Update app.py

app.py

@@ -2,9 +2,9 @@ import streamlit as st
 from transformers import AutoModelForCausalLM, AutoTokenizer
 import torch
 
-# Load the model and tokenizer
+# Load the model and tokenizer with memory optimizations
 model_name = "Tom158/Nutri_Assist"
-model = AutoModelForCausalLM.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, device_map="auto")
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 
 # Set pad token if not already set
@@ -16,25 +16,17 @@ st.title("Nutrition Chatbot")
 user_input = st.text_input("Ask me about nutrition:")
 
 if user_input:
-    # Use encode_plus to get both input_ids and attention_mask
-    inputs = tokenizer.encode_plus(user_input, return_tensors="pt", padding=True, truncation=True)
+    # Truncate input and convert to tensors
+    inputs = tokenizer.encode_plus(user_input, return_tensors="pt", padding=True, truncation=True, max_length=512)
     input_ids = inputs['input_ids']
     attention_mask = inputs['attention_mask']
-    
-    # Print the input tensors for debugging
-    st.write("Input IDs:", input_ids)
-    st.write("Attention Mask:", attention_mask)
-    
+
     # Generate output with attention mask and pad token ID
     try:
-        # Increased max_length to 100 for more space for generation
-        # Added temperature and top_k for better randomness and diversity
-        outputs = model.generate(input_ids, attention_mask=attention_mask, max_length=150, 
+        # Limit output length to save memory
+        outputs = model.generate(input_ids, attention_mask=attention_mask, max_length=100, 
                                  temperature=0.7, top_k=50, num_return_sequences=1)
-        
-        # Debugging model raw output (just the token ids)
-        st.write("Model Output (Raw Token IDs):", outputs)
-        
+
         # Decode the output and display
         decoded_output = tokenizer.decode(outputs[0], skip_special_tokens=True)
         st.write("Decoded Answer:", decoded_output)