Update src/app/main_app.py

src/app/main_app.py

@@ -110,29 +110,22 @@ html, body {
 
 def preload_models():
     """
-    Preloads Whisper, Qwen, and NLLB models only one time.
-    This prevents UI freeze or repeated loading during conversation.
+    Loads all heavy models ONCE at startup.
+    Safe for HuggingFace Spaces CPU environment.
     """
-    if st.session_state.get("models_loaded"):
-        return
-
-    from .conversation_core import load_whisper, load_partner_lm, load_nllb
-
-    with st.spinner("Loading language & speech models (one-time)…"):
-        try:
-            load_whisper()
-        except Exception:
-            pass
-        try:
-            load_partner_lm()
-        except Exception:
-            pass
-        try:
-            load_nllb()
-        except Exception:
-            pass
-
-    st.session_state["models_loaded"] = True
+    from .conversation_core import load_partner_lm, load_whisper_pipe
+
+    # Qwen LM
+    try:
+        load_partner_lm()
+    except Exception as e:
+        print("[preload_models] ERROR loading Qwen model:", e)
+
+    # Whisper ASR
+    try:
+        load_whisper_pipe()
+    except Exception as e:
+        print("[preload_models] ERROR loading Whisper pipeline:", e)
 
 
 def get_conv_manager() -> ConversationManager: