All pull requests

Pull requests

• Add optional dftd extra for dispersion corrections

  #1803·sixty-66 opened Nov 4, 2026·

  ·
  1

• Fix Matplotlib font-size side effect on calculator import

  #1802·sixty-66 opened Nov 4, 2026·

  ·
  1

• Compiled training: symbolic-shape force/stress compilation, DDP, opt-in cuEq fusion, l=4 enablement

  #1801·zhubonan opened Nov 3, 2026Contributor·

  ·

• Convert atomic references before subtracting node energies

  #1800·davrens101 opened Nov 3, 2026·

  ·

• Include pair repulsion in PolarMACE energy, forces and stress

  #1799·davrens101 opened Nov 2, 2026·

  ·
  1

• Fix learned-interaction omission in EnergyDipolesMACE node energies

  #1798·davrens101 opened Nov 2, 2026·

  ·

• Document Exponential Moving Average in training guide

  #1795·VagrantTrack149 opened Oct 28, 2026·

  ·

• [ARCH-2 3/3] Assemble MACEBackbone and its descriptor API

  reforge
  #1791·aacostadiaz opened Oct 28, 2026Collaborator·3/3·

  ·
  1

  4

• [ARCH-2 2/3] Build the interaction and product basis blocks on the backend ops

  reforge
  #1790·aacostadiaz opened Oct 28, 2026Collaborator·2/3·

  ·
  1

• [ARCH-2 1/3] Keep the convolution's paths apart and group features at the backend boundary

  reforge
  #1789·aacostadiaz opened Oct 28, 2026Collaborator·1/3·

  ·
  1

• Fix various plotting errors

  #1788·joehart2001 opened Oct 27, 2026Contributor·

  ·

• Add opt-in FP64 accumulation for ScaleShiftMACE total energies

  #1783·amphilagus opened Oct 25, 2026·

  ·

• Apply the config and property weights in every loss that skipped them

  #1781·alok-108 opened Oct 24, 2026·

  ·
  1

• [DATA-3 3/4] Collate graphs in torch without torch_geometric

  reforge
  #1780·aacostadiaz opened Oct 23, 2026Collaborator·3/4·

  ·
  1

  2

• [DATA-3 2/4] Describe the flat graph dict as data and collate it in numpy

  reforge
  #1779·aacostadiaz opened Oct 23, 2026Collaborator·2/4·

  ·
  1

  4

• [DATA-3 4/4] Pad graph batches to a budget with inert structures

  reforge
  #1777·aacostadiaz opened Oct 22, 2026Collaborator·4/4·

  ·
  1

• Apply per-config weights in WeightedHuberEnergyForcesStressLoss

  #1769·mayuriphad opened Oct 18, 2026·

  ·
  1

  2

• Take the stage-two learning rate from the command line when resuming

  #1766·alok-108 opened Oct 17, 2026·

  ·
  1

  1

• PolarMACE: opt-in regularised Fukui normalisation and charge-dipole cap

  #1762·lollcat opened Oct 15, 2026Contributor·

  ·

• Reproduce the released MACE-Polar parameterisation from the CLI

  #1761·lollcat opened Oct 15, 2026Contributor·

  ·

• Avoid retraining completed checkpoint epochs on restart

  #1753·kumar-miskin opened Oct 7, 2026·

  ·
  1

  1

• implement noise-resilient training (NRT)

  #1752·terry-berry067 opened Oct 6, 2026·

  ·

• Register multipolar MACELES readouts in optimizer param groups

  #1750·djkim-kr opened Oct 5, 2026Contributor·

  ·
  1

  3

• Multiple hidden layers for generic feature embedding.

  #1748·williamdavie opened Sep 30, 2026Contributor·

  ·
  1

• make it possible to turn off amsgrad

  #1743·ThomasWarford opened Sep 25, 2026Contributor·

  ·
  1