app.py

import gradio as gr
from transformers import AutoModelForCausalLM
import torch

# Load the model
model_name = "wop/kosmox-gguf"
model = AutoModelForCausalLM.from_pretrained(model_name)

# Define the chat template function
def format_chat(messages, add_generation_prompt):
    formatted = "<BOS>"
    for message in messages:
        if message['from'] == 'human':
            formatted += ' ' + message['value'] + ' '
        elif message['from'] == 'gpt':
            formatted += ' ' + message['value'] + ' '
        else:
            formatted += '<|' + message['from'] + '|> ' + message['value'] + ' '
    if add_generation_prompt:
        formatted += ' '
    return formatted

# Function to generate responses
def respond(message, history, system_message, max_tokens, temperature, top_p):
    # Prepare the chat history
    messages = [{"from": "system", "value": system_message}]
    for user_msg, bot_msg in history:
        if user_msg:
            messages.append({"from": "human", "value": user_msg})
        if bot_msg:
            messages.append({"from": "gpt", "value": bot_msg})
    messages.append({"from": "human", "value": message})

    # Format the chat input for the model
    chat_input = format_chat(messages, add_generation_prompt=False)

    # Tokenize input (assuming model can handle raw text inputs internally)
    inputs = torch.tensor([ord(c) for c in chat_input]).unsqueeze(0)  # Dummy tokenization

    # Generate response
    with torch.no_grad():
        outputs = model.generate(
            input_ids=inputs,
            max_length=max_tokens,
            temperature=temperature,
            top_p=top_p,
            do_sample=True
        )
    
    response = ''.join([chr(t) for t in outputs[0].tolist() if t < 256])  # Dummy decoding
    yield response.strip()

# Define the Gradio interface
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new 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 (nucleus sampling)"),
    ],
)

# Launch the demo
if __name__ == "__main__":
    demo.launch()