Issues

• VariationalDropout modules do not support CUDA device

  locuslab/torchdeq#13 · Filter by author lcdavis13 opened on Apr 5, 2026

• stop_mode='rel' seems to be ignored (solver iterates until 1e-5 despite 5% tolerance)

  locuslab/torchdeq#12 · Filter by author QuietNoviceCoder opened on Nov 17, 2025

• torch.compile failed due to modified backward pass

  locuslab/torchdeq#11 · Filter by author sablin39 opened on Nov 14, 2025

• jac_loss has no gradient for training.

  locuslab/torchdeq#10 · Filter by author QuietNoviceCoder opened on Oct 29, 2025
  3

• [New Feature] torchdeq implementation for *Reversible Deep Equilibrium Models*

  locuslab/torchdeq#9 · Filter by author sablin39 opened on Sep 27, 2025

• Returned fixed point solution sometimes has requires_grad=False

  locuslab/torchdeq#8 · Filter by author Boshi-Wang opened on Jun 27, 2024

• Fixed-point is not returned when indexing is set

  locuslab/torchdeq#7 · Filter by author BurgerAndreas opened on May 30, 2024

• Unexpected behaviour of indexing?

  locuslab/torchdeq#6 · Filter by author BurgerAndreas opened on May 25, 2024
  2

• Custom autograd fails with torchdeq in eval mode

  locuslab/torchdeq#4 · Filter by author BurgerAndreas opened on Mar 31, 2024
  1

• Implementation of torchdeq.utils.mixed_init different from original paper

  locuslab/torchdeq#3 · Filter by author jklim01 opened on Mar 15, 2024