OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2025-12-06 12:20:52 +01:00
Code Issues Actions 38 Packages Projects Releases Wiki Activity
gh/ezyang/628/base
pytorch/test/test_fake_quant.py
Gao, Xiang 45e4b614d1 Per channel quantization performance improvement (#33772) 
Summary:
Benchmark:
NVIDIA GTX 1650 + AMD Ryzen Threadripper 3970X
```python
import torch
print(torch.__version__)

for i in range(1000):
    torch.randn(1024 * 128, device='cuda')

def cuda(e):
    a = torch.randn(2 ** e, 32, device='cuda')
    s = torch.randn(32, device='cuda')
    z = torch.randn(32, device='cuda')
    torch.cuda.synchronize()
    %timeit torch.fake_quantize_per_channel_affine(a, s, z, 1, -999, 999); torch.cuda.synchronize()

def cpu(e):
    a = torch.randn(2 ** e, 32, device='cpu')
    s = torch.randn(32, device='cpu')
    z = torch.randn(32, device='cpu')
    %timeit torch.fake_quantize_per_channel_affine(a, s, z, 1, -999, 999);

for i in range(10, 24):
    cuda(i)
print()
for i in range(10, 32):
    cpu(i)
```
Before
```
1.5.0a0+9bc922d
849 µs ± 44.8 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
817 µs ± 30.4 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
814 µs ± 2.93 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
1.11 ms ± 1.32 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
1.19 ms ± 4.19 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
1.6 ms ± 5.58 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
2.44 ms ± 14.1 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
4.14 ms ± 2.55 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
7.41 ms ± 2.46 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
13.9 ms ± 2.3 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
26.9 ms ± 254 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
52.6 ms ± 260 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
104 ms ± 176 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
207 ms ± 1.24 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

249 µs ± 158 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each)
420 µs ± 230 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each)
766 µs ± 391 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each)
1.45 ms ± 574 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each)
2.84 ms ± 34.6 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
5.69 ms ± 83 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
7.29 ms ± 2.58 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
7.32 ms ± 13.5 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
17.4 ms ± 38.6 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
47.5 ms ± 264 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
187 ms ± 1.19 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
379 ms ± 5.05 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
652 ms ± 11.4 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
1.22 s ± 4.58 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
2.34 s ± 8.77 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
4.56 s ± 7.15 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
8.97 s ± 33.6 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
17.8 s ± 32.3 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
35.2 s ± 167 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
```
After
```
1.5.0a0+a7ec8cc
92.5 µs ± 2.03 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)
97.7 µs ± 469 ns per loop (mean ± std. dev. of 7 runs, 10000 loops each)
109 µs ± 4.73 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)
119 µs ± 6.17 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)
146 µs ± 1.84 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)
211 µs ± 2.45 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
347 µs ± 4.18 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
624 µs ± 14.9 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
1.17 ms ± 16.8 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
2.25 ms ± 48.3 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
4.43 ms ± 220 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
8.51 ms ± 44.3 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
16.9 ms ± 30.2 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
33.7 ms ± 7.64 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

201 µs ± 234 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each)
285 µs ± 465 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each)
287 µs ± 214 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each)
287 µs ± 221 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each)
287 µs ± 761 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each)
347 µs ± 399 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each)
675 µs ± 213 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each)
1.34 ms ± 643 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each)
4.82 ms ± 34.7 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
10.7 ms ± 88.5 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
20.3 ms ± 25.6 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
39.4 ms ± 242 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
78.8 ms ± 2 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
153 ms ± 786 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
285 ms ± 911 µs per loop (mean ± std. dev. of 7 runs, 1 loop each)
541 ms ± 1.09 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
1.03 s ± 1.67 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
1.97 s ± 8.59 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
3.81 s ± 10.2 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
```

Fixes https://github.com/pytorch/pytorch/issues/33647
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33772

Differential Revision: D20112531

Pulled By: ngimel

fbshipit-source-id: f90e3ef1b5be8276851637f3e1251cb8f1af411f
2020-02-26 10:19:25 -08:00

315 lines
15 KiB
Python
Raw Permalink Blame History

import torch
import torch.cuda
import torch.jit
import numpy as np
from hypothesis import given
from hypothesis import strategies as st
import torch.testing._internal.hypothesis_utils as hu
hu.assert_deadline_disabled()
from torch.testing._internal.common_utils import run_tests, TestCase
from torch.quantization import FakeQuantize
from torch.quantization import default_observer, default_per_channel_weight_observer
import io
import unittest
# Reference method for fake quantize
def _fake_quantize_per_tensor_affine_reference(X, scale, zero_point, quant_min, quant_max):
res = (torch.clamp(torch.round(X * (1.0 / scale) + zero_point), quant_min, quant_max) - zero_point) * scale
return res
# Reference method for the gradient of the fake quantize operator
def _fake_quantize_per_tensor_affine_grad_reference(dY, X, scale, zero_point, quant_min, quant_max):
Xq = torch.round(X * (1.0 / scale) + zero_point)
mask = (Xq >= quant_min) * (Xq <= quant_max)
res = torch.zeros_like(dY)
res[mask] = dY[mask]
return res
# Helper function used to simulate per-channel fake-quant against any axis
def _permute_to_axis_zero(X, axis):
new_axis_list = list(range(X.dim()))
new_axis_list[axis] = 0
new_axis_list[0] = axis
y = X.permute(tuple(new_axis_list))
return y, new_axis_list
# Reference method for fake quantize
def _fake_quantize_per_channel_affine_reference(X, per_channel_scale, per_channel_zero_point, axis, quant_min, quant_max):
X, permute_axis_list = _permute_to_axis_zero(X, axis)
res = torch.zeros_like(X)
for i in range(X.size()[0]):
res[i] = (torch.clamp(torch.round(X[i] * (1.0 / per_channel_scale[i]) +
per_channel_zero_point[i]), quant_min, quant_max) - per_channel_zero_point[i]) * per_channel_scale[i]
out = res.permute(tuple(permute_axis_list))
return out
# Reference method for the gradient of the fake quantize operator
def _fake_quantize_per_channel_affine_grad_reference(dY, X, per_channel_scale, per_channel_zero_point, axis, quant_min, quant_max):
X, permute_axis_list = _permute_to_axis_zero(X, axis)
Xq = torch.zeros_like(X)
for i in range(X.size()[0]):
Xq[i] = torch.round(X[i] * (1.0 / per_channel_scale[i]) + per_channel_zero_point[i])
Xq = Xq.permute(tuple(permute_axis_list))
mask = (Xq >= quant_min) * (Xq <= quant_max)
res = torch.zeros_like(dY)
res[mask] = dY[mask]
return res
def to_tensor(X, device):
return torch.tensor(X).to(device=torch.device(device), dtype=torch.float32)
NP_RANDOM_SEED = 19
tolerance = 1e-6
class TestFakeQuantizePerTensor(TestCase):
@given(device=st.sampled_from(['cpu', 'cuda'] if torch.cuda.is_available() else ['cpu']),
X=hu.tensor(shapes=hu.array_shapes(1, 5,),
qparams=hu.qparams(dtypes=torch.quint8)))
def test_forward_per_tensor(self, device, X):
r"""Tests the forward path of the FakeQuantizePerTensorAffine op.
"""
np.random.seed(NP_RANDOM_SEED)
X, (scale, zero_point, torch_type) = X
quant_min = torch.iinfo(torch_type).min
quant_max = torch.iinfo(torch_type).max
X = to_tensor(X, device)
Y = _fake_quantize_per_tensor_affine_reference(X.cpu(), scale, zero_point, quant_min, quant_max)
Y_prime = torch.fake_quantize_per_tensor_affine(
X, scale, zero_point, quant_min, quant_max)
np.testing.assert_allclose(Y, Y_prime.cpu(), rtol=tolerance, atol=tolerance)
@given(device=st.sampled_from(['cpu', 'cuda'] if torch.cuda.is_available() else ['cpu']),
X=hu.tensor(shapes=hu.array_shapes(1, 5,),
qparams=hu.qparams(dtypes=torch.quint8)))
@unittest.skip("temporarily disable the test")
def test_backward_per_tensor(self, device, X):
r"""Tests the backward method.
"""
np.random.seed(NP_RANDOM_SEED)
X, (scale, zero_point, torch_type) = X
quant_min = torch.iinfo(torch_type).min
quant_max = torch.iinfo(torch_type).max
X = to_tensor(X, device)
X.requires_grad_()
Y = _fake_quantize_per_tensor_affine_reference(X.cpu(), scale, zero_point, quant_min, quant_max)
Y_prime = torch.fake_quantize_per_tensor_affine(
X, scale, zero_point, quant_min, quant_max)
dout = torch.rand(X.shape, dtype=torch.float).to(device)
dX = _fake_quantize_per_tensor_affine_grad_reference(
dout, X, scale, zero_point, quant_min, quant_max)
Y_prime.backward(dout)
np.testing.assert_allclose(dX.cpu(), X.grad.cpu().detach().numpy(), rtol=tolerance, atol=tolerance)
@given(device=st.sampled_from(['cpu', 'cuda'] if torch.cuda.is_available() else ['cpu']),
X=hu.tensor(shapes=hu.array_shapes(1, 5,),
qparams=hu.qparams(dtypes=torch.quint8)))
# https://github.com/pytorch/pytorch/issues/30604
@unittest.skip("temporarily disable the test")
def test_numerical_consistency_per_tensor(self, device, X):
r"""Comparing numerical consistency between CPU quantize/dequantize op and the CPU fake quantize op
"""
np.random.seed(NP_RANDOM_SEED)
X, (scale, zero_point, torch_type) = X
quant_min = torch.iinfo(torch_type).min
quant_max = torch.iinfo(torch_type).max
X = to_tensor(X, device)
# quantize_per_tensor and dequantize are only implemented in CPU
Y = torch.dequantize(torch.quantize_per_tensor(X.cpu(), scale, zero_point, torch_type))
Y_prime = torch.fake_quantize_per_tensor_affine(
X, scale, zero_point, quant_min, quant_max)
np.testing.assert_allclose(Y, Y_prime.cpu(), rtol=tolerance, atol=tolerance)
@given(device=st.sampled_from(['cpu', 'cuda'] if torch.cuda.is_available() else ['cpu']),
X=hu.tensor(shapes=hu.array_shapes(1, 5,),
qparams=hu.qparams(dtypes=[torch.quint8])),
)
def test_fq_module(self, device, X):
np.random.seed(NP_RANDOM_SEED)
X, (scale, zero_point, torch_type) = X
quant_min = torch.iinfo(torch_type).min
quant_max = torch.iinfo(torch_type).max
X = to_tensor(X, device)
X.requires_grad_()
fq_module = torch.quantization.default_fake_quant().to(device)
Y_prime = fq_module(X)
assert fq_module.scale is not None
assert fq_module.zero_point is not None
Y = _fake_quantize_per_tensor_affine_reference(X, fq_module.scale, fq_module.zero_point, quant_min, quant_max)
np.testing.assert_allclose(Y.cpu().detach().numpy(), Y_prime.cpu().detach().numpy(), rtol=tolerance, atol=tolerance)
# Test backward
dout = torch.rand(X.shape, dtype=torch.float, device=device)
Y_prime.backward(dout)
dX = _fake_quantize_per_tensor_affine_grad_reference(dout, X, fq_module.scale, fq_module.zero_point, quant_min, quant_max)
np.testing.assert_allclose(dX.cpu().numpy(), X.grad.cpu().detach().numpy(), rtol=tolerance, atol=tolerance)
def test_fq_serializable(self):
observer = default_observer
quant_min = 0
quant_max = 255
fq_module = FakeQuantize(observer, quant_min, quant_max)
X = torch.tensor([-5, -3.5, -2, 0, 3, 5, 7], dtype=torch.float32)
y_ref = fq_module(X)
state_dict = fq_module.state_dict()
self.assertEqual(state_dict['scale'], 0.094488)
self.assertEqual(state_dict['zero_point'], 53)
b = io.BytesIO()
torch.save(state_dict, b)
b.seek(0)
loaded_dict = torch.load(b)
loaded_fq_module = FakeQuantize(observer, quant_min, quant_max)
loaded_fq_module.load_state_dict(loaded_dict)
for key in state_dict:
self.assertEqual(state_dict[key], loaded_fq_module.state_dict()[key])
self.assertEqual(loaded_fq_module.calculate_qparams(), fq_module.calculate_qparams())
def test_fake_quant_control(self):
torch.manual_seed(42)
X = torch.rand(20, 10, dtype=torch.float32)
fq_module = torch.quantization.default_fake_quant()
# Output of fake quant is not identical to input
Y = fq_module(X)
self.assertNotEqual(Y, X)
torch.quantization.disable_fake_quant(fq_module)
X = torch.rand(20, 10, dtype=torch.float32)
Y = fq_module(X)
# Fake quant is disabled,output is identical to input
self.assertEqual(Y, X)
scale = fq_module.scale
zero_point = fq_module.zero_point
torch.quantization.disable_observer(fq_module)
torch.quantization.enable_fake_quant(fq_module)
X = 10.0 * torch.rand(20, 10, dtype=torch.float32) - 5.0
Y = fq_module(X)
self.assertNotEqual(Y, X)
# Observer is disabled, scale and zero-point do not change
self.assertEqual(fq_module.scale, scale)
self.assertEqual(fq_module.zero_point, zero_point)
torch.quantization.enable_observer(fq_module)
Y = fq_module(X)
self.assertNotEqual(Y, X)
# Observer is enabled, scale and zero-point are different
self.assertNotEqual(fq_module.scale, scale)
self.assertNotEqual(fq_module.zero_point, zero_point)
class TestFakeQuantizePerChannel(TestCase):
@given(device=st.sampled_from(['cpu', 'cuda'] if torch.cuda.is_available() else ['cpu']),
X=hu.per_channel_tensor(shapes=hu.array_shapes(1, 5,),
qparams=hu.qparams(dtypes=torch.quint8)))
def test_forward_per_channel(self, device, X):
r"""Tests the forward path of the FakeQuantizePerTensorAffine op.
"""
np.random.seed(NP_RANDOM_SEED)
X, (scale, zero_point, axis, torch_type) = X
quant_min = torch.iinfo(torch_type).min
quant_max = torch.iinfo(torch_type).max
X = to_tensor(X, device)
scale = to_tensor(scale, device)
zero_point = torch.tensor(zero_point).to(dtype=torch.int64, device=device)
Y = _fake_quantize_per_channel_affine_reference(X.cpu(), scale.cpu(), zero_point.cpu(), axis, quant_min, quant_max)
Y_prime = torch.fake_quantize_per_channel_affine(
X, scale, zero_point, axis, quant_min, quant_max)
np.testing.assert_allclose(Y, Y_prime.cpu(), rtol=tolerance, atol=tolerance)
@given(device=st.sampled_from(['cpu', 'cuda'] if torch.cuda.is_available() else ['cpu']),
X=hu.per_channel_tensor(shapes=hu.array_shapes(1, 5,),
qparams=hu.qparams(dtypes=torch.quint8)))
def test_backward_per_channel(self, device, X):
r"""Tests the backward method.
"""
np.random.seed(NP_RANDOM_SEED)
X, (scale, zero_point, axis, torch_type) = X
quant_min = torch.iinfo(torch_type).min
quant_max = torch.iinfo(torch_type).max
X = to_tensor(X, device)
scale = to_tensor(scale, device)
zero_point = torch.tensor(zero_point).to(dtype=torch.int64, device=device)
X.requires_grad_()
Y_prime = torch.fake_quantize_per_channel_affine(
X, scale, zero_point, axis, quant_min, quant_max)
dout = torch.rand(X.shape, dtype=torch.float).to(device)
dX = _fake_quantize_per_channel_affine_grad_reference(
dout, X, scale, zero_point, axis, quant_min, quant_max)
Y_prime.backward(dout)
np.testing.assert_allclose(dX.cpu().detach().numpy(), X.grad.cpu().detach().numpy(), rtol=tolerance, atol=tolerance)
@given(device=st.sampled_from(['cpu', 'cuda'] if torch.cuda.is_available() else ['cpu']),
X=hu.per_channel_tensor(shapes=hu.array_shapes(1, 5,),
qparams=hu.qparams(dtypes=torch.quint8)))
@unittest.skip("temporarily disable the test")
def test_numerical_consistency_per_channel(self, device, X):
r"""Comparing numerical consistency between CPU quantize/dequantize op and the CPU fake quantize op
"""
np.random.seed(NP_RANDOM_SEED)
X, (scale, zero_point, axis, torch_type) = X
quant_min = torch.iinfo(torch_type).min
quant_max = torch.iinfo(torch_type).max
X = to_tensor(X, device)
scale = to_tensor(scale, device)
zero_point = torch.tensor(zero_point).to(dtype=torch.int64, device=device)
# quantize_linear and dequantize are only implemented in CPU
Y = torch.dequantize(torch.quantize_per_channel(X.cpu(), scale.cpu(), zero_point.cpu(), axis, torch_type))
Y_prime = torch.fake_quantize_per_channel_affine(
X, scale, zero_point, axis, quant_min, quant_max)
np.testing.assert_allclose(Y, Y_prime.cpu(), rtol=tolerance, atol=tolerance)
@given(device=st.sampled_from(['cpu', 'cuda'] if torch.cuda.is_available() else ['cpu']),
X=hu.per_channel_tensor(shapes=hu.array_shapes(2, 5,),
qparams=hu.qparams(dtypes=torch.qint8)))
def test_fq_module(self, device, X):
np.random.seed(NP_RANDOM_SEED)
X, (scale, zero_point, axis, torch_type) = X
quant_min = torch.iinfo(torch_type).min
quant_max = torch.iinfo(torch_type).max
X = to_tensor(X, device)
X.requires_grad_()
fq_module = FakeQuantize(default_per_channel_weight_observer, quant_min, quant_max, ch_axis=axis).to(device)
Y_prime = fq_module(X)
assert fq_module.scale is not None
assert fq_module.zero_point is not None
Y = _fake_quantize_per_channel_affine_reference(X, fq_module.scale,
fq_module.zero_point, axis, quant_min, quant_max)
np.testing.assert_allclose(Y.cpu().detach().numpy(), Y_prime.cpu().detach().numpy(), rtol=tolerance, atol=tolerance)
# Test backward
dout = torch.rand(X.shape, dtype=torch.float, device=device)
Y_prime.backward(dout)
dX = _fake_quantize_per_channel_affine_grad_reference(dout, X, fq_module.scale,
fq_module.zero_point, axis, quant_min, quant_max)
np.testing.assert_allclose(dX.cpu().numpy(), X.grad.cpu().detach().numpy(), rtol=tolerance, atol=tolerance)
def test_fq_serializable(self):
observer = default_per_channel_weight_observer
quant_min = -128
quant_max = 127
fq_module = FakeQuantize(observer, quant_min, quant_max)
X = torch.tensor([[-5, -3.5, -2, 0, 3, 5, 7], [1, 3, 2, 5, 6.5, 8, 10]], dtype=torch.float32)
y_ref = fq_module(X)
state_dict = fq_module.state_dict()
self.assertEqual(state_dict['scale'], [0.054902, 0.078431])
self.assertEqual(state_dict['zero_point'], [0, 0])
b = io.BytesIO()
torch.save(state_dict, b)
b.seek(0)
loaded_dict = torch.load(b)
for key in state_dict:
self.assertEqual(state_dict[key], loaded_dict[key])
if __name__ == '__main__':
run_tests()
Reference in New Issue View Git Blame Copy Permalink