fix match index

neuralmagic · Aug 10, 2023 · 7b9374a · 7b9374a
1 parent 320253c
commit 7b9374a
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/src/sparseml/exporters/transforms/matmul_to_qlinearmatmul.py b/src/sparseml/exporters/transforms/matmul_to_qlinearmatmul.py
@@ -97,7 +97,7 @@ def transform(self, model: ModelProto) -> ModelProto:
             for quantize_linear_parent in [match.parents[0][0], match.parents[1][0]]:
                 if graph.get_init_by_name(quantize_linear_parent.input[0]):
                     continue
-                next_node = graph.get_node_single_child(match.children[-1])
+                next_node = graph.get_node_single_child(match.children[0][-1])
                 if next_node.op_type in COMMON_QUANTIZABLE_OP_TYPES:
                     # the Q/DQ block after this matmul directly feeds into another
                     # quantizable op, it is more likely that this Q/DQ is relevant