app.py

import gradio as gr
import subprocess
import os

from huggingface_hub import HfApi, snapshot_download
from gradio_huggingfacehub_search import HuggingfaceHubSearch

from apscheduler.schedulers.background import BackgroundScheduler

HF_TOKEN = os.environ.get("HF_TOKEN")


def process_model(
    model_id: str,
    file_path: str,
    file_path_dropdown: str,
    key: str,
    value: str,
    oauth_token: gr.OAuthToken | None,
):
    if oauth_token.token is None:
        raise ValueError("You must be logged in to use gguf-metadata-updater")

    print(f"Model ID: {model_id}")
    print(f"file_path: {file_path}")
    print(f"file_path_dropdown: {file_path_dropdown}")
    print(f"key: {key}")
    print(f"value: {value}")

    MODEL_NAME = model_id.split("/")[-1]

    if file_path_dropdown:
        FILE_PATH = file_path_dropdown
    else:
        FILE_PATH = file_path

    print(f"FILE_PATH: {FILE_PATH}")

    try:
        api = HfApi(token=oauth_token.token)

        FILE_NAME = file_path.split("/")[-1]

        api.snapshot_download(
            repo_id=model_id,
            allow_patterns=FILE_PATH,
            local_dir=f"{MODEL_NAME}",
        )
        print("Model downloaded successully!")

        metadata_update = f"python3 llama.cpp/gguf-py/scripts/gguf_set_metadata.py {MODEL_NAME}/{FILE_PATH} {key} {value} --force"
        subprocess.run(metadata_update, shell=True)
        print(f"Model metadata {key} updated to {value} successully!")

        # Upload gguf files
        api.upload_folder(
            folder_path=MODEL_NAME,
            repo_id=model_id,
            allow_patterns=["*.gguf"],
        )
        print("Uploaded successfully!")

        return f"Processing complete"
    except Exception as e:
        return f"Something went wrong: {e}"


with gr.Blocks() as demo:
    gr.Markdown("You must be logged in to use GGUF metadata updated.")
    gr.LoginButton(min_width=250)

    model_id = HuggingfaceHubSearch(
        label="Hub Model ID",
        placeholder="Search for model id on Huggingface",
        search_type="model",
    )

    file_path = gr.Textbox(lines=1, label="File path")

    file_path_dropdown = gr.Dropdown(["None"], label="File", visible=False)

    key = gr.Textbox(lines=1, label="Key")

    value = gr.Textbox(lines=1, label="Value")

    iface = gr.Interface(
        fn=process_model,
        inputs=[model_id, file_path, file_path_dropdown, key, value],
        outputs=[
            gr.Markdown(label="output"),
        ],
        title="Update metadata for a GGUF file",
        description="The space takes an HF repo, a file within that repo, a metadata key, and new metadata value to update it to.",
        api_name=False,
    )

    def updateFilePath(model_id: HuggingfaceHubSearch):
        try:
            api = HfApi()
            files = []
            for file in api.list_repo_tree(
                repo_id=model_id,
                recursive=True,
            ):
                if "of-0000" in file.path and "0001-of-" not in file.path:
                    pass
                elif file.path.endswith("gguf"):
                    files.append(file.path)

            return gr.update(visible=False), gr.update(visible=True, choices=files)
        except Exception:
            return gr.update(visible=True), gr.update(visible=False)

    model_id.change(
        fn=updateFilePath, inputs=model_id, outputs=[file_path, file_path_dropdown]
    )


def restart_space():
    HfApi().restart_space(
        repo_id="bartowski/gguf-metadata-updated", token=HF_TOKEN, factory_reboot=True
    )


scheduler = BackgroundScheduler()
scheduler.add_job(restart_space, "interval", seconds=21600)
scheduler.start()

# Launch the interface
demo.queue(default_concurrency_limit=1, max_size=5).launch(debug=True, show_api=False)