Issues

• [KDA][SM90][Accuracy] Intracard CP accumulates large errors on 64K slow-decay heads

  inclusionAI/cuLA#126 · Filter by author ginowu opened on Sep 30, 2026
  1

  

• [Feature] Add GDN2 support

  enhancement
  inclusionAI/cuLA#112 · Filter by author icavan opened on Jul 23, 2026
  1

  

• Rewrite SM100 KDA Forward in CuteDSL

  enhancement
  inclusionAI/cuLA#92 · Filter by author icavan opened on Jun 12, 2026
  1

  

• [Feature] Add sm80 support

  inclusionAI/cuLA#88 · Filter by author icavan opened on Jun 10, 2026
  1

  

• Lightning MTP (Multi-Token Prediction) support

  enhancement
  inclusionAI/cuLA#82 · Filter by author icavan opened on May 26, 2026
  

• [Feature] Add GVA support for Lightning

  enhancementgood first issuehelp wanted
  inclusionAI/cuLA#81 · Filter by author icavan opened on May 25, 2026
  1

  

• [Feature] Integrate Lightning SM90 Implementation via CuTe DSL

  good first issuehelp wantedsm90
  inclusionAI/cuLA#76 · Filter by author zheyang0825 opened on May 22, 2026
  3

  

• WIP-RFC: Multi-GPU Unified Interface for cuLA

  enhancementhelp wantedintegration
  inclusionAI/cuLA#50 · Filter by author icavan opened on Apr 12, 2026
  2

  

• Explore high-occupancy kernel variants with smaller chunk sizes

  enhancementperformanceresearch
  inclusionAI/cuLA#49 · Filter by author icavan opened on Apr 11, 2026
  

• Lightning-attn supports pretransposed states

  good first issuehelp wantedperformance
  inclusionAI/cuLA#32 · Filter by author icavan opened on Apr 5, 2026
  3

  

• Prefill Output Pretransposed States [V, K] for Decode

  help wantedinferenceperformance
  inclusionAI/cuLA#28 · Filter by author icavan opened on Apr 5, 2026
  2

  

• # RFC: Split cuLA into Core Kernel Layer and FLA Transitional Wrapper

  enhancementintegration
  inclusionAI/cuLA#25 · Filter by author icavan opened on Apr 4, 2026
  3

  

• H20 Lightning Attention decode kernel — performance profiling & tuning

  inferenceperformancesm90
  inclusionAI/cuLA#21 · Filter by author icavan opened on Apr 3, 2026
  1

  

• More aggressive fusion of small neighboring kernels (cumsum etc.) for inference

  enhancementhelp wantedinference
  inclusionAI/cuLA#18 · Filter by author icavan opened on Apr 3, 2026
  2

  

• Optimize KDA prefill kernel (SM10X)

  enhancementperformance
  inclusionAI/cuLA#16 · Filter by author icavan opened on Apr 3, 2026
  

• Kernel-level compute-communication overlapping for CP linear attention kernels via nvshmem

  distributedenhancementperformance
  inclusionAI/cuLA#15 · Filter by author icavan opened on Apr 3, 2026
  3

  

• KDA Backward pass optimizations

  enhancementhelp wantedsm10xsm90train
  inclusionAI/cuLA#14 · Filter by author icavan opened on Apr 3, 2026
  1

  

• Modular GDN forward / backward kernels (compatible with Kimi CP)

  enhancementhelp wanted
  inclusionAI/cuLA#13 · Filter by author icavan opened on Apr 3, 2026
  7

  

• Small B/H/S optimizations

  enhancementhelp wantedperformance
  inclusionAI/cuLA#11 · Filter by author icavan opened on Apr 3, 2026
  3

  

• Larger chunk size and 2-CTA on SM10X for improved throughput

  enhancementperformanceresearch
  inclusionAI/cuLA#10 · Filter by author icavan opened on Apr 3, 2026
  1

  

• Polynomial approximation to mitigate the exponential bottleneck (Flash-Attention-4)

  help wantedperformanceresearch
  inclusionAI/cuLA#9 · Filter by author icavan opened on Apr 3, 2026

• Integrate cuLA into flash-linear-attention via FLA's kernel dispatch mechanism

  enhancementgood first issueintegration
  inclusionAI/cuLA#8 · Filter by author icavan opened on Apr 3, 2026
  1