app.py

import csv
import os
from datetime import datetime
from typing import Optional, Union

import gradio as gr
from huggingface_hub import HfApi, Repository

from onnx_export import convert

from apscheduler.schedulers.background import BackgroundScheduler

DATASET_REPO_URL = "https://huggingface.co/datasets/optimum/exporters"
DATA_FILENAME = "data.csv"
DATA_FILE = os.path.join("data", DATA_FILENAME)

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

DATADIR = "exporters_data"

repo: Optional[Repository] = None
# if HF_TOKEN:
#     repo = Repository(local_dir=DATADIR, clone_from=DATASET_REPO_URL, token=HF_TOKEN)


def onnx_export(token: str, model_id: str, task: str, opset: Union[int, str]) -> str:
    if token == "" or model_id == "":
        return """
        ### Invalid input 🐞

        Please fill a token and model name.
        """
    try:
        if opset == "":
            opset = None
        else:
            opset = int(opset)

        api = HfApi(token=token)

        error, commit_info = convert(api=api, model_id=model_id, task=task, opset=opset)
        if error != "0":
            return error

        print("[commit_info]", commit_info)

        # save in a private dataset
        if repo is not None:
            repo.git_pull(rebase=True)
            with open(os.path.join(DATADIR, DATA_FILE), "a") as csvfile:
                writer = csv.DictWriter(
                    csvfile, fieldnames=["model_id", "pr_url", "time"]
                )
                writer.writerow(
                    {
                        "model_id": model_id,
                        "pr_url": commit_info.pr_url,
                        "time": str(datetime.now()),
                    }
                )
            commit_url = repo.push_to_hub()
            print("[dataset]", commit_url)

        pr_revision = commit_info.pr_revision.replace("/", "%2F")

        return f"#### This model was successfully exported and a PR was open using your token, here: [{commit_info.pr_url}]({commit_info.pr_url}). If you would like to use the exported model without waiting for the PR to be approved, head to https://huggingface.co/{model_id}/tree/{pr_revision}"
    except Exception as e:
        return f"#### Error: {e}"


TTILE_IMAGE = """
<div
    style="
        display: block;
        margin-left: auto;
        margin-right: auto;
        width: 50%;
    "
>
<img src="https://i.ibb.co/m5VnjSsQ/Blue-and-White-Illustrative-Profile-Twitter-Header.png"/>
</div>
"""

TITLE = """
<div
    style="
        display: inline-flex;
        align-items: center;
        text-align: center;
        max-width: 1400px;
        gap: 0.8rem;
        font-size: 2.2rem;
    "
>
<h1 style="font-weight: 900; margin-bottom: 10px; margin-top: 10px;">
    Export transformers model to ONNX with HF Optimum exporters.
</h1>
</div>
"""

# for some reason https://huggingface.co/settings/tokens is not showing as a link by default?
DESCRIPTION = """
This Space enables automatic export of Hugging Face transformers PyTorch models to [ONNX](https://onnx.ai/). It creates a pull request on the target model repository, allowing model owners to review and merge the ONNX export, making their models accessible across a wide range of devices and platforms.

Once exported, the model can be seamlessly integrated with [HF Optimum](https://huggingface.co/docs/optimum/), maintaining compatibility with the transformers API. For detailed implementation, check out [this comprehensive guide](https://huggingface.co/docs/optimum/main/en/onnxruntime/usage_guides/models).

Quick Start Guide:
1. Obtain a read-access token from [https://huggingface.co/settings/tokens](https://huggingface.co/settings/tokens) (read access is sufficient for PR creation)
2. Enter a model ID from the Hub (e.g., [textattack/distilbert-base-cased-CoLA](https://huggingface.co/textattack/distilbert-base-cased-CoLA))
3. Click "Export to ONNX"
4. Done! You'll receive feedback on the export status and, if successful, the URL of the created pull request

Important Note: For models exceeding 2 GB, the ONNX export will be saved in an `onnx/` subfolder. When loading such models with Optimum, remember to include the `subfolder="onnx"` parameter."""

with gr.Blocks() as demo:
    gr.HTML(TTILE_IMAGE)
    gr.HTML(TITLE)

    with gr.Row():
        with gr.Column(scale=50):
            gr.Markdown(DESCRIPTION)

        with gr.Column(scale=50):
            input_token = gr.Textbox(
                max_lines=1,
                label="Hugging Face token",
            )
            input_model = gr.Textbox(
                max_lines=1,
                label="Model name",
                placeholder="textattack/distilbert-base-cased-CoLA",
            )
            input_task = gr.Textbox(
                value="auto",
                max_lines=1,
                label='Task (can be left to "auto", will be automatically inferred)',
            )
            onnx_opset = gr.Textbox(
                placeholder="for example 14, can be left blank",
                max_lines=1,
                label="ONNX opset (optional, can be left blank)",
            )

            btn = gr.Button("Export to ONNX")
            output = gr.Markdown(label="Output")

    btn.click(
        fn=onnx_export,
        inputs=[input_token, input_model, input_task, onnx_opset],
        outputs=output,
    )

def restart_space():
    HfApi().restart_space(repo_id="onnx/export", token=HF_TOKEN, factory_reboot=True)

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

demo.launch()