Skip to content

Commit

Permalink
[luci/pass] Fix typo and rename functions in QuantizeWeightsWithGPTQPass
Browse files Browse the repository at this point in the history 
This commit renames the function 'fake_quantize_weights_basic' to 'fake_quantize' and 'fake_quantize_weights_with_gptq' to 'fake_quantize_with_gptq' for consistency.

ONE-DCO-1.0-Signed-off-by: y01000.you <[email protected]>
  • Loading branch information
y01000.you committed Dec 6, 2024
1 parent ae82151 commit e3aa0d7
Showing 1 changed file with 22 additions and 23 deletions.
45 changes: 22 additions & 23 deletions compiler/luci/pass/src/QuantizeWeightsWithGPTQPass.cpp
View file
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
/*
* Copyright (c) 2024 Samsung Electronics Co., Ltd. All Rights Reserved
* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
* Copyright (c) 2024 Samsung Electronics Co., Ltd. All Rights Reserved.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
Expand Down Expand Up @@ -93,7 +93,7 @@ size_t calculate_qauntized_value(CircleConst *node, uint32_t *indices, loco::Ten
return static_cast<int32_t>(std::round((data_clipped - min[idx_channel]) * scaling_factor_inv));
}

void apply_damping_to_hessian(std::vector<float> &hessian, uint32_t num_size)
void apply_dampening_to_hessian(std::vector<float> &hessian, uint32_t num_size)
{
float damp = 0;
float percdamp = .01;
Expand Down Expand Up @@ -352,10 +352,10 @@ void transpose_to_upper_triangular(std::vector<float> &matrix, uint32_t num_size
}
}

void asymmetric_wquant_per_channel_basic(CircleConst *node, std::vector<float> &min,
std::vector<float> &max, std::vector<float> &scaling_factor,
std::vector<int64_t> &zp, std::vector<float> &nudged_min,
std::vector<float> &nudged_max, loco::DataType output_type) const
void asymmetric_wquant_per_channel(CircleConst *node, std::vector<float> &min,
std::vector<float> &max, std::vector<float> &scaling_factor,
std::vector<int64_t> &zp, std::vector<float> &nudged_min,
std::vector<float> &nudged_max, loco::DataType output_type)
{
assert(node->dtype() == loco::DataType::FLOAT32);
assert(output_type == loco::DataType::U8 || output_type == loco::DataType::U4);
Expand Down Expand Up @@ -388,11 +388,10 @@ void asymmetric_wquant_per_channel_basic(CircleConst *node, std::vector<float> &
}
}

void asymmetric_wquant_per_channel_with_gptq(CircleConst *node, std::vector<float> &min,
std::vector<float> &max, std::vector<float> &scaling_factor,
std::vector<int64_t> &zp, std::vector<float> &nudged_min,
std::vector<float> &nudged_max, std::vector<float> &hessian,
loco::DataType output_type) const
void asymmetric_wquant_per_channel_with_gptq(
CircleConst *node, std::vector<float> &min, std::vector<float> &max,
std::vector<float> &scaling_factor, std::vector<int64_t> &zp, std::vector<float> &nudged_min,
std::vector<float> &nudged_max, std::vector<float> &hessian, loco::DataType output_type)
{
assert(node->dtype() == loco::DataType::FLOAT32);
assert(output_type == loco::DataType::U8 || output_type == loco::DataType::U4);
Expand All @@ -414,7 +413,7 @@ void asymmetric_wquant_per_channel_with_gptq(CircleConst *node, std::vector<floa
uint32_t size_hessian = static_cast<uint32_t>(sqrt(hessian.size()));

// Calculate hessian inverse
apply_damping_to_hessian(hessian, size_hessian);
apply_dampening_to_hessian(hessian, size_hessian);
cholesky_decomposition(hessian, size_hessian);
cholesky_inverse(hessian, size_hessian);
cholesky_decomposition(hessian, size_hessian);
Expand Down Expand Up @@ -544,7 +543,7 @@ class QuantizeWeightsWithGPTQ final : public luci::CircleNodeMutableVisitor<void
QuantizationGranularity _granularity;
std::unordered_map<const luci::CircleNode *, std::vector<float>> *_hessian_map;

void fake_quantize_weights_basic(luci::CircleConst *weights) const
void fake_quantize(luci::CircleConst *weights) const
{
if (_granularity != luci::QuantizationGranularity::ChannelWise)
{
Expand All @@ -567,8 +566,8 @@ class QuantizeWeightsWithGPTQ final : public luci::CircleNodeMutableVisitor<void
std::vector<float> scaling_factor(min.size());
std::vector<int64_t> zp(min.size());

asymmetric_wquant_per_channel_basic(weights, min, max, scaling_factor, zp, nudged_min, nudged_max,
_output_type);
asymmetric_wquant_per_channel(weights, min, max, scaling_factor, zp, nudged_min, nudged_max,
_output_type);
asymmetric_wdequant_per_channel(weights, scaling_factor, nudged_min);

auto quantparam = std::make_unique<CircleQuantParam>();
Expand All @@ -580,7 +579,7 @@ class QuantizeWeightsWithGPTQ final : public luci::CircleNodeMutableVisitor<void
weights->quantparam(std::move(quantparam));
}

void fake_quantize_weights_with_gptq(luci::CircleConst *weights, std::vector<float> &hessian) const
void fake_quantize_with_gptq(luci::CircleConst *weights, std::vector<float> &hessian) const
{
if (_granularity != luci::QuantizationGranularity::ChannelWise)
{
Expand All @@ -603,8 +602,8 @@ class QuantizeWeightsWithGPTQ final : public luci::CircleNodeMutableVisitor<void
std::vector<float> scaling_factor(min.size());
std::vector<int64_t> zp(min.size());

asymmetric_wquant_per_channel_with_gptq(weights, min, max, scaling_factor, zp, nudged_min, nudged_max,
hessian, _output_type);
asymmetric_wquant_per_channel_with_gptq(weights, min, max, scaling_factor, zp, nudged_min,
nudged_max, hessian, _output_type);
asymmetric_wdequant_per_channel(weights, scaling_factor, nudged_min);

auto quantparam = std::make_unique<CircleQuantParam>();
Expand Down Expand Up @@ -650,7 +649,7 @@ class QuantizeWeightsWithGPTQ final : public luci::CircleNodeMutableVisitor<void

auto hessian = (*_hessian_map)[node];

fake_quantize_weights_with_gptq(new_weights, hessian);
fake_quantize_with_gptq(new_weights, hessian);
}

void visit(luci::CircleDepthwiseConv2D *node)
Expand All @@ -665,7 +664,7 @@ class QuantizeWeightsWithGPTQ final : public luci::CircleNodeMutableVisitor<void
auto new_weights = luci::clone(weights);
node->filter(new_weights);

fake_quantize_weights_basic(new_weights);
fake_quantize(new_weights);
}

void visit(luci::CircleTransposeConv *node)
Expand All @@ -680,7 +679,7 @@ class QuantizeWeightsWithGPTQ final : public luci::CircleNodeMutableVisitor<void
auto new_weights = luci::clone(weights);
node->filter(new_weights);

fake_quantize_weights_basic(new_weights);
fake_quantize(new_weights);
}

void visit(luci::CircleFullyConnected *node)
Expand All @@ -696,7 +695,7 @@ class QuantizeWeightsWithGPTQ final : public luci::CircleNodeMutableVisitor<void

auto hessian = (*_hessian_map)[node];

fake_quantize_weights_with_gptq(new_weights, hessian);
fake_quantize_with_gptq(new_weights, hessian);
}
};

Expand Down

0 comments on commit e3aa0d7

Please sign in to comment.