From a16537f601897af9ed5e72219a58d3507dd53734 Mon Sep 17 00:00:00 2001 From: Liu Liu Date: Mon, 16 Dec 2024 19:33:25 -0500 Subject: [PATCH] Revert the parameter selection since it is adversarial on M4. --- lib/nnc/mfa/v2/AttentionDescriptor.cpp | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/lib/nnc/mfa/v2/AttentionDescriptor.cpp b/lib/nnc/mfa/v2/AttentionDescriptor.cpp index 6bda3c247..7b8dbd4c1 100644 --- a/lib/nnc/mfa/v2/AttentionDescriptor.cpp +++ b/lib/nnc/mfa/v2/AttentionDescriptor.cpp @@ -454,10 +454,10 @@ std::vector AttentionDescriptor::forwardMixed(MTL::Device if (device->supportsFamily(MTL::GPUFamily(1009))) { return { AttentionParameterRow(32, 16, 128, 16, { AttentionOperand::Q, AttentionOperand::O }), - AttentionParameterRow(64, 16, 128, 32, { AttentionOperand::Q, AttentionOperand::O }), - AttentionParameterRow(160, 32, 128, 32, { AttentionOperand::O }), - AttentionParameterRow(224, 32, 128, 32, { AttentionOperand::Q }), - AttentionParameterRow(384, 32, 128, 32, {}) + AttentionParameterRow(96, 16, 128, 32, { AttentionOperand::Q, AttentionOperand::O }), + AttentionParameterRow(160, 16, 128, 32, { AttentionOperand::O }), + AttentionParameterRow(224, 16, 128, 32, { AttentionOperand::Q }), + AttentionParameterRow(384, 16, 128, 32, {}) }; } else { return {