ML Rust projects
The ML Rust organization runs this catalogue and maintains these projects. Their tiers follow the same published rules, and Featured also needs approval from a maintainer outside the organization.
All projects
Order: strongest tier first, then most downloads, then name, with unmeasured values last. A library counts downloads over 30 days. An app counts binary downloads per release. An app with none counts its crate downloads over 30 days.
What do the tiers mean?
- Featured
- Shown first. It passes every Featured gate.
- Listed
- Shown in the normal listing. No other tier applies.
- New
- A project whose first release is recent.
- Archived
- No longer maintained, or ended by its author.
Tiers come from published rules over raw signals. Stars alone never qualify a project.
Active dependents: crates from other owners that depend on this project in a current release. One owner counts at most once.
Movement needs more history.
| Project | Downloads | Dependents | Stars | Latest release | Runs on (GPU) | Native code | License |
|---|---|---|---|---|---|---|---|
| Listed 11 projects | |||||||
|
regexr Listed
A high-performance regex engine built from scratch with JIT compilation and SIMD acceleration |
8,809 downloads over 30 days | 4 | 28 | 0.6.2 | GPU support unknown | No native code found | Apache-2.0 OR MIT |
|
splintr Listed
Tokenizer for Rust and Python: pure Rust by default, with 18 bundled vocabularies and four backends covering BPE, SentencePiece and WordPiece. Its output matches Hugging Face tokenizers id for id on 14 model families, including GPT-2, Qwen, Whisper, T5 and BERT. Python bindings |
8,204 downloads over 30 days | 7 | 63 | 0.22.2 | GPU support unknown | Binds native code: pcre2-sys (feature pcre2) | MIT |
|
numr Listed
The base for numerical code in Rust, in the role NumPy has in Python: n-dimensional tensors, linear algebra, FFT, statistics and autograd on CPU, CUDA and WebGPU. The backends use native kernels, with no cuBLAS, MKL or other vendor math library. |
1,673 downloads over 30 days | 7 | 59 | 0.15.2 | CUDA, WebGPU | Binds native code: ring (feature distributed) | Apache-2.0 |
|
boostr Listed
ML framework on numr for training and inference on CPU, CUDA and WebGPU: attention, quantization, inference scheduling and distributed execution. Models include LLaMA, Mamba2, Qwen3.5, Whisper, Kokoro, VITS, VoxCPM2, Silero VAD, CLAP and super-resolution. Loads SafeTensors and GGUF. |
1,265 downloads over 30 days | 7 | 20 | 0.11.2 | CUDA, WebGPU | Binds native code: ring (feature distributed) | Apache-2.0 |
|
nexar Listed
Distributed runtime that replaces MPI between nodes: point-to-point messages, 11 collectives and RPC over QUIC, with no mpirun or MPI library. |
1,191 downloads over 30 days | 5 | 5 | 0.2.1 | CUDA | Binds native code: ring. Build uses a native toolchain: bindgen, pkg-config | Apache-2.0 |
|
solvr Listed
Scientific computing on numr, in the role SciPy has in Python, with a module map for SciPy users: optimization, ODE, DAE and PDE solvers, interpolation, statistics, signal processing, spatial algorithms and clustering. |
31 downloads over 30 days | 0 | 11 | 0.3.1 | CUDA, WebGPU | No native code found | Apache-2.0 |
|
Trains GPT-style language models from a YAML config and a tokenized dataset: dense GQA transformers, Mamba 1, 2 and 3, MLA and MoE layers on CPU or CUDA. Runs the training loop, optimizer, checkpoints, evaluation and metrics on boostr, and builds speech corpora through voxr. |
27 downloads over 30 days | Not applicable | 8 | 0.1.1 | CUDA | Binds native code: nvml-wrapper-sys (feature cuda) | Apache-2.0 |
|
treeboost Listed
Gradient-boosted decision trees for tabular data: trains a baseline model in one call, with full control over boosting modes and configs, and Scalar, AVX-512, SVE2, WGPU and CUDA backends. Python bindings |
18 downloads over 30 days | 0 | 9 | 0.1.2 | CUDA, WebGPU | No native code found | Apache-2.0 |
|
Reactive notebook environment for Rust |
10 binary downloads per release | Not applicable | 49 | 0.1.4 | GPU support unknown | No native code found | Apache-2.0 |
|
Inference server for LLMs, speech and embeddings: OpenAI-compatible routes with streaming, plus speech synthesis (TTS) and transcription (ASR). Serves Llama, Mistral, Qwen, Mamba2, MLA+MoE and hybrid models from SafeTensors, AWQ, GPTQ and GGUF. |
8 binary downloads per release | Not applicable | 7 | 0.7.1 | CUDA, WebGPU | Binds native code: ring (feature distributed) | Apache-2.0 |
|
An open protocol that gives AI agents a secure, sandboxed place to run code and use tools — with lean context usage. AXP works alongside MCP |
Not measured | Not applicable | 3 | Not measured | Not measured | Not measured | Apache-2.0 |
| New 5 projects | |||||||
|
voxr New
Audio and speech on boostr: compressed-audio decode, resampling, DSP and grapheme to phoneme, with speech synthesis (Kokoro, VITS), voice cloning (VoxCPM2), transcription with word timings (Whisper), voice activity (Silero) and audio-text scoring (CLAP), on CPU or CUDA. |
642 downloads over 30 days | 2 | 2 | 0.9.1 | CUDA | No native code found | Apache-2.0 |
|
containr New
Media containers for vidr and voxr: reads and writes MP4, MOV, Matroska and WebM, with integer timestamps and raw edit lists. |
206 downloads over 30 days | 2 | 0 | 0.1.1 | GPU support unknown | No native code found | Apache-2.0 |
|
rav1r New
AV1 encoder derived from rav1e, used by vidr. Its output stays bit-exact with upstream. |
56 downloads over 30 days | 1 | 0 | 0.1.1 | GPU support unknown | Binds native code: aom-sys (feature decode_test), libdav1d-sys (feature decode_test_dav1d). Build uses a native toolchain: cc (feature asm) | BSD-2-Clause |
|
GPU video library on numr and boostr: NVDEC, NVENC and Vulkan Video for H.264, HEVC, AV1 and VP9, CPU decoders, MP4 and MKV through containr, frames as GPU tensors, and vidr-remaster for restoration and upscaling, run on real clips. |
30 downloads over 30 days | Not applicable | 0 | 0.1.1 | CUDA, WebGPU, Vulkan | Binds native code: dav1d-sys (feature dav1d) | Apache-2.0 |
|
Model conversion and quantization with no Python: reads SafeTensors, GGUF, TCF and PyTorch state dicts, writes SafeTensors, GGUF and TCF, and quantizes to Q2_K through Q8_0 and IQ4 formats, including Q4_K_M mixes. GPTQ, AWQ, NF4, FP8, EXL2 and ONNX input are not implemented. Python bindings |
23 downloads over 30 days | Not applicable | 0 | 0.2.1 | GPU support unknown | No native code found | Apache-2.0 |