app.py

""" A simple Gradio web app to interact with the AlpineLLM model """

import base64
import gradio as gr
import os
import shutil
import torch

from huggingface_hub import hf_hub_download

from config_util import Config
from demo_inference import AlpineLLMInference
from style import custom_css

HF_TOKEN = os.environ.get("HF_TOKEN", None)


def download_model(cfg):
    """ Download the model weights from Hugging Face Hub """
    model_path = hf_hub_download(
        repo_id=cfg.repo_id, 
        filename=cfg.model_name, 
        token=HF_TOKEN,
        cache_dir=cfg.cache_dir
    )
    return model_path


def image_to_base64_data_url(filepath: str) -> str:
    """ Convert an image file to a Base64 data URL for embedding in HTML """
    try:
        ext = os.path.splitext(filepath)[1].lower()
        mime_types = {".jpg": "image/jpeg", ".jpeg": "image/jpeg", ".png": "image/png", ".gif": "image/gif", ".webp": "image/webp", ".bmp": "image/bmp"}
        mime_type = mime_types.get(ext, "image/jpeg")
        with open(filepath, "rb") as image_file:
            encoded_string = base64.b64encode(image_file.read()).decode("utf-8")
        return f"data:{mime_type};base64,{encoded_string}"
    except Exception as e:
        print(f"Error encoding image to Base64: {e}")
        return ""


def start_app():
    """ Start the web app via Gradio with custom layout """
    GOOGLE_FONTS_URL = "<link href='https://fonts.googleapis.com/css2?family=Noto+Sans+SC:wght@400;700&display=swap' rel='stylesheet'>"
    LOGO_IMAGE_PATH = "assets/background_round.png"
    logo_data_url = image_to_base64_data_url(LOGO_IMAGE_PATH) if os.path.exists(LOGO_IMAGE_PATH) else ""
    with gr.Blocks(head=GOOGLE_FONTS_URL, css=custom_css, theme=gr.themes.Soft()) as app:
        gr.HTML("""
        <div class="app-header">
            <h1>AlpineLLM Live Demo</h1>
            <p>
                A domain-specific language model for alpine storytelling. <br>
                Try asking about mountain adventures! 🏔️ <br>
                <strong>Author:</strong> <a href="https://borzyszkowski.github.io/">Bartek Borzyszkowski</a>
            </p>
        </div>
        """)
        
        gr.HTML(f"""
            <div class="app-header">
                <img src="{logo_data_url}" alt="AlpineLLM" style="max-height:10%; width: auto; margin: 10px auto; display: block;">
            </div>
            <div class="quick-links">
                <a href="https://github.com/Borzyszkowski/AlpineLLM" target="_blank">GitHub</a> | <a href="https://huggingface.co/Borzyszkowski/AlpineLLM-Tiny-10M-Base" target="_blank">Model Page</a>
            </div>
            <div class="notice">
                <strong>Heads up:</strong> This space shows a free CPU-only demo of the model, so inference may take a few seconds. Text generation of the tiny model may lack full coherence due to its limited size and character-level tokenization. Consider using the source repository to load larger pretrained weights and run inference on a GPU. 
            </div>
            <br>
            """)

        gr.Markdown("<h3> About AlpineLLM</h3>")
        gr.Markdown(
            "<p>"
            "AlpineLLM-Tiny-10M-Base is a lightweight base language model with ~10.8 million trainable parameters. It was pre-trained from scratch on raw text corpora drawn primarily from public-domain literature on alpinism, including expedition narratives and climbing essays. <br><br>"
            "This demo showcases the model's text generation capabilities within its specialized domain. Please note that AlpineLLM is a base model, and it has not been fine-tuned for downstream tasks such as summarization or dialogue. Its outputs reflect patterns learned directly from the training texts. <br><br>"
            "</p>"
        )
        with gr.Row():
            with gr.Column(scale=2):
                prompt = gr.Textbox(
                    lines=8,
                    label="Your alpine prompt...",
                    placeholder="A dawn climb on the Matterhorn..."
                )
                max_tokens = gr.Slider(50, 1000, value=300, step=10, label="Max output tokens")
                generate_btn = gr.Button("⛏️ Generate")

            with gr.Column(scale=2):
                output = gr.Textbox(lines=15, label="Generated Alpine Story", interactive=False)
        gr.Markdown("<br>")

        # Bind button click to inference
        generate_btn.click(
            fn=inference.generate_text,
            inputs=[prompt, max_tokens],
            outputs=output
        )

    app.launch(server_name="0.0.0.0", server_port=7860)


if __name__ == '__main__':
    os.chdir(os.path.dirname(os.path.abspath(__file__)))

    # Define the configuration
    cfg = {
        'cuda_id': 0,
        'model_type': 'transformer',
        'repo_id': "Borzyszkowski/AlpineLLM-Tiny-10M-Base",
        'model_name': "best_model.pt",
        'cache_dir': "./model-cache",
    }
    cfg = Config(cfg)

    # Define the hyperparameters
    hyperparam_cfg={
        "embedding_dim": 384,
        "num_heads": 6,
        "num_layers": 6,
        "dropout": 0.2,
        "context_len": 256,
        "lr": 3e-4,
    }
    hyperparam_cfg = Config(hyperparam_cfg)

    # Ensure model weights are available
    shutil.rmtree(cfg.cache_dir, ignore_errors=True)
    cfg.load_weights_path = download_model(cfg)

    # Start the application
    inference = AlpineLLMInference(cfg, hyperparam_cfg)
    start_app()