app.py

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
import gradio as gr

# Model and tokenizer paths
model_name = "ahmedbasemdev/llama-3.2-3b-ChatBot"

# Load the model
print("Loading the model...")
model = AutoModelForCausalLM.from_pretrained(model_name)

# Apply dynamic quantization to reduce model size and improve CPU performance
print("Applying quantization...")
model = torch.quantization.quantize_dynamic(
    model,  # Model to quantize
    {torch.nn.Linear},  # Layers to quantize (e.g., Linear layers)
    dtype=torch.qint8,  # Quantized data type
)

# Load the tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Define the inference function
def single_inference(question):
    messages = []
    messages.append({"role": "user", "content": question})

    # Tokenize the input
    input_ids = tokenizer.apply_chat_template(
        messages,
        add_generation_prompt=True,
        return_tensors="pt"
    ).to("cpu")  # Ensure everything runs on CPU

    # Generate a response
    terminators = [
        tokenizer.eos_token_id,
        tokenizer.convert_tokens_to_ids("<|eot_id|>")
    ]

    outputs = model.generate(
        input_ids,
        max_new_tokens=256,
        eos_token_id=terminators,
        do_sample=True,
        temperature=0.2,
    )
    response = outputs[0][input_ids.shape[-1]:]
    output = tokenizer.decode(response, skip_special_tokens=True)
    return output

# Gradio interface
print("Setting up Gradio app...")
interface = gr.Interface(
    fn=single_inference,
    inputs="text",
    outputs="text",
    title="Chatbot",
    description="Ask me anything!"
)

# Launch the Gradio app
interface.launch()