raven

modern scientific computing for OCaml

nx • ᚱ rune • ᚲ kaun • ᚨ brot • ᛃ talon

ᛞ hugin • ᛈ quill • ᚠ fehu • ᛋ sowilo

raven

Raven is an ecosystem of composable libraries for numerical computing in OCaml. Tensors, automatic differentiation, neural networks, dataframes, plotting, tokenization, computer vision, reinforcement learning, and interactive notebooks — each library does one thing well, and they compose cleanly together.

Built on OCaml 5's effect system, Raven uses function transformations — grad, vmap, jit — that compose freely because they are nested effect handlers. The same code that runs a training loop on your laptop can target Metal or CUDA. Types catch shape and dtype mismatches at compile time.

see it in action

Tokenize text, build a classifier, and train it — three libraries working together:

open Kaun

(* Tokenize text with Brot *)
let tokenizer = Brot.from_file "tokenizer.json" |> Result.get_ok
let encode text = Brot.encode_ids tokenizer text

(* Build a model with Kaun *)
let model = Layer.sequential [
  Layer.embedding ~vocab_size:30522 ~embed_dim:128 ();
  Layer.relu ();
  Layer.linear ~in_features:128 ~out_features:2 ();
]

(* Train with automatic differentiation — Rune under the hood *)
let trainer = Train.make ~model
  ~optimizer:(Optim.adam ~lr:(Optim.Schedule.constant 1e-3) ())
let st = Train.init trainer ~dtype:Nx.Float32
let st = Train.fit trainer st train_data

the ecosystem

foundation

nx N-dimensional arrays with pluggable backends (NumPy)

ᚱ rune Automatic differentiation and vectorizing maps (JAX)

machine learning

ᚲ kaun Neural networks and training (Flax / PyTorch)

ᚨ brot Tokenization for language models (HuggingFace Tokenizers)

ᛋ sowilo Differentiable computer vision (OpenCV)

ᚠ fehu Reinforcement learning environments (Gymnasium)

data and visualization

ᛃ talon Dataframes with type-safe columns (pandas / Polars)

ᛞ hugin Publication-quality plotting (Matplotlib)

tools

ᛈ quill Interactive notebooks as markdown files (Jupyter)

kaun-board Training dashboard and metrics logger

get involved

Raven is built in public. We need your help:

Try it out — install the libraries, run the examples, tell us what breaks
Report issues — found a bug? Missing a feature? Let us know
Join the conversation — API decisions happen in GitHub discussions
Contribute — check out good first issues

View on GitHub →

support the project

Building a scientific computing ecosystem takes time and focus. Your sponsorship helps us deliver on our roadmap — GPU backends, performance parity, and comprehensive documentation.

Support Raven →