Fix custom annotation (#6978)

cccclai · web-flow · commit 359e9d3a09e3 · 2024-11-20T11:08:59.000-08:00
Summary: Looks like it's added in pytorch/executorch#6849, maybe it was using the old api for the default 8bit quantization Differential Revision: D66219251
diff --git a/backends/qualcomm/quantizer/custom_annotation.py b/backends/qualcomm/quantizer/custom_annotation.py
@@ -78,7 +78,7 @@ def annotate_single_in_single_out(
         )
 
     def annotate_matmul_input1(node: Node):
-        quantization_config_8a8w = get_default_8bit_qnn_ptq_config(
+        quantization_config_8a8w = get_8a8w_qnn_ptq_config(
             act_symmetric=True, act_observer=MinMaxObserver
         )
         while isinstance(node, Node) and node.op == "call_function":

Original file line number	Diff line number	Diff line change
`@@ -78,7 +78,7 @@ def annotate_single_in_single_out(`
`78`	`78`	`)`
`79`	`79`
`80`	`80`	`def annotate_matmul_input1(node: Node):`
`81`		`- quantization_config_8a8w = get_default_8bit_qnn_ptq_config(`
	`81`	`+ quantization_config_8a8w = get_8a8w_qnn_ptq_config(`
`82`	`82`	`act_symmetric=True, act_observer=MinMaxObserver`
`83`	`83`	`)`
`84`	`84`	`while isinstance(node, Node) and node.op == "call_function":`