Update app.py

app.py

@@ -7,12 +7,11 @@ import threading
 MODEL_NAME = "Xerv-AI/MAXWELL"
 
 print("Loading model on CPU... this may take a few minutes.")
-# We load in 8-bit or float16 because 4-bit (bitsandbytes) is GPU-only
 tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_NAME,
     device_map="cpu",
-    torch_dtype=torch.float32  # CPU requires float32 for stability
+    torch_dtype=torch.float32 
 )
 
 # --- 2. INFERENCE LOGIC ---
@@ -29,7 +28,7 @@ def stream_maxwell(message, history):
     
     gen_kwargs = dict(
         **inputs,
-        max_new_tokens=512, # Reduced for CPU speed
+        max_new_tokens=512,
         temperature=0.3,
         do_sample=True,
         streamer=streamer,
@@ -48,20 +47,21 @@ def stream_maxwell(message, history):
             display_text = display_text.replace("</reasoning>", "</i></details>\n\n")
         yield display_text
 
-# --- 3. UI DESIGN ---
+# --- 3. UI DESIGN (Fixed for Gradio 4.0+) ---
 custom_css = """
-footer {visibility: hidden}
+footer {visibility: hidden !important;}
 .gradio-container {background-color: #121212 !important; color: white !important;}
 details { background: #1A1A1A; border-left: 2px solid #3b82f6; padding: 10px; margin: 10px 0; color: #A0A0A0; }
+summary { cursor: pointer; color: #5c94ff; font-weight: bold; }
 """
 
-demo = gr.ChatInterface(
-    fn=stream_maxwell,
-    title="M. (CPU Mode)",
-    description="The computational throne is currently on backup power (CPU). Expect slower response times.",
-    css=custom_css,
-    theme=gr.themes.Default(primary_hue="blue", neutral_hue="zinc")
-)
+# Wrap ChatInterface in Blocks to apply the CSS
+with gr.Blocks(css=custom_css, theme=gr.themes.Default(primary_hue="blue", neutral_hue="zinc")) as demo:
+    gr.ChatInterface(
+        fn=stream_maxwell,
+        title="M. (CPU Mode)",
+        description="The computational throne is currently on backup power (CPU).",
+    )
 
 if __name__ == "__main__":
     demo.queue().launch()