pytorch/functorch

Health v2 · maintenance only

Flags: archived

Adoption not part of the score

1434 stars · 108 forks observed · 2026-08-28

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

functorch is a library providing JAX-like composable function transforms (vmap, grad, vjp, jvp) for PyTorch, enabling efficient per-sample gradients, ensembles, and Jacobian/Hessian computation. Its APIs have been integrated into PyTorch 2.0 as torch.func, with the standalone package deprecated.

Use cases

• compute per-sample gradients in pytorch
• run model ensembles efficiently on one machine
• compute jacobians and hessians of neural networks
• batch inner-loop tasks in MAML meta-learning
• vectorize functions over batches with vmap
• compose grad and vmap transforms like JAX

When to choose

• you need composable vmap/grad transforms on PyTorch versions before 2.0
• you want JAX-style function transforms in an existing PyTorch codebase
• you need efficient per-sample gradients or batched Jacobians/Hessians

When to avoid

• you are on PyTorch 2.0 or newer - use torch.func instead
• you want a standalone actively developed library - this is deprecated and merged into PyTorch
• you need transforms for frameworks other than PyTorch

Facets

library · maturity maintenance

machine-learning deep-learning math machine-learning deep-learning python cross-platform pytorch jax autodiff vmap gradients function-transforms jacobians hessians algorithms gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="pytorch/functorch")