app.py

import streamlit as st
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# Load the model and tokenizer with memory optimizations
model_name = "Tom158/Nutri_Assist"
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
if model.config.pad_token_id is None:
    model.config.pad_token_id = model.config.eos_token_id

# Streamlit App Interface
st.title("Nutrition Chatbot")
user_input = st.text_input("Ask me about nutrition:")

if user_input:
    # 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']

    # Generate output with attention mask and pad token ID
    try:
        # 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)

        # Decode the output and display
        decoded_output = tokenizer.decode(outputs[0], skip_special_tokens=True)
        st.write("Decoded Answer:", decoded_output)
    except Exception as e:
        st.write("Error generating output:", str(e))