All pull requests

Pull requests

• fix(dbn): make UnsupervisedDBN accessors safe before the model is built

  #454·iloveeyjafjalla opened Oct 17, 2026·

  ·
  1

• fix(telemetry): make the optional kaiwu caller-context hook explicit

  #453·iloveeyjafjalla opened Oct 16, 2026·

  ·
  1

• docs(example): describe what scripts/eval.sh actually evaluates

  #452·iloveeyjafjalla opened Oct 16, 2026·

  ·
  1

• fix(docs): set CheckpointManager.save_dir in the CIM sampler snippets

  #451·iloveeyjafjalla opened Oct 16, 2026·

  ·
  1

• fix(example): strip the UTF-8 BOM preceding the eval_clustering.sh shebang

  #450·iloveeyjafjalla opened Oct 16, 2026·

  ·
  1

  1

• fix(qdiffusion): preserve proposal inputs for decode backward

  #449·Cornation opened Oct 15, 2026·

  ·

• perf(qdiffusion): skip unused token history in generation

  #448·Cornation opened Oct 15, 2026·

  ·
  1

  1

• fix(qvae_cell): retain singleton tail cells in BatchNorm training

  #447·messere1 opened Oct 15, 2026·

  ·
  1

• fix(qvae_cell): exclude missing annotations from supervised evaluation

  #446·messere1 opened Oct 15, 2026·

  ·
  1

• fix(qvae_cell): score energy without drawing unused posterior samples

  #445·messere1 opened Oct 15, 2026·

  ·
  1

• feat(example): make synthetic feature-selection datasets reproducible by seed

  #444·messere1 opened Oct 15, 2026·

  ·
  1

• perf(dplm): retain compact CPU embeddings after BOS/EOS pooling

  #443·messere1 opened Oct 15, 2026·

  ·
  1

• fix(rbm_digits): derive the plot_images layout from the data width

  #441·doufupiya opened Oct 15, 2026·

  ·
  1

• fix(qvae_cell): quote the output directory in train_v2.sh

  #439·doufupiya opened Oct 15, 2026·

  ·
  1

• Add molecular design reproduction with CIM, SQA and SA results

  #438·xxxxxys opened Oct 15, 2026·

  ·

• fix(typo): fix some typo errors

  #435·enkilee opened Oct 13, 2026·

  ·

• Add QDEQ: a reusable DEQ fixed-point plugin

  #434·Renata1027 opened Oct 13, 2026·

  ·

• fix(example): clean up BM sampling workers when training raises

  #433·balckwell opened Oct 12, 2026·

  ·

• fix(example): reject malformed FASTA records without dropping sequence data

  #432·balckwell opened Oct 12, 2026·

  ·

• fix(example): weight downstream MLP epoch losses by processed samples

  #431·balckwell opened Oct 12, 2026·

  ·

• fix(example): initialize and regularize the sequential MNIST decoder

  #430·balckwell opened Oct 12, 2026·

  ·

• fix(example): exclude missing annotations from clustering agreement scores

  #429·balckwell opened Oct 12, 2026·

  ·

• fix(example): reject empty DPLM epochs instead of reporting zero loss

  #428·balckwell opened Oct 12, 2026·

  ·

• fix(example): reject incomplete aligned generation quality inputs

  #427·balckwell opened Oct 12, 2026·

  ·

• fix(qvae): respect the unit-interval support of exponential smoothing

  #426·balckwell opened Oct 12, 2026·

  ·