151 lines
5.0 KiB
Python
151 lines
5.0 KiB
Python
# Copyright (c) 2025 PaddlePaddle Authors. 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
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
|
|
import unittest
|
|
|
|
import numpy as np
|
|
|
|
import paddle
|
|
from paddle import _legacy_C_ops
|
|
|
|
|
|
class TestQuantizeLinerAPI(unittest.TestCase):
|
|
"""
|
|
test for quantize_linear and dequantize_linear
|
|
"""
|
|
|
|
def setUp(self):
|
|
np.random.seed(2025)
|
|
paddle.disable_static()
|
|
|
|
def run_case(self, function_name, xshape, axis, bit_length, qmin, qmax):
|
|
func = getattr(_legacy_C_ops, function_name, None)
|
|
if func is None:
|
|
raise ValueError(
|
|
f"No function named '{function_name}' found in _legacy_C_ops."
|
|
)
|
|
|
|
x_np = np.random.uniform(-0.1, 0.1, xshape).astype("float32")
|
|
|
|
x_paddle = paddle.to_tensor(
|
|
x_np, dtype="float32", place=paddle.XPUPlace(0)
|
|
)
|
|
x_paddle_cpu = paddle.to_tensor(
|
|
x_np, dtype="float32", place=paddle.CPUPlace()
|
|
)
|
|
|
|
zero_paddle = paddle.to_tensor(
|
|
[0], dtype="float32", place=paddle.XPUPlace(0)
|
|
)
|
|
zero_paddle_cpu = paddle.to_tensor(
|
|
[0], dtype="float32", place=paddle.CPUPlace()
|
|
)
|
|
|
|
if axis == -1:
|
|
scale_paddle = paddle.to_tensor(
|
|
[0.5], dtype="float32", place=paddle.XPUPlace(0)
|
|
)
|
|
scale_paddle_cpu = paddle.to_tensor(
|
|
[0.5], dtype="float32", place=paddle.CPUPlace()
|
|
)
|
|
elif axis == 0:
|
|
scale_np = np.random.uniform(-0.1, 0.1, xshape[0]).astype("float32")
|
|
scale_paddle = paddle.to_tensor(
|
|
scale_np, dtype="float32", place=paddle.XPUPlace(0)
|
|
)
|
|
scale_paddle_cpu = paddle.to_tensor(
|
|
scale_np, dtype="float32", place=paddle.CPUPlace()
|
|
)
|
|
elif axis == 1:
|
|
scale_np = np.random.uniform(-0.1, 0.1, xshape[1]).astype("float32")
|
|
scale_paddle = paddle.to_tensor(
|
|
scale_np, dtype="float32", place=paddle.XPUPlace(0)
|
|
)
|
|
scale_paddle_cpu = paddle.to_tensor(
|
|
scale_np, dtype="float32", place=paddle.CPUPlace()
|
|
)
|
|
else:
|
|
raise AssertionError(
|
|
"quant axis other than -1, 0, 1 is not supported in XPU"
|
|
)
|
|
|
|
paddle.set_device("xpu")
|
|
y_xpu = func(
|
|
x_paddle,
|
|
scale_paddle,
|
|
zero_paddle,
|
|
"quant_axis",
|
|
axis,
|
|
"bit_length",
|
|
bit_length,
|
|
"qmin",
|
|
qmin,
|
|
"qmax",
|
|
qmax,
|
|
)
|
|
paddle.set_device("cpu")
|
|
y_cpu = func(
|
|
x_paddle_cpu,
|
|
scale_paddle_cpu,
|
|
zero_paddle_cpu,
|
|
"quant_axis",
|
|
axis,
|
|
"bit_length",
|
|
bit_length,
|
|
"qmin",
|
|
qmin,
|
|
"qmax",
|
|
qmax,
|
|
)
|
|
np.testing.assert_allclose(y_xpu.numpy(), y_cpu.numpy(), atol=0, rtol=0)
|
|
|
|
def test_quantize(self):
|
|
for axis in [-1, 0, 1]:
|
|
self.run_case("quantize_linear", [3, 5], axis, 4, -8, 7)
|
|
self.run_case("quantize_linear", [10, 12], axis, 4, -8, 7)
|
|
self.run_case("quantize_linear", [10, 12], axis, 8, -128, 127)
|
|
self.run_case("quantize_linear", [10, 12, 15], axis, 4, -8, 7)
|
|
self.run_case("quantize_linear", [10, 12, 15], axis, 8, -128, 127)
|
|
|
|
def test_dequantize(self):
|
|
for axis in [-1, 0, 1]:
|
|
self.run_case("dequantize_linear", [3, 5], axis, 4, -8, 7)
|
|
self.run_case("dequantize_linear", [10, 12], axis, 4, -8, 7)
|
|
self.run_case("dequantize_linear", [10, 12], axis, 8, -128, 127)
|
|
self.run_case("dequantize_linear", [10, 12, 15], axis, 4, -8, 7)
|
|
self.run_case("dequantize_linear", [10, 12, 15], axis, 8, -128, 127)
|
|
|
|
def test_weight_only_linear_empty_batch_xpu(self):
|
|
paddle.disable_static()
|
|
paddle.set_device("xpu")
|
|
|
|
x = paddle.empty([0, 1, 512], dtype="float16")
|
|
weight_int32 = paddle.randint(low=-128, high=127, shape=[1024, 512])
|
|
weight = paddle.cast(weight_int32, "int8")
|
|
bias = paddle.zeros([1024], dtype="float16")
|
|
weight_scale = paddle.ones([1024], dtype="float16")
|
|
|
|
out = paddle.nn.quant.weight_only_linear(
|
|
x,
|
|
weight=weight,
|
|
bias=bias,
|
|
weight_scale=weight_scale,
|
|
weight_dtype="int8",
|
|
)
|
|
self.assertEqual(list(out.shape), [0, 1024])
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|