From 518327d6317756f02cacdc0433d48ac57d328e99 Mon Sep 17 00:00:00 2001 From: Ziang Li Date: Fri, 27 Feb 2026 00:25:16 -0800 Subject: [PATCH] Fix error in Blackwell document of referring to Mxf4 format as NVF4 (#2977) * Update blackwell_functionality.md Fixed the error of referring to Mxf4 format as NVF4. * Correct NVF4 output matrix reference in documentation --- media/docs/cpp/blackwell_functionality.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/media/docs/cpp/blackwell_functionality.md b/media/docs/cpp/blackwell_functionality.md index 8b410e1c..6774d9dc 100644 --- a/media/docs/cpp/blackwell_functionality.md +++ b/media/docs/cpp/blackwell_functionality.md @@ -443,11 +443,11 @@ which are listed above. Several examples of block scaled dense GEMM kernels can be found in [examples/72_blackwell_narrow_precision_gemm](https://github.com/NVIDIA/cutlass/tree/main/examples/72_blackwell_narrow_precision_gemm/) directory: 1. [NVF4 Gemm with block scaling](https://github.com/NVIDIA/cutlass/tree/main/examples/72_blackwell_narrow_precision_gemm/72a_blackwell_nvfp4_bf16_gemm.cu) 2. [NVF4 Gemm with block scaling and NVF4 output matrix](https://github.com/NVIDIA/cutlass/tree/main/examples/72_blackwell_narrow_precision_gemm/72b_blackwell_nvfp4_nvfp4_gemm.cu) -3. [Mixed precision Nvf4 x Mxf8 GEMM with block scaling](https://github.com/NVIDIA/cutlass/tree/main/examples/72_blackwell_narrow_precision_gemm/72c_blackwell_mixed_mxfp8_bf16_gemm.cu) +3. [Mixed precision Mxf4 x Mxf8 GEMM with block scaling](https://github.com/NVIDIA/cutlass/tree/main/examples/72_blackwell_narrow_precision_gemm/72c_blackwell_mixed_mxfp8_bf16_gemm.cu) Several examples of block scaled sparse GEMM kernels can be found in [examples/84_blackwell_narrow_precision_sparse_gemm](https://github.com/NVIDIA/cutlass/tree/main/examples/84_blackwell_narrow_precision_sparse_gemm) directory: 1. [NVF4 Gemm with block scaling](https://github.com/NVIDIA/cutlass/tree/main/examples/84_blackwell_narrow_precision_sparse_gemm/84a_blackwell_nvfp4_bf16_sparse_gemm.cu) -2. [Mixed precision Nvf4 x Mxf8 GEMM with block scaling](https://github.com/NVIDIA/cutlass/tree/main/examples/84_blackwell_narrow_precision_sparse_gemm/84b_blackwell_mixed_mxfp8_bf16_sparse_gemm.cu) +2. [Mixed precision Mxf4 x Mxf8 GEMM with block scaling](https://github.com/NVIDIA/cutlass/tree/main/examples/84_blackwell_narrow_precision_sparse_gemm/84b_blackwell_mixed_mxfp8_bf16_sparse_gemm.cu) Collective builder interface expects the same arguments as any other CUTLASS 3.x kernels as described [here](gemm_api_3x.md#collective-builder-for-collectivemmas) with a small difference for Collective MMA builder interface.