hub-stats.py

# /// script
# requires-python = ">=3.10"
# dependencies = [
#     "pandas>=2.2.2",
#     "aiohttp",
#     "python-dotenv>=1.0.1",
#     "huggingface-hub>=0.24.3",
#     "tenacity>=9.0.0",
#     "pyarrow>=17.0.0",
#     "requests",
# ]
# ///

import json
import os
import asyncio
import time

import pandas as pd
import aiohttp
import requests.utils
from dotenv import load_dotenv
from huggingface_hub import HfApi
from tenacity import retry, stop_after_attempt, wait_exponential
import pyarrow as pa
import pyarrow.parquet as pq

load_dotenv()

CACHE_DIR = ".hf_cache"
os.makedirs(CACHE_DIR, exist_ok=True)

# Get token from environment (works in HF Jobs)
HF_TOKEN = os.environ.get("HF_TOKEN")
api = HfApi(token=HF_TOKEN)
USER_ID = api.whoami()["name"]
REPO_ID = f"{USER_ID}/hub-stats"

print(f"🚀 Hugging Face Hub Stats Collector")
print(f"📊 Dataset will be uploaded to: {REPO_ID}")
print(f"👤 User: {USER_ID}")
print("-" * 50)

ENDPOINT_CONFIGS = {
    "models": {
        "limit": 1000,
        "params": {
            "full": "true",
            "config": "true",
            "expand[]": [
                "gguf",
                "downloadsAllTime",
                "transformersInfo",
                "cardData",
                "safetensors",
                "baseModels",
                "author",
                "likes",
                "inferenceProviderMapping",
                "downloads",
                "siblings",
                "tags",
                "pipeline_tag",
                "lastModified",
                "createdAt",
                "config",
                "library_name",
            ],
        },
    },
    "datasets": {
        "limit": 1000,
        "params": {
            "full": "true",
            "expand[]": [
                "author",
                "cardData",
                "citation",
                "createdAt",
                "disabled",
                "description",
                "downloads",
                "downloadsAllTime",
                "gated",
                "lastModified",
                "likes",
                "paperswithcode_id",
                "private",
                "siblings",
                "sha",
                "tags",
                "trendingScore",
            ],
        },
    },
    "spaces": {"limit": 1000, "params": {"full": "true"}},
    "posts": {"limit": 50, "params": {"skip": 0}},
    "daily_papers": {
        "limit": 50,
        "params": {},
        "base_url": "https://huggingface.co/api/daily_papers",
    },
    "arxiv_papers": {
        "limit": 100,
        "params": {},
        "base_url": "https://huggingface.co/api/papers",
    },
}


def parse_link_header(link_header):
    if not link_header:
        return None
    links = requests.utils.parse_header_links(link_header)
    for link in links:
        if link.get("rel") == "next":
            return link.get("url")
    return None


def to_json_string(x):
    if isinstance(x, (dict, list)):
        return json.dumps(x)
    if x is None or pd.isna(x):
        return None
    return str(x)


def stringify_nested_columns(df):
    for col in df.columns:
        if df[col].map(lambda value: isinstance(value, (dict, list))).any():
            df[col] = df[col].apply(to_json_string)
    return df


def process_dataframe(df, endpoint):
    if len(df) == 0:
        return df

    if endpoint == "posts":
        if "author" in df.columns:
            author_df = pd.json_normalize(df["author"])
            author_cols = ["avatarUrl", "followerCount", "fullname", "name"]
            for col in author_cols:
                if col in author_df.columns:
                    df[col] = author_df[col]
            df = df.drop("author", axis=1)

        for ts_col in ["publishedAt", "updatedAt"]:
            if ts_col in df.columns:
                df[ts_col] = pd.to_datetime(df[ts_col]).dt.tz_localize(None)

    elif endpoint == "daily_papers":
        if "paper" in df.columns:
            paper_df = pd.json_normalize(df["paper"], errors="ignore").add_prefix(
                "paper_"
            )
            df = pd.concat([df.drop("paper", axis=1), paper_df], axis=1)

        for ts_col in ["publishedAt", "paper_publishedAt"]:
            if ts_col in df.columns:
                df[ts_col] = pd.to_datetime(df[ts_col], errors="coerce").dt.tz_localize(
                    None
                )

    elif endpoint == "arxiv_papers":
        for ts_col in ["publishedAt", "submittedOnDailyAt"]:
            if ts_col in df.columns:
                df[ts_col] = pd.to_datetime(df[ts_col], errors="coerce").dt.tz_localize(
                    None
                )

    else:
        for field in ["createdAt", "lastModified"]:
            if field in df.columns:
                df[field] = pd.to_datetime(df[field], errors="coerce").dt.tz_localize(
                    None
                )

    if "gated" in df.columns:
        df["gated"] = df["gated"].astype(str)

    for col in ["cardData", "config", "gguf"]:
        if col in df.columns:
            df[col] = df[col].apply(to_json_string)

    if endpoint == "arxiv_papers":
        df = stringify_nested_columns(df)

    return df


def save_parquet(df, output_file):
    df.to_parquet(output_file, index=False, engine="pyarrow")


@retry(stop=stop_after_attempt(5), wait=wait_exponential(multiplier=1, min=4, max=60))
async def fetch_data_page(session, url, params=None, headers=None):
    async with session.get(url, params=params, headers=headers) as response:
        response.raise_for_status()
        return await response.json(), response.headers.get("Link")


def jsonl_to_parquet(endpoint, jsonl_file, output_file):
    if not os.path.exists(jsonl_file):
        print(f"✗ {jsonl_file} not found")
        return 0

    # Collect all dataframes first to get unified schema
    all_dfs = []
    with open(jsonl_file, "r") as f:
        for line in f:
            line = line.strip()
            if not line:
                continue
            data = json.loads(line)
            if endpoint == "posts":
                items = data.get("socialPosts", [])
            else:
                items = data

            if not items:
                continue

            df = pd.DataFrame(items)
            if df.empty:
                continue

            df = process_dataframe(df, endpoint)
            all_dfs.append(df)

    if not all_dfs:
        print(f"  No data found for {endpoint}")
        return 0

    # Concatenate all dataframes - pandas will unify types
    combined_df = pd.concat(all_dfs, ignore_index=True)
    total_rows = len(combined_df)

    # Write to parquet with controlled row group sizes
    row_group_size = 50_000
    table = pa.Table.from_pandas(combined_df, preserve_index=False)
    writer = pq.ParquetWriter(output_file, table.schema)
    for i in range(0, total_rows, row_group_size):
        chunk = table.slice(i, min(row_group_size, total_rows - i))
        writer.write_table(chunk)
    writer.close()

    return total_rows


async def create_parquet_files(skip_upload=False):
    start_time = time.time()
    endpoints = [
        "daily_papers",
        "arxiv_papers",
        "models",
        "spaces",
        "datasets",
        "posts",
    ]
    created_files = []
    jsonl_files = {}

    async with aiohttp.ClientSession() as session:
        for endpoint in endpoints:
            print(f"Fetching {endpoint}...")

            config = ENDPOINT_CONFIGS[endpoint]
            base_url = config.get("base_url", f"https://huggingface.co/api/{endpoint}")
            params = {"limit": config["limit"]}
            params.update(config["params"])

            headers = {"Accept": "application/json"}
            url = base_url
            page = 0

            jsonl_file = os.path.join(CACHE_DIR, f"{endpoint}_raw.jsonl")
            with open(jsonl_file, "w") as f:
                pass  # truncate

            while url:
                if endpoint == "posts":
                    params["skip"] = page * params["limit"]

                try:
                    data, link_header = await fetch_data_page(
                        session, url, params, headers
                    )

                    with open(jsonl_file, "a") as f:
                        f.write(json.dumps(data) + "\n")

                    if endpoint == "posts":
                        total_items = data.get("numTotalItems", 0)
                        items_on_page = len(data.get("socialPosts", []))
                        if (page + 1) * params[
                            "limit"
                        ] >= total_items or items_on_page == 0:
                            url = None
                        else:
                            url = base_url
                    else:
                        url = parse_link_header(link_header)
                        if url:
                            params = {}

                    page += 1

                except Exception as e:
                    print(f"Error on page {page} for {endpoint}: {e}")
                    await asyncio.sleep(2)
                    if page > 0:
                        url = None
                    else:
                        raise

            print(f"  Raw data for {endpoint} saved to {jsonl_file}")
            jsonl_files[endpoint] = jsonl_file

    # Convert JSONL -> Parquet with streaming writer
    for endpoint in endpoints:
        jsonl_file = jsonl_files.get(endpoint)
        if not jsonl_file or not os.path.exists(jsonl_file):
            continue

        print(f"Processing {endpoint} from JSONL...")
        output_file = os.path.join(CACHE_DIR, f"{endpoint}.parquet")
        total_rows = jsonl_to_parquet(endpoint, jsonl_file, output_file)
        print(f"✓ {endpoint}: {total_rows:,} rows -> {output_file}")
        created_files.append(output_file)

        if not skip_upload:
            upload_to_hub(output_file, REPO_ID)

    elapsed = time.time() - start_time
    return created_files, elapsed


def recreate_from_jsonl():
    endpoints = [
        "daily_papers",
        "arxiv_papers",
        "models",
        "spaces",
        "datasets",
        "posts",
    ]

    for endpoint in endpoints:
        jsonl_file = os.path.join(CACHE_DIR, f"{endpoint}_raw.jsonl")
        if not os.path.exists(jsonl_file):
            print(f"✗ {jsonl_file} not found")
            continue

        print(f"Recreating {endpoint} from {jsonl_file}...")
        output_file = os.path.join(CACHE_DIR, f"{endpoint}.parquet")
        total_rows = jsonl_to_parquet(endpoint, jsonl_file, output_file)
        print(f"✓ {endpoint}: {total_rows:,} rows -> {output_file}")


def upload_to_hub(file_path, repo_id):
    try:
        api.upload_file(
            path_or_fileobj=file_path,
            path_in_repo=os.path.basename(file_path),
            repo_id=repo_id,
            repo_type="dataset",
        )
        print(f"✓ Uploaded {os.path.basename(file_path)} to {repo_id}")
        return True
    except Exception as e:
        print(f"✗ Failed to upload {os.path.basename(file_path)}: {e}")
        return False


def main(skip_upload=False):
    created_files, elapsed = asyncio.run(create_parquet_files(skip_upload=skip_upload))

    print(f"\nCompleted in {elapsed:.2f} seconds")
    print(f"Created {len(created_files)} parquet files:")

    for file in created_files:
        size = os.path.getsize(file)
        pf = pq.ParquetFile(file)
        rows = pf.metadata.num_rows
        print(f"  {os.path.basename(file)}: {rows:,} rows, {size:,} bytes")

    if skip_upload:
        print(f"\nRaw JSONL files saved to {CACHE_DIR}/ for recreation")
        print("Use 'python app.py --recreate' to recreate parquet files from JSONL")


if __name__ == "__main__":
    import sys

    if "--recreate" in sys.argv:
        recreate_from_jsonl()
    else:
        skip_upload = "--skip-upload" in sys.argv
        main(skip_upload=skip_upload)