All pull requests

Pull requests

• fix: Support zero-size shapes in variance scaling initializers

  #683·georgeguimaraes opened Oct 19, 2026Member·

  ·

• fix: Support zero-length sequences in RNN unrolls

  #682·georgeguimaraes opened Oct 19, 2026Member·

  ·

• Fix sparse categorical cross entropy for softmax outputs

  #680·dajiaohuang opened Sep 17, 2026·

  ·

• Keep metric accumulators and transforms during validation

  #679·georgeguimaraes opened Sep 17, 2026Member·

  ·

• Document how initializers read custom parameter shapes

  #678·georgeguimaraes opened Sep 17, 2026Member·

  ·

• Add Axon.stop_grad/2

  #677·georgeguimaraes opened Sep 17, 2026Member·

  ·

• Let a loss function take the batch inputs

  #676·georgeguimaraes opened Sep 17, 2026Member·

  ·

• feat: Add a learning rate range test

  #675·georgeguimaraes opened Sep 15, 2026Member·

  ·

• docs: Say that dense works on inputs of any rank

  #674·georgeguimaraes opened Sep 15, 2026Member·

  ·

• feat: Add quantile (pinball) loss

  #672·georgeguimaraes opened Sep 15, 2026Member·

  ·

• Build train steps from objective functions

  #669·seanmor5 opened Aug 23, 2026Contributor·

  ·
  1

• Initialize training step state in the types the step produces

  #668·seanmor5 opened Aug 23, 2026Contributor·

  ·
  1

• Add Axon.Pruning with magnitude-based unstructured pruning

  #666·seanmor5 opened Aug 23, 2026Contributor·

  ·
  1

  2

• Add parameter constraints

  #665·seanmor5 opened Aug 23, 2026Contributor·

  ·
  1

  5

• Add axis names to Axon.input and an Axon.rename layer

  #664·seanmor5 opened Aug 23, 2026Contributor·

  ·
  1

  1

• Merge training updates into a single Axon.ModelState.update call

  #663·seanmor5 opened Aug 23, 2026Contributor·

  ·
  1

• Optionally keep gradients in the training step state

  #661·seanmor5 opened Aug 23, 2026Contributor·

  ·
  1

  1

• Diagnose partial initial model states and document Axon.build

  #660·seanmor5 opened Aug 23, 2026Contributor·

  ·
  1

• Add training loop fusion

  #658·seanmor5 opened Aug 16, 2026Contributor·

  ·
  1

• Add Axon.Loop.stream/4 and implement run/4 on top of it

  #655·seanmor5 opened Aug 15, 2026Contributor·

  ·

• Fix quantization metadata loss and non-f32 weight quantization

  #654·seanmor5 opened Aug 15, 2026Contributor·

  ·
  2

• Add scaled dot-product attention

  #650·seanmor5 opened Aug 15, 2026Contributor·

  ·
  1

• Add Axon.capture

  #626·seanmor5 opened Jan 31, 2026Contributor·

  ·
  8