""" Community fork of MariaK/Check-my-progress-Audio-Course. Fixes the upstream RUNTIME_ERROR: the original calls certification() eagerly at module load AND the Unit 7 ("demo") branch downloads usernames.csv from the `huggingface-course/audio-course-u7-hands-on` dataset, which has been deleted (404) — with no try/except, so the Space crashed on startup. Changes (behaviour-preserving for Units 4/5/6): 1. Do NOT run certification() at import time (the table starts empty). 2. Guard the Unit 7 download so a deleted/gated dataset no longer crashes the app; Unit 7 is reported as "auto-check unavailable" instead. 3. Handle an empty/blank username gracefully. Units 4/5/6 are verified exactly as upstream (against your models on the Hub). """ import os import re import gradio as gr import pandas as pd import requests from huggingface_hub import HfApi, ModelCard, hf_hub_download from huggingface_hub.repocard import metadata_load def pass_emoji(passed): return "✅" if passed is True 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): models = api.list_models(author=hf_username, filter=[task]) user_model_ids = [x.modelId for x in models] match task: case "audio-classification": dataset = "marsyas/gtzan" case "automatic-speech-recognition": dataset = "PolyAI/minds14" case "text-to-speech": dataset = "" case _: print("Unsupported task") dataset = "" if dataset == "": return user_model_ids dataset_specific_models = [] for model in user_model_ids: meta = get_metadata(model) if meta is None: continue try: if meta["datasets"] == [dataset]: dataset_specific_models.append(model) except Exception: continue return dataset_specific_models def calculate_best_result(user_models, task): best_model = "" if task == "audio-classification": best_result = -100 larger_is_better = True elif task == "automatic-speech-recognition": best_result = 100 larger_is_better = False for model in user_models: meta = get_metadata(model) if meta is None: continue metric = parse_metrics(model, task) if metric is None: continue if larger_is_better: if metric > best_result: best_result = metric best_model = meta["model-index"][0]["name"] else: if metric < best_result: best_result = metric best_model = meta["model-index"][0]["name"] return best_result, best_model def get_metadata(model_id): try: readme_path = hf_hub_download(model_id, filename="README.md") return metadata_load(readme_path) except requests.exceptions.HTTPError: return None def extract_metric(model_card_content, task): 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): card = ModelCard.load(model) return extract_metric(card.content, task) def certification(hf_username): hf_username = (hf_username or "").strip() 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: unit["passed"] = pass_emoji(unit["passed_"]) if not hf_username: continue match unit["task"]: case "audio-classification": try: m = get_user_models(hf_username, task="audio-classification") best_result, best_model_id = calculate_best_result( m, task="audio-classification" ) unit["best_result"] = best_result unit["best_model_id"] = best_model_id if unit["best_result"] >= unit["baseline_metric"]: unit["passed_"] = True unit["passed"] = pass_emoji(unit["passed_"]) except Exception: print("No relevant models / metrics for audio classification") case "automatic-speech-recognition": try: m = get_user_models( hf_username, task="automatic-speech-recognition" ) best_result, best_model_id = calculate_best_result( m, task="automatic-speech-recognition" ) unit["best_result"] = best_result unit["best_model_id"] = best_model_id if unit["best_result"] <= unit["baseline_metric"]: unit["passed_"] = True unit["passed"] = pass_emoji(unit["passed_"]) except Exception: print("No relevant models / metrics for ASR") case "text-to-speech": try: m = get_user_models(hf_username, task="text-to-speech") if m: unit["best_result"] = 0 unit["best_model_id"] = m[0] unit["passed_"] = True unit["passed"] = pass_emoji(unit["passed_"]) except Exception: print("No relevant models for TTS") case "demo": # Guarded: the upstream usernames dataset was deleted (404). try: path = hf_hub_download( USERNAMES_DATASET_ID, repo_type="dataset", filename="usernames.csv", token=HF_TOKEN, ) users = pd.read_csv(path) if hf_username in users["username"].tolist(): unit["best_result"] = 0 unit["best_model_id"] = "Demo check passed" unit["passed_"] = True unit["passed"] = pass_emoji(unit["passed_"]) except Exception: unit["best_model_id"] = ( "Unit 7 auto-check unavailable — upstream usernames dataset " "deleted; verify your public demo via the Unit 7 assessment space" ) case _: print("Unknown task") df = pd.DataFrame(results_certification) return df[ ["passed", "unit", "task", "baseline_metric", "best_result", "best_model_id"] ] with gr.Blocks() as demo: gr.Markdown( """ # 🏆 Check your progress in the Audio Course (community fork) 🏆 > Fork of `MariaK/Check-my-progress-Audio-Course` that fixes the upstream > startup crash (the Unit 7 check downloaded a now-deleted dataset). Units > 4/5/6 are verified exactly as in the original, against your models on the Hub. - Certificate of completion: **pass 3 of 4** assignments. - Honors certificate: **pass 4 of 4**. Your trained-model metric must be equal to or better than the baseline. Unit 7's automatic check is unavailable upstream (deleted dataset); use the [Unit 7 assessment space](https://huggingface.co/spaces/huggingface-course/audio-course-u7-assessment) (or a working fork) to confirm your public demo. Enter your Hugging Face username to check your progress: """ ) hf_username = gr.Textbox( placeholder="VoicesColeby", label="Your Hugging Face Username" ) check_progress_button = gr.Button(value="Check my progress") output = gr.components.Dataframe(value=None) check_progress_button.click(fn=certification, inputs=hf_username, outputs=output) demo.launch()