import csv
import os
from datetime import datetime
from typing import Optional

import gradio as gr
from huggingface_hub import HfApi, Repository

from optimum_neuron_export import convert
from gradio_huggingfacehub_search import HuggingfaceHubSearch
from apscheduler.schedulers.background import BackgroundScheduler

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

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

DATADIR = "neuron_exports_data"

repo: Optional[Repository] = None
# Uncomment if you want to push to dataset repo with token
# if HF_TOKEN:
#     repo = Repository(local_dir=DATADIR, clone_from=DATASET_REPO_URL, token=HF_TOKEN)


def neuron_export(model_id: str, task: str) -> str:
    if not model_id:
        return f"### Invalid input 🐞 Please specify a model name, got {model_id}"

    try:
        api = HfApi(token=HF_TOKEN)  # Use HF_TOKEN if available, else anonymous
        token = HF_TOKEN  # Pass token to convert only if available

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

        print("[commit_info]", commit_info)

        # Save in a private dataset if repo initialized
        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"#### Success 🔥 This model was successfully exported and a PR was opened: [{commit_info.pr_url}]({commit_info.pr_url}). To use the model before the PR is approved, go to https://huggingface.co/{model_id}/tree/{pr_revision}"

    except Exception as e:
        return f"#### Error: {e}"


TITLE_IMAGE = """
<div style="display: block; margin-left: auto; margin-right: auto; width: 50%;">
<img src="https://huggingface.co/spaces/optimum/neuron-export/resolve/main/huggingfaceXneuron.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;">
    🤗 Optimum Neuron Model Exporter
</h1>
</div>
"""

DESCRIPTION = """
Export 🤗 Transformers models hosted on the Hugging Face Hub to AWS Neuron-optimized format for Inferentia/Trainium acceleration.

*Features:*
- Automatically opens PR with Neuron-optimized model
- Preserves original model weights
- Adds proper tags to model card

*Note:*
- PR creation requires the Space owner to have a valid write token set via HF_WRITE_TOKEN
"""

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

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

        with gr.Column(scale=50):
            input_model = HuggingfaceHubSearch(
                label="Hub model ID",
                placeholder="Search for model ID on the hub",
                search_type="model",
            )
            input_task = gr.Textbox(
                value="auto",
                max_lines=1,
                label='Task (can be left to "auto", will be automatically inferred)',
            )
            btn = gr.Button("Export to Neuron")
            output = gr.Markdown(label="Output")

    btn.click(
        fn=neuron_export,
        inputs=[input_model, input_task],
        outputs=output,
    )


if __name__ == "__main__":
    def restart_space():
        if HF_TOKEN:
            HfApi().restart_space(repo_id="optimum/neuron-export", token=HF_TOKEN, factory_reboot=True)

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

    demo.launch()