Issues

Issues

• SentencePieceTokenizer (Unigram, from tokenizer.json): equal-score segmentations are ordered differently from Hugging Face tokenizers

  dotnet/machinelearning#7909 · opened on Nov 12, 2026

• SentencePieceTokenizer: charsmap entries that map to an empty string are not applied, so control characters become <unk>

  dotnet/machinelearning#7908 · opened on Nov 12, 2026

• CI failure: SdcaMulticlassSupportVectorMachine value not in range

  dotnet/machinelearning#7901 · opened on Nov 11, 2026

• Microsoft.ML.TorchSharp QuestionAnswer: training takes the answer's last token as inclusive but the decoder cuts the span as exclusive — every answer comes back without its last token (a one-token answer is the empty string)

  dotnet/machinelearning#7890 · opened on Nov 11, 2026

• Microsoft.ML.TorchSharp NamedEntityRecognition: predicted tags come one word to the right (first slot is the start token's, the last word's tag is missing) — a model that tags every word correctly scores as if it had learned nothing

  dotnet/machinelearning#7889 · opened on Nov 11, 2026

• DataOperationsCatalog.Cache: reading a vector column while the cache is still filling can throw ArgumentOutOfRangeException ("valuesCount") from CacheDataView.ColumnCache.ImplVec.Fetch (Microsoft.ML 5.0.1, arm64)

  dotnet/machinelearning#7884 · opened on Nov 9, 2026

• KMeans ignores example weights in centroid updates and default initialization

  dotnet/machinelearning#7876 · opened on Nov 3, 2026

• Make Word.MergeAll sub-quadratic

  dotnet/machinelearning#7872 · opened on Nov 2, 2026

• SentencePieceTokenizer (Unigram) emits one <unk> per unknown character — SentencePiece merges a run of unknown characters into one <unk>

  dotnet/machinelearning#7859 · opened on Oct 25, 2026

• TorchSharp text trainers (TextClassification, SentenceSimilarity, NER, QA) drop every input character above U+00FF — non-Latin text reaches the model as nothing

  dotnet/machinelearning#7858 · opened on Oct 25, 2026
  1

• SymSGD classification test hangs on Linux ARM32 Helix

  dotnet/machinelearning#7837 · opened on Oct 21, 2026

• TensorFlow image classification tests running out of memory on Linux x64

  dotnet/machinelearning#7836 · opened on Oct 21, 2026

• AutoML BinaryClassificationExperiment.Execute throws NullReferenceException in ConcatenateEntity.ToString when ColumnInformation names a column the preFeaturizer creates

  dotnet/machinelearning#7825 · opened on Oct 21, 2026

• ONNX export of a reloaded NgramExtractingTransformer (useAllLengths: true) writes min_gram_length = ngramLength — the exported graph counts only the longest n-grams

  dotnet/machinelearning#7824 · opened on Oct 21, 2026

• AutoML BinaryClassificationExperiment ignores ColumnInformation.ExampleWeightColumnName — trainers are fitted unweighted

  dotnet/machinelearning#7823 · opened on Oct 20, 2026

• Cloning imported Arrow strings across buffer boundaries corrupts null metadata and breaks row access

  dotnet/machinelearning#7801 · opened on Oct 14, 2026

• ToArrowRecordBatches accumulates arrays from previous batches, producing invalid record batches

  dotnet/machinelearning#7799 · opened on Oct 14, 2026

• Imported Arrow strings lose nulls when exported in nonzero-offset record batches

  dotnet/machinelearning#7800 · opened on Oct 14, 2026

• Update oneDAL to a newer version

  dotnet/machinelearning#7792 · opened on Oct 14, 2026 · ML.NET 7.0

• Add missing XML documentation to public APIs

  dotnet/machinelearning#7788 · opened on Oct 13, 2026
  1

• Remove TensorFlow

  dotnet/machinelearning#7778 · opened on Oct 12, 2026
  1

• CI failure: SQL Server process failed to start in DatabaseLoaderTests

  dotnet/machinelearning#7758 · opened on Oct 1, 2026 · ML.NET Future
  2

• SentencePieceTokenizer.Create throws IndexOutOfRangeException for Unigram models without BOS, such as T5

  dotnet/machinelearning#7756 · opened on Oct 1, 2026
  1

• LlamaTokenizer encodes some runs of repeated characters differently from SentencePiece

  dotnet/machinelearning#7751 · opened on Oct 1, 2026
  1

• QuestionAnsweringTrainer cannot produce answers with characters above U+00FF (answer is decoded one byte per character)

  dotnet/machinelearning#7749 · opened on Oct 1, 2026
  1