update default values for weight quatization (#1564)

microsoft · Aug 6, 2019 · 16087f3 · 16087f3
1 parent 7ee8aca
commit 16087f3
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git a/onnxruntime/python/tools/quantization/README.md b/onnxruntime/python/tools/quantization/README.md
@@ -62,7 +62,7 @@ onnx.save(quantized_model, 'path/to/the/quantized_model.onnx')
 See below for a description of all the options to quantize():
 
 - **model**: ModelProto to quantize
-- **per_channel**: *default: True*
+- **per_channel**: *default: False*
     If True, weights of Conv nodes are quantized per output channel.
     If False, they are quantized per tensor. Refer [QLinearConv](https://github.com/onnx/onnx/blob/master/docs/Operators.md#qlinearconv) for more information.
 - **nbits**: *default: 8*

diff --git a/onnxruntime/python/tools/quantization/quantize.py b/onnxruntime/python/tools/quantization/quantize.py
@@ -990,7 +990,7 @@ def _quantize_matmul(self, node, new_nodes_list):
         return [node]
 
 
-def quantize(model, per_channel=True, nbits=8, quantization_mode=QuantizationMode.IntegerOps,
+def quantize(model, per_channel=False, nbits=8, quantization_mode=QuantizationMode.IntegerOps,
     static=False, asymmetric_input_types=False, input_quantization_params=None, output_quantization_params=None):
     '''
         Given an onnx model, create a quantized onnx model and save it into a file