Pull requests

• Say what else exists, and what we reuse instead of rebuilding

  #46·ringger opened Sep 20, 2026·

  ·

• Add torchlingo.diagnostics: checks that name what is wrong

  #45·ringger opened Sep 20, 2026·

  ·

• Teach the failure modes by breaking a model, not by listing them

  #44·ringger opened Sep 20, 2026·

  ·

• Surface attention for the translation the model generated

  #43·ringger opened Sep 20, 2026·

  ·

• PR and task conventions: stop stacking, and name them unambiguously

  #42·ringger opened Sep 20, 2026·

  ·

• Make the training script say why it stopped, and the comparison say what varied

  #38·ringger opened Sep 19, 2026·

  ·

• Stop val_losses from interleaving two different measurements

  #37·ringger opened Sep 18, 2026·

  ·

• Retrain the tutorial checkpoint for 36 epochs on the recovered corpus

  #34·ringger opened Sep 18, 2026·

  ·
  2

• Measure what the decoding knobs buy, and connect beam search to prior coursework

  #26·ringger opened Sep 17, 2026·

  ·
  1

• Add resumable training checkpoints (carries #1 forward)

  #17·ringger opened Sep 14, 2026·

  ·
  3