nit: Adjusted order to match Conv-Act-Pool sequence

gcunhase · gcunhase · commit 9b99aeb2fa18 · 2025-10-17T16:01:32.000-04:00
Signed-off-by: gcunhase &lt;4861122+gcunhase@users.noreply.github.com&gt;
diff --git a/modelopt/onnx/quantization/partitioning.py b/modelopt/onnx/quantization/partitioning.py
@@ -136,7 +136,7 @@ def _is_fusible_mul(mul_node: Node) -> bool:
                 and _is_cask_fusible(consumer_node, partition_node_outputs)
             )
             or (
-                consumer_node.op in ["MaxPool", "BatchNormalization", "Relu"]
+                consumer_node.op in ["BatchNormalization", "Relu", "MaxPool"]
                 and get_fusible_backbone(consumer_node, graph)
             )
             or _is_on_non_residual_path(consumer_node)

Original file line number	Diff line number	Diff line change
`@@ -136,7 +136,7 @@ def _is_fusible_mul(mul_node: Node) -> bool:`
`136`	`136`	`and _is_cask_fusible(consumer_node, partition_node_outputs)`
`137`	`137`	`)`
`138`	`138`	`or (`
`139`		`- consumer_node.op in ["MaxPool", "BatchNormalization", "Relu"]`
	`139`	`+ consumer_node.op in ["BatchNormalization", "Relu", "MaxPool"]`
`140`	`140`	`and get_fusible_backbone(consumer_node, graph)`
`141`	`141`	`)`
`142`	`142`	`or _is_on_non_residual_path(consumer_node)`