app.py

import os
from threading import Thread
from typing import Iterator

import gradio as gr  # Importing Gradio for creating UI interfaces.
import spaces  # Import for using Hugging Face Spaces functionalities.
import torch  # PyTorch library for deep learning applications.
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer  # Import necessary components from Hugging Face's Transformers.

# Constants for maximum token lengths and defaults.
MAX_MAX_NEW_TOKENS = 2048
DEFAULT_MAX_NEW_TOKENS = 1024
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))

# Initial description for the UI interface, showcasing the AI version and creator.
DESCRIPTION = """\
# Masher AI v6.1 7B
This Space demonstrates Masher AI v6.1 7B by Maheswar.
"""

# Check for GPU availability, append a warning to the description if running on CPU.
if not torch.cuda.is_available():
    DESCRIPTION += "\n<p>Running on CPU! This demo does not work on CPU.</p>"

# If a GPU is available, load the model and tokenizer with specific configurations.
if torch.cuda.is_available():
    model_id = "mahiatlinux/MasherAI-7B-v6.1"
    model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", load_in_4bit=True)
    tokenizer = AutoTokenizer.from_pretrained(model_id)
    tokenizer.use_default_system_prompt = False

# Define a function decorated to use GPU and enable queue for processing the generation tasks.
@spaces.GPU(enable_queue=True)
def generate(
    message: str,
    chat_history: list[tuple[str, str]],
    system_prompt: str,
    max_new_tokens: int = 1024,
    temperature: float = 0.1,
    top_p: float = 0.9,
    top_k: int = 50,
    repetition_penalty: float = 1.2,
) -> Iterator[str]:
    # Preparing conversation history for processing.
    conversation = []
    # Adding system prompt.
    # conversation.append({"from": "human", "value": system_prompt})
    # Extending the conversation history with user and assistant interactions.
    for user, assistant in chat_history:
        conversation.extend([{"from": "human", "value": user}, {"from": "gpt", "value": assistant}])
    # Adding the latest message from the user to the conversation.
    conversation.append({"from": "human", "value": message})

    # Tokenize and prepare the input, handle exceeding token lengths.
    input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt", add_generation_prompt=True)
    if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
        input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
        gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
    input_ids = input_ids.to(model.device)

    # Setup for asynchronous text generation.
    streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)
    generate_kwargs = dict(
        {"input_ids": input_ids},
        streamer=streamer,
        max_new_tokens=max_new_tokens,
        do_sample=True,
        top_p=top_p,
        top_k=top_k,
        temperature=temperature,
        num_beams=1,
        repetition_penalty=repetition_penalty,
    )
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()

    # Collect and yield generated outputs as they become available.
    outputs = []
    for text in streamer:
        outputs.append(text)
        yield "".join(outputs)

# Setup Gradio interface for chat, including additional controls for the generation parameters.
chat_interface = gr.ChatInterface(
    fn=generate,
    fill_height=True,
    additional_inputs=[
        gr.Textbox(label="System prompt", lines=6),
        gr.Slider(
            label="Max new tokens",
            minimum=1,
            maximum=MAX_MAX_NEW_TOKENS,
            step=1,
            value=DEFAULT_MAX_NEW_TOKENS,
        ),
        gr.Slider(
            label="Temperature",
            minimum=0.1,
            maximum=4.0,
            step=0.1,
            value=0.6,
        ),
        gr.Slider(
            label="Top-p (nucleus sampling)",
            minimum=0.05,
            maximum=1.0,
            step=0.05,
            value=0.9,
        ),
        gr.Slider(
            label="Top-k",
            minimum=1,
            maximum=1000,
            step=1,
            value=50,
        ),
        gr.Slider(
            label="Repetition penalty",
            minimum=1.0,
            maximum=2.0,
            step=0.05,
            value=1.2,
        ),
    ],
    stop_btn=None,
    examples=[
        # Examples to assist users in starting conversations with the AI.
    ],
)

chatbot=gr.Chatbot(height=450, label='Gradio ChatInterface')
# Setup and launch the Gradio demo with Blocks API.
with gr.Blocks(css="style.css", fill_height=True) as demo:
    gr.Markdown(DESCRIPTION)
    chat_interface.render()

# Main entry point to start the web application if this script is run directly.
if __name__ == "__main__":
    demo.queue(max_size=20).launch()