nvidia awq only use QuantFormat.QDQ quant format (#22429)

anujj · web-flow · commit ec7aa63b3a51 · 2024-10-14T20:32:59.000-07:00
nvidia awq only use QuantFormat.QDQ quant format
diff --git a/onnxruntime/python/tools/quantization/matmul_4bits_quantizer.py b/onnxruntime/python/tools/quantization/matmul_4bits_quantizer.py
@@ -280,7 +280,7 @@ def __init__(
 
         super().__init__(
             algorithm="nvidia_awq",
-            quant_format=quant_format,
+            quant_format=QuantFormat.QDQ,
             op_types_to_quantize=None,  # Assuming op_types_to_quantize is handled elsewhere
             quant_axes=None,  # Assuming quant_axes is handled elsewhere
         )

Original file line number	Diff line number	Diff line change
`@@ -280,7 +280,7 @@ def __init__(`
`280`	`280`
`281`	`281`	`super().__init__(`
`282`	`282`	`algorithm="nvidia_awq",`
`283`		`- quant_format=quant_format,`
	`283`	`+ quant_format=QuantFormat.QDQ,`
`284`	`284`	`op_types_to_quantize=None, # Assuming op_types_to_quantize is handled elsewhere`
`285`	`285`	`quant_axes=None, # Assuming quant_axes is handled elsewhere`
`286`	`286`	`)`