app.py

import gradio as gr
from huggingface_hub import HfApi, hf_hub_download
from huggingface_hub.repocard import metadata_load
from huggingface_hub import ModelCard
import requests
import re
import pandas as pd
import os

# --------------------
# Helper functions
# --------------------
def pass_emoji(passed):
    return "✅" if passed else "❌"

api = HfApi()
USERNAMES_DATASET_ID = "huggingface-course/audio-course-u7-hands-on"
HF_TOKEN = os.environ.get("HF_TOKEN")

def get_user_models(hf_username, task):
    """
    List the user's models for a given task
    """
    try:
        models = api.list_models(author=hf_username, filter=[task])
    except Exception:
        return []

    user_model_ids = [x.modelId for x in models]

    # map task to dataset
    if task == "audio-classification":
        dataset = 'marsyas/gtzan'
    elif task == "automatic-speech-recognition":
        dataset = 'PolyAI/minds14'
    elif task == "text-to-speech":
        dataset = ""
    else:
        print(f"Unsupported task: {task}")
        return []

    dataset_specific_models = []
    for model in user_model_ids:
        try:
            meta = get_metadata(model)
            if meta is None:
                continue
            if dataset == "" or meta.get("datasets") == [dataset]:
                dataset_specific_models.append(model)
        except Exception:
            continue

    return dataset_specific_models

def get_metadata(model_id):
    """Load model metadata safely"""
    try:
        readme_path = hf_hub_download(model_id, filename="README.md", token=HF_TOKEN)
        return metadata_load(readme_path)
    except requests.exceptions.HTTPError:
        return None
    except Exception:
        return None

def extract_metric(model_card_content, task):
    """Extract metric from model card content"""
    accuracy_pattern = r"(?:Accuracy|eval_accuracy): (\d+\.\d+)"
    wer_pattern = r"Wer: (\d+\.\d+)"
    pattern = accuracy_pattern if task == "audio-classification" else wer_pattern
    match = re.search(pattern, model_card_content)
    return float(match.group(1)) if match else None

def parse_metrics(model, task):
    try:
        card = ModelCard.load(model)
        return extract_metric(card.content, task)
    except Exception:
        return None

def calculate_best_result(user_models, task):
    """Calculate best result for a task"""
    best_model = ""
    best_result = -100 if task == "audio-classification" else 100
    larger_is_better = task == "audio-classification"

    for model in user_models:
        metric = parse_metrics(model, task)
        if metric is None:
            continue
        if (larger_is_better and metric > best_result) or (not larger_is_better and metric < best_result):
            best_result = metric
            meta = get_metadata(model)
            if meta:
                best_model = meta.get('model-index', [{}])[0].get("name", model)
    return best_result, best_model

# --------------------
# Certification logic
# --------------------
def certification(hf_username):
    results_certification = [
        {"unit": "Unit 4: Audio Classification", "task": "audio-classification", "baseline_metric": 0.87, "best_result": 0, "best_model_id": "", "passed_": False},
        {"unit": "Unit 5: Automatic Speech Recognition", "task": "automatic-speech-recognition", "baseline_metric": 0.37, "best_result": 0, "best_model_id": "", "passed_": False},
        {"unit": "Unit 6: Text-to-Speech", "task": "text-to-speech", "baseline_metric": 0, "best_result": 0, "best_model_id": "", "passed_": False},
        {"unit": "Unit 7: Audio applications", "task": "demo", "baseline_metric": 0, "best_result": 0, "best_model_id": "", "passed_": False},
    ]

    for unit in results_certification:
        task = unit["task"]
        if task == "audio-classification":
            try:
                models = get_user_models(hf_username, task)
                best_result, best_model_id = calculate_best_result(models, task)
                unit["best_result"] = best_result
                unit["best_model_id"] = best_model_id
                unit["passed_"] = best_result >= unit["baseline_metric"]
            except Exception:
                pass
        elif task == "automatic-speech-recognition":
            try:
                models = get_user_models(hf_username, task)
                best_result, best_model_id = calculate_best_result(models, task)
                unit["best_result"] = best_result
                unit["best_model_id"] = best_model_id
                unit["passed_"] = best_result <= unit["baseline_metric"]
            except Exception:
                pass
        elif task == "text-to-speech":
            try:
                models = get_user_models(hf_username, task)
                if models:
                    unit["best_result"] = 0
                    unit["best_model_id"] = models[0]
                    unit["passed_"] = True
            except Exception:
                pass
        elif task == "demo":
            try:
                u7_file = hf_hub_download(USERNAMES_DATASET_ID, repo_type="dataset", filename="usernames.csv", token=HF_TOKEN)
                u7_users = pd.read_csv(u7_file)
                if hf_username in u7_users['username'].tolist():
                    unit["best_result"] = 0
                    unit["best_model_id"] = "Demo check passed"
                    unit["passed_"] = True
            except Exception:
                pass

        unit["passed"] = pass_emoji(unit["passed_"])

    df = pd.DataFrame(results_certification)
    df = df[['passed', 'unit', 'task', 'baseline_metric', 'best_result', 'best_model_id']]
    return df

# --------------------
# Gradio UI
# --------------------
with gr.Blocks() as demo:
    gr.Markdown("""
    # 🏆 Check your progress in the Audio Course 🏆
    - Pass 3 out of 4 assignments for a certificate.
    - Pass 4 out of 4 assignments for honors.
    - For Unit 7, first check your demo with the [Unit 7 assessment space](https://huggingface.co/spaces/huggingface-course/audio-course-u7-assessment).
    - Make sure your models are uploaded to Hub and public.
    """)
    
    hf_username_input = gr.Textbox(label="Your Hugging Face Username", placeholder="MariaK")
    check_button = gr.Button("Check my progress")
    output_table = gr.Dataframe()
    
    check_button.click(fn=certification, inputs=hf_username_input, outputs=output_table)

demo.launch()