patrick-kidger/diffrax

Health v2 · maintenance only

Adoption not part of the score

2089 stars · 186 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Diffrax is a JAX-based library providing numerical differential equation solvers for ODEs, SDEs, and CDEs. It is fully autodifferentiable and GPU-capable, with support for neural differential equations and multiple adjoint methods for backpropagation.

Use cases

• solve ODEs in JAX with GPU acceleration
• train neural differential equations with backpropagation
• simulate stochastic differential equations differentiably
• integrate controlled differential equations for time series models
• run differentiable physics or dynamical systems simulations
• vmapped batch integration of differential equations

When to choose

• you need differentiable, GPU-capable ODE/SDE/CDE solvers within the JAX ecosystem
• you are training neural differential equations or scientific ML models
• you want many solver choices (Tsit5, Dopri8, symplectic, implicit) in one unified library

When to avoid

• you work outside the JAX/Python ecosystem
• you need a general-purpose non-differentiable solver without GPU requirements
• you need symbolic or exact analytical solutions rather than numerical integration

Facets

library · maturity active

machine-learning simulation math gpu-computing machine-learning deep-learning data-science python cross-platform jax differential-equations ode sde cde neural-differential-equations numerical-solvers autodiff scientific-computing dynamical-systems algorithms gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="patrick-kidger/diffrax")