From 0ae07549e0746fbda8816d59e0921a3e3b0ec089 Mon Sep 17 00:00:00 2001
From: Guy Jacob <guy.jacob@intel.com>
Date: Tue, 12 Feb 2019 17:53:41 +0200
Subject: [PATCH] Post-train quant bugfix: squeeze scale factor shape when
quantizing bias
---
distiller/quantization/range_linear.py | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/distiller/quantization/range_linear.py b/distiller/quantization/range_linear.py
index c4e797f..234bfe8 100644
--- a/distiller/quantization/range_linear.py
+++ b/distiller/quantization/range_linear.py
@@ -359,7 +359,7 @@ class RangeLinearQuantParamLayerWrapper(RangeLinearQuantWrapper):
self.has_bias = hasattr(wrapped_module, 'bias') and wrapped_module.bias is not None
if self.has_bias:
if self.preset_act_stats:
- linear_quantize_clamp(wrapped_module.bias.data, self.accum_scale, 0,
+ linear_quantize_clamp(wrapped_module.bias.data, self.accum_scale.squeeze(), 0,
self.accum_min_q_val, self.accum_max_q_val, inplace=True)
else:
b_scale, b_zero_point = _get_quant_params_from_tensor(wrapped_module.bias, num_bits_params, self.mode)
@@ -387,7 +387,7 @@ class RangeLinearQuantParamLayerWrapper(RangeLinearQuantWrapper):
if self.has_bias:
# Re-quantize bias to match x * w scale: b_q' = (in_scale * w_scale / b_scale) * (b_q + b_zero_point)
self.wrapped_module.bias.data = linear_quantize_clamp(self.base_b_q + self.b_zero_point,
- self.accum_scale / self.b_scale, 0,
+ self.accum_scale.squeeze() / self.b_scale, 0,
self.accum_min_q_val, self.accum_max_q_val)
# Note the main terms within the summation is:
--
GitLab