kernels

Modules

modelopt.torch.kernels.common	Common (non-domain-specific) kernels.
modelopt.torch.kernels.quantization	Quantization kernels: conv (implicit GEMM) and gemm (tensor_quant + Triton FP4/FP8).
modelopt.torch.kernels.sparsity	Sparsity kernels: attention (Triton skip-softmax backends) and gemm (placeholder).

ModelOpt kernel library: common, quantization (conv, gemm), sparsity (attention, gemm).