: {str(e)}]", "error"