sheaf-lang 1.2.0

A functional language for differentiable programming

Sheaf is a functional language for differentiable computation. It is inspired by Clojure and is designed to operate within the Python/JAX machine learning ecosystem.

Sheaf has three main goals:

• Clojure Paradigm: Homoiconicity (code is data), immutability and minimalist syntax.
• Native XLA Performance: Sheaf code runs on GPUs/TPUs through JAX.
• Seamless Interop with Python: Compiled to pure JAX functions natively callable from Python.

Sample code:

(defn transformer-block [x layer-p config]
  (as-> x h
    (-> h   ;; 1. Self-Attention
        (layer-norm (get layer-p :ln1) 2)
        (multi-head-attention layer-p config)
        (first)  ;; Attention output
        (+ h))   ;; Residual 1

    (-> h   ;; 2. MLP
        (layer-norm (get layer-p :ln2) 2)
        (mlp (get layer-p :mlp))
        (+ h)))) ;; Residual 2

Python interoperability

from sheaf import Sheaf
shf = Sheaf("mlp.shf")

for x, y in zip(X, shf.forward(X, params)):
    print(f"  {x} -> {y[0]:.3f}")

• Website
• Reference documentation
• Quick Start guide