From 12e6a585ae020cdcef9ba61aae45eec5ba29591d Mon Sep 17 00:00:00 2001 From: Laith Sakka Date: Mon, 12 Oct 2026 10:09:57 -0700 Subject: [PATCH] =?UTF-8?q?[Fix][Compile]=20Declare=20group=5Fshape=20in?= =?UTF-8?q?=20the=20static=20FP8=20quant=20fusion=20=E2=80=A6=20(#61108)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Signed-off-by: Laith Sakka Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com> --- vllm/compilation/passes/fusion/matcher_utils.py | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/vllm/compilation/passes/fusion/matcher_utils.py b/vllm/compilation/passes/fusion/matcher_utils.py index 6d959c3067..80be65a978 100644 --- a/vllm/compilation/passes/fusion/matcher_utils.py +++ b/vllm/compilation/passes/fusion/matcher_utils.py @@ -481,8 +481,13 @@ class MatcherQuantFP8(MatcherCustomOp): if self.quant_key.scale.static: assert scale is not None + group_shape = self.quant_key.scale.group_shape _, result = auto_functionalized( - self.QUANT_OP, result=result, input=input, scale=scale + self.QUANT_OP, + result=result, + input=input, + scale=scale, + group_shape=[group_shape.row, group_shape.col], ) return result, scale else: