Search results

FA4: Flash Attention 4 with WGMMA for SM120

enhancementNew feature or requestNew feature or request

Status: Open.

#192 In m96-chan/PyGPUkit;

· m96-chan opened on Jan 16, 2026

SM120 (Blackwell) FA3 Attention Optimization

Status: Open.

#191 In m96-chan/PyGPUkit;

· m96-chan opened on Jan 16, 2026

Refactor LLaMA 4 kernels to modular nn/ structure

Status: Open.

#190 In m96-chan/PyGPUkit;

· m96-chan opened on Jan 16, 2026

perf(diffusion): FLUX.1 transformer performance optimization

enhancementNew feature or requestNew feature or request

Status: Open.

#187 In m96-chan/PyGPUkit;

· m96-chan opened on Jan 1, 2026

Add comprehensive NumPy comparison tests for GPU kernels

Status: Open.

#186 In m96-chan/PyGPUkit;

· m96-chan opened on Jan 1, 2026

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

Status: Open.

#184 In m96-chan/PyGPUkit;

· m96-chan opened on Jan 1, 2026

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

enhancementNew feature or requestNew feature or request

v0.3Advanced: Triton backend, advanced opsAdvanced: Triton backend, advanced ops

Status: Open.

#182 In m96-chan/PyGPUkit;

· m96-chan opened on Jan 1, 2026

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

enhancementNew feature or requestNew feature or request

v0.2Scheduler: memory + bandwidth guaranteesScheduler: memory + bandwidth guarantees

Status: Open.

#181 In m96-chan/PyGPUkit;

· m96-chan opened on Jan 1, 2026

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

enhancementNew feature or requestNew feature or request

v0.3Advanced: Triton backend, advanced opsAdvanced: Triton backend, advanced ops

Status: Open.

#180 In m96-chan/PyGPUkit;

· m96-chan opened on Jan 1, 2026

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

enhancementNew feature or requestNew feature or request

v0.3Advanced: Triton backend, advanced opsAdvanced: Triton backend, advanced ops

Status: Open.

#176 In m96-chan/PyGPUkit;

· m96-chan opened on Dec 31, 2025

feat(nn): Add Retention / RetNet mechanism

Status: Open.

#172 In m96-chan/PyGPUkit;

· m96-chan opened on Dec 30, 2025

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

Status: Open.

#170 In m96-chan/PyGPUkit;

· m96-chan opened on Dec 30, 2025