From abd8d6d6467f79b9a69a29c22a23f244b1362760 Mon Sep 17 00:00:00 2001 From: JasonHonKL Date: Fri, 14 Aug 2026 15:37:15 +0800 Subject: [PATCH] fix quantize matrix multiplication floor issue --- mlx/backend/metal/quantized.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/mlx/backend/metal/quantized.cpp b/mlx/backend/metal/quantized.cpp index f659e16c93..5db3893226 100644 --- a/mlx/backend/metal/quantized.cpp +++ b/mlx/backend/metal/quantized.cpp @@ -649,7 +649,7 @@ void qvm_split_k( constexpr int bk = 32; int bn = std::min(group_size, 32) * num_simdgroups; MTL::Size group_dims = MTL::Size(bk, num_simdgroups, 1); - MTL::Size grid_dims = MTL::Size(M, N / bn, B); + MTL::Size grid_dims = MTL::Size(M, (N + bn - 1) / bn, B); auto x_shape = x.shape(); auto x_strides = x.strides();