Commits

Branch selector

 main

User selector

All users

All time

Commit History

Commits on Jun 10, 2026

• Merge pull request #926 from NatLibFi/issue895-nn-ensemble-pytorch

  Show description for f6030a3
  osma

  authored
  f6030a3

  Copy full SHA for f6030a3

Commits on Jun 2, 2026

• Merge branch 'main' into issue895-nn-ensemble-pytorch

  osma

  committed
  e318f64

  Copy full SHA for e318f64

• Merge pull request #941 from NatLibFi/fix-schemathesis-flaky-test

  Show description for 14b7443
  juhoinkinen

  authored
  14b7443

  Copy full SHA for 14b7443

• Do not skip server errors

  juhoinkinen

  committed
  45404a8

  Copy full SHA for 45404a8

• Set up hypothesis strategy to test only given project-ids

  juhoinkinen

  committed
  32a234d

  Copy full SHA for 32a234d

• exclude schemathesis check for server errors (5xx) to avoid failing tests when 503 is a valid response (e.g. untrained project)

  osma

  committed
  d1f8849

  Copy full SHA for d1f8849

• robustness fix: catch AnnifException in REST API and turn it into a proper REST error

  osma

  committed
  b57773f

  Copy full SHA for b57773f

• remove unused constant EVAL_BATCH_SIZE

  osma

  committed
  e6b934c

  Copy full SHA for e6b934c

• fix early stopping patience calculation to match conventions

  osma

  committed
  55eeaa3

  Copy full SHA for 55eeaa3

• update README for CUDA 12.6 and 13.0 specific pytorch extras

  osma

  committed
  93e4063

  Copy full SHA for 93e4063

• use binary targets for ndcg_batch

  osma

  committed
  0bc987f

  Copy full SHA for 0bc987f

• adjust LMDBDataset __getitem__ and get_subset type hints and docstrings to match reality

  osma

  committed
  b442c80

  Copy full SHA for b442c80

• adjust pytorch and lmdb dependencies for latest versions (torch now supports CUDA 12.6 and 13.0 + 13.2 but no longer 12.8); declare tqdm dependency

  osma

  committed
  ab8c413

  Copy full SHA for ab8c413

• make ndcg_batch return a standard float

  osma

  committed
  9da7ee6

  Copy full SHA for 9da7ee6

• Merge branch 'main' into issue895-nn-ensemble-pytorch

  osma

  committed
  08e7269

  Copy full SHA for 08e7269

• make early stopping subset selection deterministic

  osma

  committed
  8b2c2f4

  Copy full SHA for 8b2c2f4

Commits on Jun 1, 2026

• change model architecture to match torch_per_label_bias_global_plus_delta prototype impl

  osma

  committed
  a2c474b

  Copy full SHA for a2c474b

Commits on Apr 15, 2026

• Mock HfHubHTTPError when testing upload to nonexistent repo (#939)

  juhoinkinen

  authored
  826b35f

  Copy full SHA for 826b35f

Commits on Feb 26, 2026

• use csr_matrix when storing source predictions in LMDB to avoid massive serialization overhead

  osma

  committed
  211fde9

  Copy full SHA for 211fde9

Commits on Feb 23, 2026

• Merge branch 'main' into issue895-nn-ensemble-pytorch

  osma

  authored
  8c02e8b

  Copy full SHA for 8c02e8b

Commits on Feb 17, 2026

• Merge pull request #935 from deutsche-nationalbibliothek/issue934_fix-docs-uv-sync

  Show description for 4ab42dc
  mfakaehler

  authored
  4ab42dc

  Copy full SHA for 4ab42dc

• Fix documentation for usage of uv sync

  mfakaehler

  committed
  a62efeb

  Copy full SHA for a62efeb

Commits on Feb 12, 2026

• upgrade to torch 2.10

  osma

  committed
  d0970ac

  Copy full SHA for d0970ac

• remove print statements, log chosen best epoch

  osma

  committed
  63b5b97

  Copy full SHA for 63b5b97

Commits on Feb 11, 2026

• increase max-epochs default to 50, since we have early stopping in place

  osma

  committed
  c489082

  Copy full SHA for c489082

• mention the --group all option in README

  osma

  committed
  94f1df2

  Copy full SHA for 94f1df2

• ensure early stopping is triggered in nn_ensemble unit tests

  osma

  committed
  b127b39

  Copy full SHA for b127b39

• scale predictions wider within 0..1 range

  osma

  committed
  f18011b

  Copy full SHA for f18011b

Commits on Feb 10, 2026

• switch back to unlimited nDCG for early stopping, instead of arbitrary @1000

  osma

  committed
  4ef1bfe

  Copy full SHA for 4ef1bfe

• apply log1p scaling for targets as well (improves results a lot)

  osma

  committed
  f91c50b

  Copy full SHA for f91c50b

• fix type annotation and reformat

  osma

  committed
  46d9494

  Copy full SHA for 46d9494

• use source weights to initialize model arameters

  osma

  committed
  4156b17

  Copy full SHA for 4156b17

• define constants for EVAL_BATCH_SIZE and EARLY_STOPPING_PATIENCE for clarity

  osma

  committed
  34dde70

  Copy full SHA for 34dde70

• implement early stopping

  osma

  committed
  c047591

  Copy full SHA for c047591

• cleanups

  osma

  committed
  72698d2

  Copy full SHA for 72698d2