Update app.py

app.py

@@ -29,22 +29,20 @@ def load_model():
 
     base_model = AutoModelForCausalLM.from_pretrained(
         MODEL_ID,
-        torch_dtype=torch.float32
+        torch_dtype=torch.float32,
+        device_map=None
     )
 
-    # Load LoRA from /lora (FAIL if missing — no silent fallback)
-    model = PeftModel.from_pretrained(
-        base_model,
-        f"{MODEL_ID}/lora",
-        is_trainable=False
-    )
+    # Load LoRA if present
+    try:
+        model = PeftModel.from_pretrained(base_model, MODEL_ID)
+    except Exception:
+        model = base_model
 
     print(model.peft_config)
 
-    model.set_adapter("default")  # ensure adapter is active
     model.to("cpu")
     model.eval()
-
     return tokenizer, model
 
 
@@ -61,9 +59,7 @@ if st.button("Send") and prompt.strip():
     st.session_state.history.append(("You", prompt))
 
     system_instructions = """You are Kushina.
-
 You operate in exactly ONE of two modes.
-
 ====================
 MODE: CHAT
 ====================
@@ -73,12 +69,11 @@ Rules:
 - Neutral → neutral.
 - Serious → serious.
 - Rude → curt or dismissive.
-- Mirroring of emotions is very important and myst be talen as priority
+- Mirroring of emotions is very important and must be talen as priority
 - No enthusiasm by default.
 - No emojis unless the user uses them first.
 - Replies must be short (1–3 sentences).
 - No explanations unless explicitly asked.
-
 ====================
 MODE: CODE
 ====================
@@ -92,7 +87,6 @@ Rules:
 - Follow standard best practices.
 - Be deterministic and professional.
 - Finish the task completely.
-
 ====================
 MODE SELECTION
 ====================
@@ -105,9 +99,7 @@ Automatically switch to MODE: CODE if the user requests:
 - API
 - algorithm
 - app
-
 Otherwise, use MODE: CHAT.
-
 ====================
 IDENTITY
 ====================
@@ -163,4 +155,4 @@ for speaker, text in st.session_state.history:
     if speaker == "You":
         st.markdown(f"**You:** {text}")
     else:
-        st.markdown(f"**Niche:** {text}")
+        st.markdown(f"**Niche:** {text}")