Issues

• Regularized cosine loss returns NaN gradients for zero embeddings

  google-research/kauldron#1549 · Filter by author sylvesterkaczmarek opened on Oct 5, 2026

• Standard deviation masks weight the count but not the moments

  google-research/kauldron#1543 · Filter by author sylvesterkaczmarek opened on Oct 5, 2026

• Bug: np.float128 in ktyping/dtypes.py crashes on macOS ARM (Apple Silicon)

  google-research/kauldron#1416 · Filter by author ac12644 opened on Apr 5, 2026

• Issue with typeguard 4.5.0

  google-research/kauldron#1337 · Filter by author nmilosev opened on Feb 15, 2026

• Example of distributed training?

  google-research/kauldron#1265 · Filter by author radna0 opened on Oct 2, 2025

• Training jobs stall in TPU VM v4-64 environment regardless of workdir

  google-research/kauldron#1120 · Filter by author tucnak opened on Apr 28, 2025
  1

• Support for Callback Functions in Trainer Class

  google-research/kauldron#1078 · Filter by author Lucas-Fernandes-Martins opened on Mar 26, 2025
  7