Modules:
| Name | Description |
|---|---|
activation | Custom activation functions for diffusion models. |
adalayernorm | |
custom_op | |
fourier | |
fused_moe | |
fused_qk_norm_rope | Q/K RMSNorm followed by packed RoPE (non-interleaved or interleaved). |
fused_qk_rope | Bit-exact paired Q/K full-width interleaved RoPE for CUDA. |
indexed_modulation | Fused indexed modulation with FP32 accumulation. |
mhc | Native and eager Triton mHC post-processing, without projection or tensor caches. |
norm | |
qwen_select01_modulation | |
rope | |
swiglu7 | Interleaved clamped SwiGLU7 with a native compile path. |