From 16087f313360d81d8a3c208101f7b5297d0c4315 Mon Sep 17 00:00:00 2001 From: Ashwini Khade Date: Mon, 5 Aug 2019 21:39:37 -0700 Subject: [PATCH] update default values for weight quatization (#1564) --- onnxruntime/python/tools/quantization/README.md | 2 +- onnxruntime/python/tools/quantization/quantize.py | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/onnxruntime/python/tools/quantization/README.md b/onnxruntime/python/tools/quantization/README.md index 19c8e2cd55c64..6f0ccfe48e45e 100644 --- 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 index 3bf180e84c3cf..deb4dddf33b45 100644 --- 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