Issues

• Make non-paged CUDA optimizer updates current-stream correct and asynchronous

  heiheiha798/bitsandbytes#25 · Filter by author heiheiha798 opened on Aug 21, 2026

• Benchmark fused SM103 DP4A small-M INT8 scaled matmul

  heiheiha798/bitsandbytes#24 · Filter by author heiheiha798 opened on Aug 21, 2026
  1

• Benchmark stream-ordered lookahead prefetch for CUDA paged optimizers on B300

  heiheiha798/bitsandbytes#23 · Filter by author heiheiha798 opened on Aug 21, 2026
  1

• Fuse nested scale reconstruction into SM103 4-bit dequantization

  heiheiha798/bitsandbytes#21 · Filter by author heiheiha798 opened on Aug 21, 2026

• Make non-paged 8-bit optimizers safe for FSDP2 local shards

  heiheiha798/bitsandbytes#20 · Filter by author heiheiha798 opened on Aug 21, 2026
  1

• Respect the current CUDA stream in blockwise quantization

  heiheiha798/bitsandbytes#18 · Filter by author heiheiha798 opened on Aug 20, 2026

• Support >INT_MAX CUDA 4-bit quantization with block-aligned chunked launches

  heiheiha798/bitsandbytes#17 · Filter by author heiheiha798 opened on Aug 20, 2026
  1

• Fuse LLM.int8 correction operand preparation into one CUDA kernel

  heiheiha798/bitsandbytes#16 · Filter by author heiheiha798 opened on Aug 20, 2026
  1

• Retry tiled LLM.int8 outlier detection with an empty-mask fast path

  heiheiha798/bitsandbytes#15 · Filter by author heiheiha798 opened on Aug 20, 2026
  1

• Evaluate a tiled native detector for LLM.int8 outlier columns

  heiheiha798/bitsandbytes#14 · Filter by author heiheiha798 opened on Aug 20, 2026
  1

• Evaluate lane-pair nested-scale broadcast in CUDA SIMT 4-bit GEMM

  heiheiha798/bitsandbytes#13 · Filter by author heiheiha798 opened on Aug 20, 2026
  1

• Evaluate multi-warp CTAs for SM103 4-bit quantization

  heiheiha798/bitsandbytes#12 · Filter by author heiheiha798 opened on Aug 20, 2026
  1

• Evaluate redundant barrier removal in CUDA blockwise dequantization

  heiheiha798/bitsandbytes#10 · Filter by author heiheiha798 opened on Aug 20, 2026

• Calibrate SM103 quantile-map replicas in the Adam 8-bit optimizer

  heiheiha798/bitsandbytes#9 · Filter by author heiheiha798 opened on Aug 20, 2026
  1

• Investigate a 256-thread SM103 path for small 32-bit optimizer tensors

  heiheiha798/bitsandbytes#8 · Filter by author heiheiha798 opened on Aug 20, 2026
  1

• Investigate SM103 bias-correction broadcast in 32-bit optimizers

  heiheiha798/bitsandbytes#7 · Filter by author heiheiha798 opened on Aug 20, 2026
  1

• Investigate a persistent SM103 grid for blockwise 8-bit optimizers

  heiheiha798/bitsandbytes#6 · Filter by author heiheiha798 opened on Aug 20, 2026
  1

• Investigate SM103 launch geometry for 4-bit dequantization

  heiheiha798/bitsandbytes#5 · Filter by author heiheiha798 opened on Aug 20, 2026
  1

• Investigate SM103 SIMT-vs-MMA dispatch for fused 4-bit GEMM

  heiheiha798/bitsandbytes#3 · Filter by author heiheiha798 opened on Aug 20, 2026

• Calibrate fused 4-bit GEMM dispatch for B300 (SM103)

  heiheiha798/bitsandbytes#1 · Filter by author heiheiha798 opened on Aug 20, 2026