tenferro

tenferro-rs is a dense tensor computation workspace in Rust, inspired by PyTorch and JAX. It provides:

  • Eager execution — Tensor, TypedTensor, EagerTensor, and EagerContext with CpuBackend for immediate computation and scalar-loss reverse-mode via backward()
  • Lazy traced execution — TracedTensor with graph optimization and transform-oriented automatic differentiation (grad, vjp, jvp, HVP)
  • Einsum — with automatic contraction-tree planning
  • Linear algebra — SVD, QR, Cholesky, eigh, solve

CPU execution is fully supported; GPU support is planned.

PyTorch, JAX, and tenferro

Topic PyTorch JAX tenferro
Eager tensor torch.tensor(data) jnp.array(data) Tensor::from_vec(shape, data)
Traced tensor — staged via jit TracedTensor::from_vec(shape, data)
Execution Eager by default Eager; jit stages when asked Eager (Tensor / EagerTensor) or lazy traced (TracedTensor + .eval())
Eager gradients loss.backward() — EagerTensor::backward() with accumulation
Transform AD torch.autograd.grad(...) jax.grad, jax.vjp, jax.jvp, hvp via composition loss.grad(&x), .vjp(), .jvp()
Einsum torch.einsum(...) jnp.einsum(...) einsum(...) / eager_einsum(...)
Device Device on tensors jax.device_put(...) Backend owned by Engine or CpuBackend