Issues

• matmul_kernel_persistent runs fp32 inputs through TF32 — silent precision loss vs torch.mm

  thinking-machines-lab/batch_invariant_ops#23 · Filter by author chenwenxiaolive opened on Jun 22, 2026

• Incompatible with all public PyTorch/Triton releases - tl.range flatten parameter not available

  thinking-machines-lab/batch_invariant_ops#14 · Filter by author hkraemer opened on Jan 21, 2026
  2

• is log_softmax undeterminsitic?

  thinking-machines-lab/batch_invariant_ops#13 · Filter by author Intsigstephon opened on Nov 22, 2025
  1

• Low matmul performance

  thinking-machines-lab/batch_invariant_ops#12 · Filter by author AKKamath opened on Oct 13, 2025
  2

• VM crashes during installation using https://github.com/Chillee/vllm.git

  thinking-machines-lab/batch_invariant_ops#8 · Filter by author shivani405 opened on Oct 5, 2025
  2

• Failed to replicate batch invariant using vLLM inference

  thinking-machines-lab/batch_invariant_ops#6 · Filter by author fopdoodle8 opened on Sep 26, 2025
  1