ElBruno.HuggingFace.Downloader 1.6.1

📥 ElBruno.HuggingFace.Downloader

A .NET library and CLI tool to download files (ONNX models, tokenizers, voice presets, etc.) from Hugging Face Hub repositories with progress reporting, caching, and authentication support.

Project website: elbruno.github.io/ElBruno.HuggingFace.Downloader. See the website guide for publishing and local preview instructions.

Features

• 📦 Download any file from public or private Hugging Face repositories
• 📊 Rich progress reporting with stages (Checking → Downloading → Validating → Complete)
• 🔑 HF_TOKEN authentication for gated/private repositories (env var or explicit)
• 🌐 Custom Hub endpoints via HF_ENDPOINT or downloader options
• 🔒 Atomic writes using temp files to avoid partial/corrupt downloads
• 🔁 Resumable downloads using HTTP range requests for interrupted large files
• 📌 Revision pinning with resolved commit metadata and optional expected commit enforcement
• 🗂️ Remote file listing — enumerate files in a repo (models, datasets, spaces) before downloading, no need to know exact paths in advance
• ✅ Required vs optional files — optional files fail silently
• 🧾 Manifest-based bundles with SHA-256 and size validation
• 📏 HEAD requests to resolve total download size before starting
• ⏭️ Skip existing files — only downloads what's missing
• 🖥️ Cross-platform cache directory helpers (Windows, Linux, macOS)
• 💉 DI-friendly with IServiceCollection extension methods
• 📝 ILogger integration for structured logging

📦 NuGet Packages

Package	Version	Downloads	Description
ElBruno.HuggingFace.Downloader			Core library for downloading files from Hugging Face Hub repositories
ElBruno.HuggingFace.Downloader.Cli			CLI tool (hfdownload) for managing Hugging Face downloads

Installation

Library (NuGet)

dotnet add package ElBruno.HuggingFace.Downloader

.NET Framework

The package also targets netstandard2.0, so it works on .NET Framework 4.6.2 and later. DownloadRequest and the other request types use required and init properties, which need C# 11 or later (the samples below also use collection expressions, so latest is the simplest choice). Passing options or an HttpClient to the HuggingFaceDownloader constructor also needs a reference to System.Net.Http:

<PropertyGroup>
  <LangVersion>latest</LangVersion>
</PropertyGroup>

<ItemGroup>
  <Reference Include="System.Net.Http" />
</ItemGroup>

CLI Tool

dotnet tool install -g ElBruno.HuggingFace.Downloader.Cli

Once installed, use the hfdownload command:

# Download model files
hfdownload download sentence-transformers/all-MiniLM-L6-v2 onnx/model.onnx tokenizer.json

# Download a specific tag into a revision-aware cache directory
hfdownload download sentence-transformers/all-MiniLM-L6-v2 onnx/model.onnx --revision v1.0

# Restart from scratch instead of resuming a preserved partial download
hfdownload download sentence-transformers/all-MiniLM-L6-v2 onnx/model.onnx --no-resume

# Ensure a manifest-defined bundle with integrity validation
hfdownload download --manifest ./phi4-bundle.json -o ./models/phi4

# Check if files exist locally
hfdownload check sentence-transformers/all-MiniLM-L6-v2 onnx/model.onnx tokenizer.json

# List remote files in a repo without knowing filenames up front
hfdownload files sentence-transformers/all-MiniLM-L6-v2

# List remote files in a dataset repo, as JSON
hfdownload files my-org/my-dataset --repo-type dataset --format json

# List cached models
hfdownload list

# Delete a cached model
hfdownload delete sentence-transformers/all-MiniLM-L6-v2

# See all commands
hfdownload --help

See the full CLI Reference for all commands and options.

Quick Start (Library)

1) Download model files

using ElBruno.HuggingFace;

using var downloader = new HuggingFaceDownloader();

var request = new DownloadRequest
{
    RepoId = "sentence-transformers/all-MiniLM-L6-v2",
    LocalDirectory = "./models/miniLM",
    RequiredFiles = ["onnx/model.onnx", "tokenizer.json"],
    OptionalFiles = ["tokenizer_config.json", "vocab.txt"],
    Revision = "refs/pr/42"
};

await downloader.DownloadFilesAsync(request);

Console.WriteLine($"Resolved commit: {request.ResolvedCommitSha}");

2) Ensure a manifest-based bundle with SHA-256 validation

var manifest = new ModelBundleManifest
{
    RepoId = "microsoft/Phi-4-mini-instruct-onnx",
    Revision = "main",
    Files =
    [
        new ModelBundleFile
        {
            Path = "onnx/model.onnx",
            Size = 123456789,
            Sha256 = "0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef"
        },
        new ModelBundleFile
        {
            Path = "tokenizer.json",
            Required = false
        }
    ]
};

var bundle = await downloader.EnsureBundleAsync(manifest, "./models/phi4");
Console.WriteLine(bundle.ResolvedManifestPath);

Manifest-driven bundle downloads validate existing files before reuse, re-download corrupted files, and write a resolved manifest to ./models/phi4/.hf.bundle.resolved.json.

3) Pin a branch or tag to the expected commit

var request = new DownloadRequest
{
    RepoId = "my-org/my-model",
    LocalDirectory = "./models/pinned",
    RequiredFiles = ["model.onnx"],
    Revision = "main",
    ExpectedCommitSha = "1234567890abcdef1234567890abcdef12345678"
};

await downloader.DownloadFilesAsync(request);
Console.WriteLine(request.ResolvedCommitSha);

When ExpectedCommitSha is set, the download fails if the resolved commit is missing or differs from the expected SHA. Downloads use the verified immutable commit URL so a moving branch cannot change between verification and transfer. If a custom endpoint or redirect hides revision headers, set Revision to the immutable commit SHA.

4) Check if files are already downloaded

bool ready = downloader.AreFilesAvailable(
    ["onnx/model.onnx", "tokenizer.json"],
    "./models/miniLM");

if (!ready)
{
    var missing = downloader.GetMissingFiles(
        ["onnx/model.onnx", "tokenizer.json"],
        "./models/miniLM");
    Console.WriteLine($"Missing {missing.Count} files");
}

5) Discover files in a repo before downloading

IReadOnlyList<RepoTreeEntry> entries = await downloader.ListRepoFilesAsync(
    "sentence-transformers/all-MiniLM-L6-v2",
    RepoType.Model);

foreach (var entry in entries.Where(e => !e.IsDirectory))
    Console.WriteLine($"{entry.Path} ({entry.SizeBytes} bytes)");

// Use the discovered paths directly as a download request
await downloader.DownloadFilesAsync(new DownloadRequest
{
    RepoId = "sentence-transformers/all-MiniLM-L6-v2",
    LocalDirectory = "./models/miniLM",
    RequiredFiles = [.. entries.Where(e => !e.IsDirectory).Select(e => e.Path)]
});

6) Track download progress

var progress = new Progress<DownloadProgress>(p =>
{
    if (p.Stage == DownloadStage.Downloading)
        Console.Write($"\r⬇️ [{p.CurrentFile}] {p.PercentComplete:F0}% (reused {ByteFormatHelper.FormatBytes(p.ResumedBytes)})");
    else
        Console.WriteLine($"{p.Stage}: {p.Message}");
});

await downloader.DownloadFilesAsync(new DownloadRequest
{
    RepoId = "sentence-transformers/all-MiniLM-L6-v2",
    LocalDirectory = "./models/miniLM",
    RequiredFiles = ["onnx/model.onnx", "tokenizer.json"],
    Progress = progress
});

7) Authentication (Private/Gated Repos)

Set the HF_TOKEN environment variable, or pass it explicitly:

var downloader = new HuggingFaceDownloader(new HuggingFaceDownloaderOptions
{
    AuthToken = "hf_your_token_here"
});

8) Use a private Hub endpoint

Set the HF_ENDPOINT environment variable, or configure the endpoint explicitly:

var downloader = new HuggingFaceDownloader(new HuggingFaceDownloaderOptions
{
    Endpoint = "https://huggingface.internal"
});

The explicit Endpoint takes precedence over HF_ENDPOINT; if neither is set, the downloader uses https://huggingface.co.

9) Dependency Injection

builder.Services.AddHuggingFaceDownloader(options =>
{
    options.Timeout = TimeSpan.FromMinutes(60);
});

// Then inject HuggingFaceDownloader in your services
public class MyModelService(HuggingFaceDownloader downloader)
{
    public async Task EnsureModelAsync()
    {
        await downloader.DownloadFilesAsync(new DownloadRequest
        {
            RepoId = "my-org/my-model",
            LocalDirectory = DefaultPathHelper.GetDefaultCacheDirectory("MyApp"),
            RequiredFiles = ["model.onnx", "tokenizer.json"]
        });
    }
}

9) Manage local cache programmatically

var cacheRoot = DefaultPathHelper.GetDefaultCacheDirectory("hfdownload");
var repoId = "sentence-transformers/all-MiniLM-L6-v2";

// List all cached repositories
var cachedRepos = downloader.ListCachedRepos(cacheRoot);

// Check whether required files are cached for one repo
bool isCached = downloader.IsCached(
    repoId,
    cacheRoot,
    ["onnx/model.onnx", "tokenizer.json"]);

// Get total cache size for one local repo directory
long size = downloader.GetCachedSize(
    Path.Combine(cacheRoot, DefaultPathHelper.SanitizeModelName(repoId)));

// Delete cached files for the repo (safe no-op if not found)
await downloader.DeleteCachedFilesAsync(repoId, cacheRoot);

Direct downloads also write ./models/.../.hf.download.resolved.json so callers can inspect the requested revision and resolved commit later.

Documentation

Related Projects

• ElBruno.PersonaPlex — Integrates this downloader to auto-download ONNX models for NVIDIA's PersonaPlex-7B-v1 full-duplex speech-to-speech model

Building from Source

git clone https://github.com/elbruno/ElBruno.HuggingFace.Downloader.git
cd ElBruno.HuggingFace.Downloader
dotnet build ElBruno.HuggingFace.Downloader.slnx
dotnet test ElBruno.HuggingFace.Downloader.slnx

Requirements

• .NET 8.0 SDK or later

🤝 Contributing

Contributions are welcome! Please:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

📄 License

This project is licensed under the MIT License — see the LICENSE file for details.

👋 About the Author

Hi! I'm ElBruno 🧡, a passionate developer and content creator exploring AI, .NET, and modern development practices.

Made with ❤️ by ElBruno

If you like this project, consider following my work across platforms:

• 📻 Podcast: No Tienen Nombre — Spanish-language episodes on AI, development, and tech culture
• 💻 Blog: ElBruno.com — Deep dives on embeddings, RAG, .NET, and local AI
• 📺 YouTube: youtube.com/elbruno — Demos, tutorials, and live coding
• 🔗 LinkedIn: @elbruno — Professional updates and insights
• 𝕏 X: @elbruno — Quick tips, releases, and tech news