All pull requests

Pull requests

• Bound out-of-range ECDF ranks by the train extremes' midranks

  #1310·LeoGrin opened Sep 22, 2026·

  ·

• Remap bucket masses by summing overlaps instead of differencing a CDF

  no changelog needed
  #1309·LeoGrin opened Sep 22, 2026·

  ·

• feat(finetuning): add resume_from_checkpoint opt-out to fit()

  #1308·aniruddhaadak80 opened Sep 22, 2026·

  ·
  1

• Fix string-based PyTorch version comparison

  #1305·mayuriphad opened Sep 22, 2026·

  ·
  1

  4

• feat(estimator): support nested inference_config__FIELD params for tuning

  #1303·aniruddhaadak80 opened Sep 22, 2026·

  ·
  1

  5

• fix(version): compare torch versions via TorchVersion in MPS and GQA gates

  #1301·aniruddhaadak80 opened Sep 22, 2026·

  ·
  1

  5

• feat(kv-cache): add kv_cache_precision="adaptive" that follows the attention backend

  #1299·ggprior opened Sep 21, 2026·

  ·
  2

• Difference the regression border remap in float32, escalating to float64 only when it loses a bucket

  #1298·jmetzen-pl opened Sep 21, 2026·

  ·
  3

• Run n_preprocessing_jobs > 1 on threads instead of processes

  #1295·LeoGrin opened Sep 20, 2026·

  ·
  4

• Pin all GitHub Actions to commit SHAs and update outdated versions

  no changelog needed
  #1293·eddiebergman opened Sep 19, 2026·

  ·
  16

• Replace the FA3 attention backend with FlashAttention-4 (Hopper and Blackwell)

  #1291·tolleybot opened Sep 18, 2026·

  ·
  1

  9

• Pool the regression ensemble in log space

  #1289·jmetzen-pl opened Sep 18, 2026·

  ·
  5

  

• Stop the regression border remap from losing bucket mass to float32

  #1288·jmetzen-pl opened Sep 17, 2026·

  ·
  22

  

• Keep an inspection of the fit input and the evidence behind each modality decision

  #1285·alanprior opened Sep 16, 2026·

  ·
  4

• Option to run equal-shape ensemble members as one batched forward pass

  #1268·Innixma opened Sep 13, 2026·4/4·

  ·
  3

• Add kv_cache_at_predict: build each member's KV cache per predict and drop it

  #1267·Innixma opened Sep 12, 2026·

  ·
  4

• Faster frozen encoder, KDI fit and dtype fixing in the per-member preprocessing

  #1265·Innixma opened Sep 12, 2026·2/4·

  ·
  3

• Encode numeric categorical columns with a numpy ordinal encoder

  #1263·Innixma opened Sep 11, 2026·1/4·

  ·
  12

• Adding Cookbook to readme

  #1262·TuanaCelik opened Sep 11, 2026·

  ·
  1

• [Automated PR (update-readme)]: Document TABPFN_MODEL_CACHE_SIZE and TABPFN_EXCLUDE_DEVICES env vars

  documentationno changelog needed
  #1244·claude[bot] opened Sep 7, 2026·

  ·

• Fix FullSupportBarDistribution tail sampling and quantiles

  #1215·ppguo opened Aug 27, 2026·

  ·
  1

  14

• test(notebooks): run the demo notebook end to end on both backends

  #1214·ggprior opened Aug 27, 2026·

  ·

• Move from ux.priorlabs.ai to platform.priorlabs.ai

  #1213·safaricd opened Aug 27, 2026·

  ·

• Bump mypy from 2.3.0 to 2.3.1

  dependenciespython
  #1210·dependabot[bot] opened Aug 27, 2026·

  ·

• Apply regression target transforms to the unnormalized target [RES-2639]

  #1199·bejaeger opened Aug 21, 2026·

  ·
  2