Can I define my own differentiation rules?

Absolutely. The skill covers the EnzymeRules system, allowing you to define custom forward and reverse passes for functions where standard AD might be insufficient or inefficient.

What is Enzyme.jl and how does this skill help?

Enzyme.jl is an automatic differentiation tool that works at the LLVM level. This skill provides the patterns and best practices needed to implement high-performance gradients in Julia code within the Claude environment.

What is the difference between Active and Duplicated annotations?

Active is typically used for scalar values in reverse mode, while Duplicated is used for mutable objects like arrays that require a shadow gradient buffer to store derivative values.

Does this skill support GPU acceleration?

Yes, it includes specific patterns for differentiating CUDA kernels using the autodiff_deferred function and GPUCompiler integration for high-performance GPU tasks.

Enzyme.jl Automatic Differentiation

Name: Enzyme.jl Automatic Differentiation
Author: plurigrid

byplurigrid

•

Data Science & ML

Implements high-performance LLVM-level automatic differentiation for Julia code on both CPU and GPU architectures.

Enzyme-autodiff is a specialized Claude Code skill designed for high-performance scientific computing and machine learning in Julia. By leveraging Enzyme's ability to perform automatic differentiation (AD) at the LLVM IR level, it generates highly optimized gradient code with minimal overhead. The skill provides comprehensive guidance on reverse and forward modes, complex type annotations like Duplicated and Active, and deep integration with GPU kernels via CUDA.jl, making it an essential tool for developers building differentiable simulations or custom neural network architectures.

Key Features

01LLVM-level IR transformation for peak performance

02Advanced memory management with Duplicated and BatchDuplicated types

038 GitHub stars

04Direct GPU kernel differentiation via CUDA.jl integration

05Extensible rule system with EnzymeRules and ChainRules support

06Support for Forward and Reverse differentiation modes

Use Cases

01Computing Jacobians and Hessians for large-scale optimization problems

02Training high-performance machine learning models using custom Julia kernels

03Optimizing complex scientific simulations with differentiable physics

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add plurigrid/asi enzyme-autodiff

For use in Claude.ai and ChatGPT

Download Skill