conv

Modules

modelopt.torch.kernels.quantization.conv.bench_implicit_gemm	Latency benchmark: implicit GEMM (quant / non-quant) vs cuDNN conv3d.
modelopt.torch.kernels.quantization.conv.implicit_gemm_cuda	Conv3D Implicit GEMM with BF16 WMMA Tensor Cores and optional fused FP4 quantization.

Implicit-GEMM CUDA kernel for quantized 3D convolution.