Update app.py

app.py

@@ -101,10 +101,11 @@ def stream_chat(message, history: list, temperature: float, max_length: int, top
     
     model = AutoModelForCausalLM.from_pretrained(
         MODEL_ID,
-        torch_dtype=torch.bfloat16,
-        low_cpu_mem_usage=True,
-        trust_remote_code=True
+        torch_dtype=torch.float32,  # Use float32 for CPU
+        trust_remote_code=True,
+        low_cpu_mem_usage=True  # Optimize for CPU
     )
+
         
     print(f'message is - {message}')
     print(f'history is - {history}')
@@ -243,7 +244,7 @@ with gr.Blocks(css=CSS, theme="soft", fill_height=True) as demo:
             ),
         ],
     ),
-    gr.Examples(EXAMPLES, [chat_input])
+    #gr.LExamples(EXAMPES, [chat_input])
 
 if __name__ == "__main__":