wehub-resource-sync
1100 lines
34 KiB
Python
1100 lines
34 KiB
Python
# Copyright (c) 2023 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 copy
|
|
import math
|
|
import struct
|
|
import unittest
|
|
|
|
import numpy as np
|
|
|
|
import paddle
|
|
import paddle.nn.quant as Q
|
|
from paddle import base
|
|
from paddle.base import core
|
|
from paddle.framework import set_default_dtype
|
|
from paddle.pir_utils import IrGuard
|
|
|
|
np.random.seed(123)
|
|
paddle.seed(123)
|
|
|
|
|
|
def convert_uint16_to_float(in_list):
|
|
in_list = np.asarray(in_list)
|
|
out = np.vectorize(
|
|
lambda x: struct.unpack(
|
|
'<f', struct.pack('<I', np.uint32(x) << np.uint32(16))
|
|
)[0],
|
|
otypes=[np.float32],
|
|
)(in_list.flat)
|
|
return np.reshape(out, in_list.shape)
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda(),
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase(unittest.TestCase):
|
|
def config(self):
|
|
self.dtype = 'float16'
|
|
self.rtol = 1e-5
|
|
self.atol = 1e-2
|
|
self.bias = True
|
|
self.batch = 1
|
|
self.token = 32
|
|
self.in_features = 64
|
|
self.out_features = 256
|
|
self.weight_dtype = "int8"
|
|
self.static = False
|
|
self.group_size = -1
|
|
|
|
def weightQuantizeCPUGPUConsistenceCheck(self, weight_float):
|
|
for arch in [70, 75, 80, 86]:
|
|
weight_gpu, weight_scale_gpu = Q.weight_quantize(
|
|
(
|
|
weight_float.cuda()
|
|
if self.weight_dtype == "int8"
|
|
else self.weight.cpu()
|
|
),
|
|
algo=(
|
|
"weight_only_int8"
|
|
if self.weight_dtype == "int8"
|
|
else "weight_only_int4"
|
|
),
|
|
arch=arch,
|
|
group_size=self.group_size,
|
|
)
|
|
weight_cpu, weight_scale_cpu = Q.weight_quantize(
|
|
weight_float.cpu(),
|
|
algo=(
|
|
"weight_only_int8"
|
|
if self.weight_dtype == "int8"
|
|
else "weight_only_int4"
|
|
),
|
|
arch=arch,
|
|
group_size=self.group_size,
|
|
)
|
|
np.testing.assert_allclose(
|
|
weight_gpu.numpy(),
|
|
weight_cpu.numpy(),
|
|
atol=1.5,
|
|
rtol=2,
|
|
)
|
|
np.testing.assert_allclose(
|
|
weight_scale_gpu.numpy(),
|
|
weight_scale_cpu.numpy(),
|
|
atol=1e-5,
|
|
rtol=1e-3,
|
|
)
|
|
pass
|
|
pass
|
|
|
|
def setUp(self):
|
|
self.config()
|
|
if self.dtype == "bfloat16" or self.weight_dtype == "int4":
|
|
self.atol = 1.3e-1
|
|
x = np.random.random((self.batch, self.token, self.in_features))
|
|
self.x = paddle.to_tensor(x, dtype=self.dtype)
|
|
if self.bias:
|
|
bias_attr = base.ParamAttr(
|
|
trainable=False,
|
|
regularizer=None,
|
|
initializer=paddle.nn.initializer.Constant(value=1.0),
|
|
)
|
|
else:
|
|
bias_attr = None
|
|
set_default_dtype(self.dtype)
|
|
self.linear = paddle.nn.Linear(
|
|
self.in_features, self.out_features, bias_attr=bias_attr
|
|
)
|
|
|
|
self.bias = self.linear.bias
|
|
self.weight = self.linear.weight
|
|
self.float_weight = self.linear.weight
|
|
self.weight_scale = None
|
|
# check weight quantize
|
|
self.weightQuantizeCPUGPUConsistenceCheck(self.float_weight)
|
|
|
|
self.weight, self.weight_scale = Q.weight_quantize(
|
|
(
|
|
self.float_weight.cuda()
|
|
if self.weight_dtype == "int8"
|
|
else self.weight.cpu()
|
|
),
|
|
algo=(
|
|
"weight_only_int8"
|
|
if self.weight_dtype == "int8"
|
|
else "weight_only_int4"
|
|
),
|
|
group_size=self.group_size,
|
|
)
|
|
|
|
def get_linear_out(self):
|
|
out = self.linear(self.x)
|
|
return out.numpy()
|
|
|
|
def get_weight_only_linear_out(self):
|
|
out = Q.weight_only_linear(
|
|
self.x,
|
|
self.weight,
|
|
bias=self.bias,
|
|
weight_scale=self.weight_scale,
|
|
weight_dtype=self.weight_dtype,
|
|
group_size=self.group_size,
|
|
)
|
|
return out.numpy()
|
|
|
|
def test_weight_only_linear(self):
|
|
out_expect = self.get_linear_out()
|
|
out_real = self.get_weight_only_linear_out()
|
|
|
|
if self.dtype == "bfloat16":
|
|
out_real = convert_uint16_to_float(out_real)
|
|
out_expect = convert_uint16_to_float(out_expect)
|
|
np.testing.assert_allclose(
|
|
out_real, out_expect, rtol=self.rtol, atol=self.atol
|
|
)
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda(),
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase1(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'float16'
|
|
self.weight_dtype = "int8"
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda(),
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase2(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'float16'
|
|
self.bias = False
|
|
self.weight_dtype = "int8"
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8,
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase3(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'bfloat16'
|
|
self.weight_dtype = "int8"
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8
|
|
or not core.is_bfloat16_supported(core.CUDAPlace(0)),
|
|
"quantized_matmul requires CUDA_ARCH >= 8 or core is not support bfloat16",
|
|
)
|
|
class WeightOnlyLinearTestCase4(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'float16'
|
|
self.weight_dtype = "int4"
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda(),
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase5(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'float16'
|
|
self.bias = False
|
|
self.weight_dtype = "int4"
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8
|
|
or not core.is_bfloat16_supported(core.CUDAPlace(0)),
|
|
"quantized_matmul requires CUDA_ARCH >= 8 or core is not support bfloat16",
|
|
)
|
|
class WeightOnlyLinearTestCase6(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'bfloat16'
|
|
self.weight_dtype = "int4"
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda(),
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase7(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'float16'
|
|
self.weight_dtype = "int8"
|
|
self.batch = 1
|
|
self.token = 1
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda(),
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase8(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'float16'
|
|
self.weight_dtype = "int8"
|
|
self.bias = False
|
|
self.batch = 1
|
|
self.token = 1
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8,
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase9(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'bfloat16'
|
|
self.weight_dtype = "int8"
|
|
self.batch = 1
|
|
self.token = 1
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8,
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase10(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'bfloat16'
|
|
self.weight_dtype = "int8"
|
|
self.bias = False
|
|
self.batch = 1
|
|
self.token = 1
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda(),
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase11(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'float16'
|
|
self.weight_dtype = "int4"
|
|
self.batch = 1
|
|
self.token = 1
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda(),
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase12(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'float16'
|
|
self.weight_dtype = "int4"
|
|
self.bias = False
|
|
self.batch = 1
|
|
self.token = 1
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8
|
|
or not core.is_bfloat16_supported(core.CUDAPlace(0)),
|
|
"quantized_matmul requires CUDA_ARCH >= 8 or core is not support bfloat16",
|
|
)
|
|
class WeightOnlyLinearTestCase13(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'bfloat16'
|
|
self.weight_dtype = "int4"
|
|
self.bias = False
|
|
self.batch = 1
|
|
self.token = 1
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8
|
|
or not core.is_bfloat16_supported(core.CUDAPlace(0)),
|
|
"quantized_matmul requires CUDA_ARCH >= 8 or core is not support bfloat16",
|
|
)
|
|
class WeightOnlyLinearTestCase14(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'bfloat16'
|
|
self.weight_dtype = "int4"
|
|
self.bias = False
|
|
self.batch = 1
|
|
self.token = 1
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8
|
|
or not core.is_bfloat16_supported(core.CUDAPlace(0)),
|
|
"quantized_matmul requires CUDA_ARCH >= 8 or core is not support bfloat16",
|
|
)
|
|
class WeightOnlyLinearTestCase15(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'bfloat16'
|
|
self.weight_dtype = "int4"
|
|
self.bias = False
|
|
self.batch = 1
|
|
self.token = 1
|
|
self.group_size = 64
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8
|
|
or not core.is_bfloat16_supported(core.CUDAPlace(0)),
|
|
"quantized_matmul requires CUDA_ARCH >= 8 or core is not support bfloat16",
|
|
)
|
|
class WeightOnlyLinearTestCase16(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'bfloat16'
|
|
self.weight_dtype = "int4"
|
|
self.bias = False
|
|
self.batch = 1
|
|
self.token = 1
|
|
self.group_size = 128
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8,
|
|
"quantized_matmul groupwise mode need CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase17(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'float16'
|
|
self.weight_dtype = "int4"
|
|
self.bias = False
|
|
self.batch = 1
|
|
self.token = 1
|
|
self.group_size = 64
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8,
|
|
"quantized_matmul groupwise mode need CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase18(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'float16'
|
|
self.weight_dtype = "int4"
|
|
self.bias = False
|
|
self.batch = 1
|
|
self.token = 1
|
|
self.group_size = 128
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8
|
|
or not core.is_bfloat16_supported(core.CUDAPlace(0)),
|
|
"quantized_matmul requires CUDA_ARCH >= 8 or core is not support bfloat16",
|
|
)
|
|
class WeightOnlyLinearTestCase19(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'bfloat16'
|
|
self.weight_dtype = "int4"
|
|
self.bias = False
|
|
self.batch = 1
|
|
self.token = 2
|
|
self.group_size = 128
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8
|
|
or not core.is_bfloat16_supported(core.CUDAPlace(0)),
|
|
"quantized_matmul requires CUDA_ARCH >= 8 or core is not support bfloat16",
|
|
)
|
|
class WeightOnlyLinearTestCase20(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'bfloat16'
|
|
self.weight_dtype = "int8"
|
|
self.bias = False
|
|
self.batch = 1
|
|
self.token = 1
|
|
self.group_size = 64
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8
|
|
or not core.is_bfloat16_supported(core.CUDAPlace(0)),
|
|
"quantized_matmul requires CUDA_ARCH >= 8 or core is not support bfloat16",
|
|
)
|
|
class WeightOnlyLinearTestCase21(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'bfloat16'
|
|
self.weight_dtype = "int8"
|
|
self.bias = False
|
|
self.batch = 1
|
|
self.token = 1
|
|
self.group_size = 128
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda(),
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase22(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'float16'
|
|
self.weight_dtype = "int8"
|
|
self.in_features = 128
|
|
self.out_features = 288
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda(),
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase23(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'float16'
|
|
self.bias = False
|
|
self.weight_dtype = "int8"
|
|
self.in_features = 128
|
|
self.out_features = 288
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8,
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase24(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'bfloat16'
|
|
self.weight_dtype = "int8"
|
|
self.in_features = 128
|
|
self.out_features = 288
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8,
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase25(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'bfloat16'
|
|
self.weight_dtype = "int4"
|
|
self.group_size = 128
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8,
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase26(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'bfloat16'
|
|
self.weight_dtype = "int4"
|
|
self.group_size = 64
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8,
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase27(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'float16'
|
|
self.weight_dtype = "int4"
|
|
self.group_size = 128
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8,
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase28(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'bfloat16'
|
|
self.weight_dtype = "int4"
|
|
self.token = 300
|
|
self.group_size = 128
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8,
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCase29(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.dtype = 'bfloat16'
|
|
self.weight_dtype = "int8"
|
|
self.token = 300
|
|
self.group_size = 128
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8,
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearTestCaseStatic(WeightOnlyLinearTestCase):
|
|
def config(self):
|
|
super().config()
|
|
self.static = True
|
|
|
|
def get_weight_only_linear_out_static(self):
|
|
paddle.enable_static()
|
|
main = paddle.static.Program()
|
|
start = paddle.static.Program()
|
|
with paddle.static.program_guard(main, start):
|
|
x = paddle.static.data("x", self.x.shape, dtype=self.x.dtype)
|
|
|
|
weight = paddle.static.data(
|
|
"weight", self.weight.shape, dtype=self.weight.dtype
|
|
)
|
|
bias = paddle.static.data(
|
|
"bias", self.bias.shape, dtype=self.bias.dtype
|
|
)
|
|
x_np = self.x.numpy()
|
|
weight_np = self.weight.numpy()
|
|
bias_np = self.bias.numpy()
|
|
if self.weight_scale is not None:
|
|
weight_scale = paddle.static.data(
|
|
"weight_scale",
|
|
self.weight_scale.shape,
|
|
dtype=self.weight_scale.dtype,
|
|
)
|
|
weight_scale_np = self.weight_scale.numpy()
|
|
else:
|
|
weight_scale = None
|
|
weight_scale_np = None
|
|
|
|
out = Q.weight_only_linear(
|
|
x,
|
|
weight,
|
|
bias,
|
|
weight_scale,
|
|
self.weight_dtype,
|
|
group_size=self.group_size,
|
|
)
|
|
feed_dict = {
|
|
'x': x_np,
|
|
'weight': weight_np,
|
|
'bias': bias_np,
|
|
"weight_scale": weight_scale_np,
|
|
}
|
|
exe = base.Executor(paddle.CUDAPlace(0))
|
|
exe.run(start)
|
|
(out,) = exe.run(main, feed=feed_dict, fetch_list=[out])
|
|
paddle.disable_static()
|
|
return out
|
|
|
|
def test_weight_quantize_and_dequantize_pir(self, algo='weight_only_int8'):
|
|
with IrGuard():
|
|
weight = (
|
|
paddle.rand(shape=(4096, 12288), dtype='float16')
|
|
* 1
|
|
/ math.sqrt(4096)
|
|
)
|
|
|
|
quant_weight, quant_scale = Q.weight_quantize(x=weight, algo=algo)
|
|
dequant_weight = Q.weight_dequantize(
|
|
quant_weight, quant_scale, algo=algo
|
|
)
|
|
exe = paddle.static.Executor(paddle.CUDAPlace(0))
|
|
res = exe.run(feed={}, fetch_list=[weight, dequant_weight])
|
|
np.testing.assert_allclose(res[0], res[1], rtol=1e-2, atol=1e-2)
|
|
|
|
def test_weight_quantize_and_dequantize_int4_pir(
|
|
self, algo='weight_only_int4'
|
|
):
|
|
with IrGuard():
|
|
weight = (
|
|
paddle.rand(shape=(4096, 12288), dtype='float16')
|
|
* 1
|
|
/ math.sqrt(4096)
|
|
)
|
|
quant_weight, quant_scale = Q.weight_quantize(x=weight, algo=algo)
|
|
dequant_weight = Q.weight_dequantize(
|
|
quant_weight, quant_scale, algo=algo
|
|
)
|
|
exe = paddle.static.Executor(paddle.CUDAPlace(0))
|
|
res = exe.run(feed={}, fetch_list=[weight, dequant_weight])
|
|
np.testing.assert_allclose(res[0], res[1], rtol=1e-1, atol=1e-1)
|
|
|
|
def test_weight_only_linear(self):
|
|
out_expect = self.get_linear_out()
|
|
|
|
out_real = self.get_weight_only_linear_out_static()
|
|
if self.dtype == "bfloat16":
|
|
out_real = convert_uint16_to_float(out_real)
|
|
out_expect = convert_uint16_to_float(out_expect)
|
|
np.testing.assert_allclose(
|
|
out_real, out_expect, rtol=self.rtol, atol=self.atol
|
|
)
|
|
|
|
with IrGuard():
|
|
out_real = self.get_weight_only_linear_out_static()
|
|
if self.dtype == "bfloat16":
|
|
out_real = convert_uint16_to_float(out_real)
|
|
out_expect = convert_uint16_to_float(out_expect)
|
|
np.testing.assert_allclose(
|
|
out_real, out_expect, rtol=self.rtol, atol=self.atol
|
|
)
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda(),
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyQuantizeCPUGPUTestCase(unittest.TestCase):
|
|
def config(self):
|
|
self.dtype = 'float16'
|
|
self.batch = 1
|
|
self.token = 32
|
|
self.in_features = 64
|
|
self.out_features = 256
|
|
self.group_size = -1
|
|
|
|
def weightQuantizeCPUGPUConsistenceCheck(self, weight_float):
|
|
for arch in [70, 75, 80, 86]:
|
|
weight_gpu, weight_scale_gpu = Q.weight_quantize(
|
|
weight_float.cuda(),
|
|
algo="weight_only_int4",
|
|
arch=arch,
|
|
group_size=self.group_size,
|
|
)
|
|
weight_cpu, weight_scale_cpu = Q.weight_quantize(
|
|
weight_float.cpu(),
|
|
algo="weight_only_int4",
|
|
arch=arch,
|
|
group_size=self.group_size,
|
|
)
|
|
np.testing.assert_allclose(
|
|
weight_gpu.numpy(),
|
|
weight_cpu.numpy(),
|
|
atol=17,
|
|
)
|
|
np.testing.assert_allclose(
|
|
weight_scale_gpu.numpy(),
|
|
weight_scale_cpu.numpy(),
|
|
atol=1e-5,
|
|
rtol=1e-3,
|
|
)
|
|
|
|
def setUp(self):
|
|
self.config()
|
|
x = np.random.random((self.batch, self.token, self.in_features))
|
|
self.x = paddle.to_tensor(x, dtype=self.dtype)
|
|
set_default_dtype(self.dtype)
|
|
if self.bias:
|
|
bias_attr = base.ParamAttr(
|
|
trainable=False,
|
|
regularizer=None,
|
|
initializer=paddle.nn.initializer.Constant(value=1.0),
|
|
)
|
|
else:
|
|
bias_attr = None
|
|
self.linear = paddle.nn.Linear(
|
|
self.in_features, self.out_features, bias_attr=bias_attr
|
|
)
|
|
|
|
self.bias = self.linear.bias
|
|
self.float_weight = self.linear.weight
|
|
self.weightQuantizeCPUGPUConsistenceCheck(self.float_weight)
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8,
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearBackwardAndWeightDequantizeTestCase(unittest.TestCase):
|
|
def test_weightonly_linear_backward(
|
|
self, algo='weight_only_int8', weight_dtype='int8'
|
|
):
|
|
x = (
|
|
paddle.rand(shape=(128, 4096), dtype='float16')
|
|
* 1
|
|
/ math.sqrt(4096)
|
|
)
|
|
x.stop_gradient = False
|
|
quant_x = copy.deepcopy(x)
|
|
quant_x.stop_gradient = False
|
|
weight = (
|
|
paddle.rand(shape=(4096, 12288), dtype='float16')
|
|
* 1
|
|
/ math.sqrt(4096)
|
|
)
|
|
|
|
quant_weight, quant_scale = Q.weight_quantize(
|
|
x=weight.cuda(), algo=algo
|
|
)
|
|
dequant_weight = Q.weight_dequantize(
|
|
quant_weight.cuda(), quant_scale, algo=algo
|
|
)
|
|
np.testing.assert_allclose(weight, dequant_weight, rtol=1e-2, atol=1e-2)
|
|
|
|
quant_out = Q.weight_only_linear(
|
|
x=quant_x,
|
|
weight=quant_weight,
|
|
weight_scale=quant_scale,
|
|
weight_dtype=weight_dtype,
|
|
)
|
|
out = paddle.matmul(x=x, y=weight)
|
|
np.testing.assert_allclose(quant_out, out, rtol=1e-2, atol=1e-2)
|
|
|
|
quant_out.backward()
|
|
out.backward()
|
|
np.testing.assert_allclose(quant_x.grad, x.grad, rtol=1e-2, atol=1e-2)
|
|
|
|
def test_weightonly_linear_backward_int4(self):
|
|
def test_weightonly_linear_backward(
|
|
self, algo='weight_only_int4', weight_dtype='int4'
|
|
):
|
|
x = (
|
|
paddle.rand(shape=(128, 4096), dtype='float16')
|
|
* 1
|
|
/ math.sqrt(4096)
|
|
)
|
|
x.stop_gradient = False
|
|
quant_x = copy.deepcopy(x)
|
|
quant_x.stop_gradient = False
|
|
weight = (
|
|
paddle.rand(shape=(4096, 12288), dtype='float16')
|
|
* 1
|
|
/ math.sqrt(4096)
|
|
)
|
|
|
|
quant_weight, quant_scale = Q.weight_quantize(
|
|
x=weight.cuda(), algo=algo
|
|
)
|
|
quant_weight = quant_weight.view(
|
|
[quant_weight.shape[0] * 2, quant_weight.shape[1] // 2]
|
|
)
|
|
dequant_weight = Q.weight_dequantize(
|
|
quant_weight.cuda(), quant_scale, algo=algo
|
|
)
|
|
np.testing.assert_allclose(
|
|
weight, dequant_weight, rtol=1e-2, atol=1e-2
|
|
)
|
|
|
|
quant_out = Q.weight_only_linear(
|
|
x=quant_x,
|
|
weight=quant_weight,
|
|
weight_scale=quant_scale,
|
|
weight_dtype=weight_dtype,
|
|
)
|
|
out = paddle.matmul(x=x, y=weight)
|
|
np.testing.assert_allclose(quant_out, out, rtol=1e-3, atol=1e-3)
|
|
|
|
quant_out.backward()
|
|
out.backward()
|
|
np.testing.assert_allclose(
|
|
quant_x.grad, x.grad, rtol=1e-3, atol=1e-3
|
|
)
|
|
|
|
def test_weightonly_linear_backward_int4_zerosize(self):
|
|
def test_weightonly_linear_backward(
|
|
self, algo='weight_only_int4', weight_dtype='int4'
|
|
):
|
|
x = (
|
|
paddle.rand(shape=(0, 4096), dtype='float16')
|
|
* 1
|
|
/ math.sqrt(4096)
|
|
)
|
|
x.stop_gradient = False
|
|
quant_x = copy.deepcopy(x)
|
|
quant_x.stop_gradient = False
|
|
weight = (
|
|
paddle.rand(shape=(0, 12288), dtype='float16')
|
|
* 1
|
|
/ math.sqrt(4096)
|
|
)
|
|
|
|
quant_weight, quant_scale = Q.weight_quantize(
|
|
x=weight.cuda(), algo=algo
|
|
)
|
|
quant_weight = quant_weight.view(
|
|
[quant_weight.shape[0] * 2, quant_weight.shape[1] // 2]
|
|
)
|
|
dequant_weight = Q.weight_dequantize(
|
|
quant_weight.cuda(), quant_scale, algo=algo
|
|
)
|
|
np.testing.assert_allclose(
|
|
weight, dequant_weight, rtol=1e-2, atol=1e-2
|
|
)
|
|
|
|
quant_out = Q.weight_only_linear(
|
|
x=quant_x,
|
|
weight=quant_weight,
|
|
weight_scale=quant_scale,
|
|
weight_dtype=weight_dtype,
|
|
)
|
|
out = paddle.matmul(x=x, y=weight)
|
|
np.testing.assert_allclose(quant_out, out, rtol=1e-3, atol=1e-3)
|
|
|
|
quant_out.backward()
|
|
out.backward()
|
|
np.testing.assert_allclose(
|
|
quant_x.grad, x.grad, rtol=1e-3, atol=1e-3
|
|
)
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8,
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinear_stream_k_TestCase(unittest.TestCase):
|
|
def test_weightonly_linear_backward_int4(self):
|
|
def test_weightonly_linear_backward(
|
|
self, algo='weight_only_int4', weight_dtype='int4'
|
|
):
|
|
x = (
|
|
paddle.rand(shape=(128, 8192), dtype='float16')
|
|
* 1
|
|
/ math.sqrt(8192)
|
|
)
|
|
x.stop_gradient = False
|
|
quant_x = copy.deepcopy(x)
|
|
quant_x.stop_gradient = False
|
|
weight = (
|
|
paddle.rand(shape=(8192, 8192), dtype='float16')
|
|
* 1
|
|
/ math.sqrt(8192)
|
|
)
|
|
|
|
quant_weight, quant_scale = Q.weight_quantize(
|
|
x=weight.cuda(), algo=algo
|
|
)
|
|
|
|
quant_out = Q.weight_only_linear(
|
|
x=quant_x,
|
|
weight=quant_weight,
|
|
weight_scale=quant_scale,
|
|
weight_dtype=weight_dtype,
|
|
)
|
|
|
|
test_weightonly_linear_backward(self)
|
|
|
|
def test_weightonly_linear_backward_int4_bf16(self):
|
|
def test_weightonly_linear_backward(
|
|
self, algo='weight_only_int4', weight_dtype='int4'
|
|
):
|
|
x = (
|
|
paddle.rand(shape=(128, 8192), dtype='bfloat16')
|
|
* 1
|
|
/ math.sqrt(8192)
|
|
)
|
|
x.stop_gradient = False
|
|
quant_x = copy.deepcopy(x)
|
|
quant_x.stop_gradient = False
|
|
weight = (
|
|
paddle.rand(shape=(8192, 8192), dtype='bfloat16')
|
|
* 1
|
|
/ math.sqrt(8192)
|
|
)
|
|
|
|
quant_weight, quant_scale = Q.weight_quantize(
|
|
x=weight.cuda(), algo=algo
|
|
)
|
|
|
|
quant_out = Q.weight_only_linear(
|
|
x=quant_x,
|
|
weight=quant_weight,
|
|
weight_scale=quant_scale,
|
|
weight_dtype=weight_dtype,
|
|
)
|
|
|
|
test_weightonly_linear_backward(self)
|
|
|
|
def test_weightonly_linear_backward_int8(self):
|
|
def test_weightonly_linear_backward(
|
|
self, algo='weight_only_int8', weight_dtype='int8'
|
|
):
|
|
x = (
|
|
paddle.rand(shape=(128, 8192), dtype='float16')
|
|
* 1
|
|
/ math.sqrt(8192)
|
|
)
|
|
x.stop_gradient = False
|
|
quant_x = copy.deepcopy(x)
|
|
quant_x.stop_gradient = False
|
|
weight = (
|
|
paddle.rand(shape=(8192, 8192), dtype='float16')
|
|
* 1
|
|
/ math.sqrt(8192)
|
|
)
|
|
|
|
quant_weight, quant_scale = Q.weight_quantize(
|
|
x=weight.cuda(), algo=algo
|
|
)
|
|
|
|
quant_out = Q.weight_only_linear(
|
|
x=quant_x,
|
|
weight=quant_weight,
|
|
weight_scale=quant_scale,
|
|
weight_dtype=weight_dtype,
|
|
)
|
|
|
|
test_weightonly_linear_backward(self)
|
|
|
|
def test_weightonly_linear_backward_int8_bf16(self):
|
|
def test_weightonly_linear_backward(
|
|
self, algo='weight_only_int8', weight_dtype='int8'
|
|
):
|
|
x = (
|
|
paddle.rand(shape=(128, 8192), dtype='bfloat16')
|
|
* 1
|
|
/ math.sqrt(8192)
|
|
)
|
|
x.stop_gradient = False
|
|
quant_x = copy.deepcopy(x)
|
|
quant_x.stop_gradient = False
|
|
weight = (
|
|
paddle.rand(shape=(8192, 8192), dtype='bfloat16')
|
|
* 1
|
|
/ math.sqrt(8192)
|
|
)
|
|
|
|
quant_weight, quant_scale = Q.weight_quantize(
|
|
x=weight.cuda(), algo=algo
|
|
)
|
|
|
|
quant_out = Q.weight_only_linear(
|
|
x=quant_x,
|
|
weight=quant_weight,
|
|
weight_scale=quant_scale,
|
|
weight_dtype=weight_dtype,
|
|
)
|
|
|
|
test_weightonly_linear_backward(self)
|
|
|
|
|
|
@unittest.skipIf(
|
|
not core.is_compiled_with_cuda()
|
|
or paddle.device.cuda.get_device_capability()[0] < 8,
|
|
"quantized_matmul requires CUDA_ARCH >= 8",
|
|
)
|
|
class WeightOnlyLinearZeroSizeWeightTestCase(unittest.TestCase):
|
|
"""Test weight_only_linear with zero-size weight tensor (first dim = 0).
|
|
|
|
When weight has shape [0, k], the grad kernel (WeightOnlyLinearGradKernel)
|
|
must skip the WeightDequantize call to avoid launching CUDA kernels that
|
|
read from the empty weight buffer.
|
|
"""
|
|
|
|
def _run_zero_weight_forward_and_grad(
|
|
self, dtype, weight_dtype, bias, in_features, out_features
|
|
):
|
|
"""Helper: forward + backward with weight.shape[0] == 0 must not crash."""
|
|
x = paddle.randn([2, 1, in_features], dtype=dtype)
|
|
x.stop_gradient = False
|
|
|
|
weight = paddle.zeros([0, in_features], dtype='int8')
|
|
weight_scale = paddle.randn([out_features], dtype=dtype)
|
|
bias_tensor = (
|
|
paddle.randn([out_features], dtype=dtype) if bias else None
|
|
)
|
|
|
|
out = Q.weight_only_linear(
|
|
x,
|
|
weight,
|
|
bias=bias_tensor,
|
|
weight_scale=weight_scale,
|
|
weight_dtype=weight_dtype,
|
|
)
|
|
|
|
# output should be all zeros since weight is empty
|
|
np.testing.assert_equal(
|
|
out.numpy(),
|
|
np.zeros(
|
|
out.shape,
|
|
dtype=np.float16 if dtype == 'float16' else np.float32,
|
|
),
|
|
)
|
|
|
|
# backward should not crash
|
|
if out.numel() > 0:
|
|
paddle.grad(
|
|
[out],
|
|
[x],
|
|
grad_outputs=[paddle.ones_like(out)],
|
|
allow_unused=True,
|
|
)
|
|
|
|
def test_zero_weight_int8_fp16_with_bias(self):
|
|
self._run_zero_weight_forward_and_grad('float16', 'int8', True, 64, 192)
|
|
|
|
def test_zero_weight_int8_fp16_no_bias(self):
|
|
self._run_zero_weight_forward_and_grad(
|
|
'float16', 'int8', False, 512, 512
|
|
)
|
|
|
|
def test_zero_weight_int4_fp16_with_bias(self):
|
|
self._run_zero_weight_forward_and_grad('float16', 'int4', True, 64, 256)
|
|
|
|
def test_zero_weight_int8_fp16_large(self):
|
|
self._run_zero_weight_forward_and_grad(
|
|
'float16', 'int8', False, 768, 2304
|
|
)
|
|
|
|
def test_zero_weight_int8_fp16_3d_input(self):
|
|
"""Specifically mirrors the original bug report shape."""
|
|
self._run_zero_weight_forward_and_grad(
|
|
'float16', 'int8', True, 128, 288
|
|
)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
unittest.main()
|