Issues

• FA4: Flash Attention 4 with WGMMA for SM120

  enhancement
  m96-chan/PyGPUkit#192 · Filter by author m96-chan opened on Jan 16, 2026

• SM120 (Blackwell) FA3 Attention Optimization

  m96-chan/PyGPUkit#191 · Filter by author m96-chan opened on Jan 16, 2026

• Refactor LLaMA 4 kernels to modular nn/ structure

  m96-chan/PyGPUkit#190 · Filter by author m96-chan opened on Jan 16, 2026

• perf(diffusion): FLUX.1 transformer performance optimization

  enhancement
  m96-chan/PyGPUkit#187 · Filter by author m96-chan opened on Jan 1, 2026

• Add comprehensive NumPy comparison tests for GPU kernels

  m96-chan/PyGPUkit#186 · Filter by author m96-chan opened on Jan 1, 2026

• test(tts): Verify Kokoro TTS implementation (#183)

  m96-chan/PyGPUkit#184 · Filter by author m96-chan opened on Jan 1, 2026

• perf(tts): Kokoro layers use CPU fallback (numpy) instead of GPU kernels

  enhancementv0.3
  m96-chan/PyGPUkit#182 · Filter by author m96-chan opened on Jan 1, 2026

• perf(llm): Batch decode missing CUDA Graph and zero-alloc optimizations

  enhancementv0.2
  m96-chan/PyGPUkit#181 · Filter by author m96-chan opened on Jan 1, 2026

• perf(asr): Whisper conv1d uses CPU fallback instead of GPU kernel

  enhancementv0.3
  m96-chan/PyGPUkit#180 · Filter by author m96-chan opened on Jan 1, 2026

• RFC: Universal Model Architecture Support (GPT/MoE/VLM/LRM/LAM/HLM/LCM)

  enhancementv0.3
  m96-chan/PyGPUkit#176 · Filter by author m96-chan opened on Dec 31, 2025

• feat(nn): Add Retention / RetNet mechanism

  m96-chan/PyGPUkit#172 · Filter by author m96-chan opened on Dec 30, 2025

• feat(nn): Add Sliding Window / Local Attention for long context

  m96-chan/PyGPUkit#170 · Filter by author m96-chan opened on Dec 30, 2025

• feat(llm): MoE (Mixture of Experts) Optimization

  enhancement
  m96-chan/PyGPUkit#164 · Filter by author m96-chan opened on Dec 30, 2025

• feat(core): Async Execution and Prefetch

  enhancement
  m96-chan/PyGPUkit#162 · Filter by author m96-chan opened on Dec 30, 2025

• feat(core): Tensor Parallelism

  enhancement
  m96-chan/PyGPUkit#161 · Filter by author m96-chan opened on Dec 30, 2025

• feat(core): Mixed Precision Policy API

  enhancement
  m96-chan/PyGPUkit#160 · Filter by author m96-chan opened on Dec 30, 2025

• feat(jit): Custom Kernel Authoring API

  enhancement
  m96-chan/PyGPUkit#157 · Filter by author m96-chan opened on Dec 30, 2025

• feat(core): Graph compilation and kernel fusion

  m96-chan/PyGPUkit#156 · Filter by author m96-chan opened on Dec 30, 2025

• feat(ops): Flash Attention 2/3 with variable sequence length

  m96-chan/PyGPUkit#155 · Filter by author m96-chan opened on Dec 30, 2025

• feat(core): Enhanced stream management and async operations

  m96-chan/PyGPUkit#154 · Filter by author m96-chan opened on Dec 30, 2025

• feat: Auto-tuning kernel dispatcher with size-aware selection

  enhancement
  m96-chan/PyGPUkit#120 · Filter by author m96-chan opened on Dec 26, 2025

• feat: Auto-quantization based on available VRAM

  enhancement
  m96-chan/PyGPUkit#119 · Filter by author m96-chan opened on Dec 26, 2025

• feat(decode): Medusa Multi-Head Speculative Decoding

  m96-chan/PyGPUkit#114 · Filter by author m96-chan opened on Dec 25, 2025

• feat(decode): External Draft Model Speculative Decoding

  m96-chan/PyGPUkit#113 · Filter by author m96-chan opened on Dec 25, 2025

• feat(quantize): Post-Training Quantization (PTQ) - INT4/AWQ/GPTQ support

  m96-chan/PyGPUkit#112 · Filter by author m96-chan opened on Dec 25, 2025