Update app.py

app.py

@@ -5,7 +5,12 @@ import torch
 # Load the Zephyr-7B-Alpha model (fully open and optimized for instruction-following)
 MODEL_NAME = "HuggingFaceH4/zephyr-7b-alpha"
 tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
-model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, torch_dtype=torch.float16, device_map="auto")
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL_NAME, 
+    torch_dtype=torch.float16, 
+    device_map="cpu",  # Forces CPU usage
+    low_cpu_mem_usage=True  # Helps reduce memory spikes
+)
 
 # Initialize conversation history if not present
 if "conversation" not in st.session_state: