app.py

import gradio as gr
from unsloth import FastLanguageModel
import torch

# Load the model and tokenizer locally
max_seq_length = 2048
model_name_or_path = "michailroussos/model_llama_8d"

# Load model and tokenizer using unsloth
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name=model_name_or_path,
    max_seq_length=max_seq_length,
    load_in_4bit=True,
)
FastLanguageModel.for_inference(model)  # Enable optimized inference

# Define the response function
def respond(message, history, system_message, max_tokens, temperature, top_p):
    # Print the inputs at the start
    print("===== Respond Function Called =====")
    print(f"Received message: {message}")
    print(f"Current history: {history}")
    
    # Prepare the messages for the model
    messages = []
    if history:
        print("Adding previous messages to the history...")
        for entry in history:
            messages.append({"role": "user", "content": entry[0]})
            messages.append({"role": "assistant", "content": entry[1]})

    # Add the current user message
    print(f"Adding current user message: {message}")
    messages.append({"role": "user", "content": message})

    # Print the messages list before tokenization
    print("Messages before tokenization:", messages)

    # Tokenize the input (prepare the data for the model)
    print("Preparing the input for the model...")
    inputs = tokenizer.apply_chat_template(
        messages,
        tokenize=True,
        add_generation_prompt=True,
        return_tensors="pt",
    ).to("cuda" if torch.cuda.is_available() else "cpu")

    # Print the tokenized inputs
    print(f"Tokenized inputs: {inputs}")

    # Generate the response
    attention_mask = inputs.ne(tokenizer.pad_token_id).long()
    print(f"Attention mask: {attention_mask}")

    try:
        generated_tokens = model.generate(
            input_ids=inputs,
            attention_mask=attention_mask,
            max_new_tokens=max_tokens,
            use_cache=True,
            temperature=temperature,
            top_p=top_p,
        )
    except Exception as e:
        print(f"Error during model generation: {e}")
        return []

    # Decode the generated response
    response = tokenizer.decode(generated_tokens[0], skip_special_tokens=True)
    print(f"Generated response: {response}")
    
    # Check and filter out unwanted system-level messages or metadata
    if "system" in response.lower():
        print("System message detected. Replacing with fallback response.")
        response = "Hello! How can I assist you today?"

    # Prepare the return format for Gradio (list of [user_message, assistant_message])
    if history is None:
        history = []
    
    # Append the new conversation turn
    history.append([message, response])

    return history

# Define the Gradio interface
demo = gr.ChatInterface(
    fn=respond,
    additional_inputs=[
        gr.Textbox(value="You are a helpful assistant.", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p"),
    ],
)

if __name__ == "__main__":
    demo.launch(share=False)  # Use share=False for local testing