From 9538457f878a811649c6472b281b33088fed9abd Mon Sep 17 00:00:00 2001 From: bhalimi <benjamin.halimi@cea.fr> Date: Fri, 14 Mar 2025 12:59:11 +0000 Subject: [PATCH] edit quantizeNormalizedNetwork() (MatMul support) --- src/PTQ/PTQ.cpp | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/src/PTQ/PTQ.cpp b/src/PTQ/PTQ.cpp index 3a24097..8a36e4f 100644 --- a/src/PTQ/PTQ.cpp +++ b/src/PTQ/PTQ.cpp @@ -1046,6 +1046,10 @@ void quantizeNormalizedNetwork(std::shared_ptr<GraphView> graphView, std::uint8_ std::shared_ptr<Node> scalingNode = getUniqueChild(node); // TODO : assert if scalingNode is a Scaling ... + // TODO : double check this ... + if (node->type() == "MatMul") + rescaling /= inputIsUnsigned ? unsignedMax : signedMax; + multiplyScalingFactor(scalingNode, rescaling) ; } -- GitLab