Issues

• ExactMatch counts fully ignored samples as correct

  Lightning-AI/torchmetrics#3585 · Filter by author AryanRG opened on Oct 18, 2026
  1

• BootStrapper computes the quantile over all outputs at once for metrics with non-scalar results

  Lightning-AI/torchmetrics#3582 · Filter by author RaphaelFakhri opened on Oct 18, 2026
  1

• Variance-weighted R2 returns NaN when every target output is constant

  Lightning-AI/torchmetrics#3580 · Filter by author AaravMehta-07 opened on Oct 16, 2026
  1

• Quick survey: your experience as a PyTorch Ecosystem maintainer

  Lightning-AI/torchmetrics#3579 · Filter by author michellearoth opened on Oct 16, 2026
  1

• Pearson and concordance return NaN for locally constant distributed shards

  Lightning-AI/torchmetrics#3577 · Filter by author kocean9-freedom opened on Oct 13, 2026
  1

• Perplexity returns inf in float16 when one update sums many tokens (follow-up to #3537)

  bug / fixhelp wanted
  Lightning-AI/torchmetrics#3575 · Filter by author arnav-verma200 opened on Oct 12, 2026
  1

• Running loses list-backed metric state after resetting the base metric

  Lightning-AI/torchmetrics#3573 · Filter by author j45856021-dev opened on Oct 12, 2026
  2

• MetricTracker.compute_all fails for samplewise metrics with different sample counts

  Lightning-AI/torchmetrics#3569 · Filter by author ktz03 opened on Oct 10, 2026

• RMSE-SW and RASE return NaN for window_size=1

  Lightning-AI/torchmetrics#3567 · Filter by author ktz03 opened on Oct 9, 2026
  1

• Variance-weighted explained variance returns NaN for constant targets

  Lightning-AI/torchmetrics#3565 · Filter by author aalquwayfili opened on Oct 8, 2026

• GeneralizedDiceScore is wrong for a class absent from a sample's target but present in predictions

  Lightning-AI/torchmetrics#3563 · Filter by author raunak2007 opened on Oct 7, 2026
  1

• Integration proposal: JLD and JLD-fast image perceptual distances

  Lightning-AI/torchmetrics#3561 · Filter by author shreshthsaini opened on Oct 7, 2026
  1

• ProcrustesDisparity returns a zero-score tuple for invalid clouds and crashes in update

  Lightning-AI/torchmetrics#3542 · Filter by author Yang1107-wzy opened on Oct 2, 2026
  1

• Perplexity returns inf when a target token's softmax probability underflows

  Lightning-AI/torchmetrics#3537 · Filter by author MohammadHijjawi97 opened on Sep 30, 2026

• FleissKappa with mode="probs" raises when there are fewer raters than categories

  Lightning-AI/torchmetrics#3536 · Filter by author MohammadHijjawi97 opened on Sep 30, 2026
  1

• MSE and RMSE reject non-contiguous inputs in the default single-output path

  Lightning-AI/torchmetrics#3534 · Filter by author AHMETHAKANBEZIR1 opened on Sep 30, 2026

• Mean absolute error wraps small-integer residuals before taking abs

  Lightning-AI/torchmetrics#3532 · Filter by author AHMETHAKANBEZIR1 opened on Sep 30, 2026

• LogCoshError silently wraps residuals for smaller integer dtypes

  Lightning-AI/torchmetrics#3528 · Filter by author AHMETHAKANBEZIR1 opened on Sep 30, 2026
  1

• R2Score reports a perfect score for a mean predictor after rescaling nonconstant targets

  Lightning-AI/torchmetrics#3523 · Filter by author galleonli opened on Sep 29, 2026
  1

• Spearman correlation flips sign for a large tie

  Lightning-AI/torchmetrics#3521 · Filter by author SashaMIT opened on Sep 29, 2026
  1

• multilabel_ranking_loss depends on label order when scores are tied

  Lightning-AI/torchmetrics#3519 · Filter by author kratos0718 opened on Sep 29, 2026
  1

• Retrieval MAP, MRR, precision and recall treat non-positive scores as not retrieved

  Lightning-AI/torchmetrics#3518 · Filter by author aalquwayfili opened on Sep 28, 2026
  1

• WER/CER return inf/nan with no warning when the total reference length is zero; jiwer defines a value for this case

  Lightning-AI/torchmetrics#3516 · Filter by author shaurya416 opened on Sep 25, 2026
  1

• Feature Request: Add Fréchet Video Distance (FVD)

  enhancement
  Lightning-AI/torchmetrics#3515 · Filter by author lavinal712 opened on Sep 25, 2026
  1

• ConcordanceCorrCoef mixes (n-1) variances with a 1/n mean term, so it does not match Lin's CCC

  bug / fixhelp wanted
  Lightning-AI/torchmetrics#3512 · Filter by author lucaluo925 opened on Sep 24, 2026
  1