wehub-resource-sync
326 lines
11 KiB
Python
326 lines
11 KiB
Python
# Copyright (c) 2024 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
|
|
from op_test import get_device_place, is_custom_device
|
|
|
|
import paddle
|
|
import paddle.nn.functional as F
|
|
from paddle import base
|
|
from paddle.base import core
|
|
from paddle.base.layer_helper import LayerHelper
|
|
|
|
|
|
def naive_residual_add(x, residual):
|
|
return np.add(x, residual)
|
|
|
|
|
|
def naive_group_norm(x, scale, bias, epsilon, groups, data_layout):
|
|
dim = x.ndim
|
|
if dim == 3:
|
|
if data_layout == "NHWC":
|
|
x = np.transpose(x, (0, 2, 1)) # NLC => NCL
|
|
N, C, L = x.shape
|
|
G = groups
|
|
x = x.reshape((N * G, -1))
|
|
mean = np.mean(x, axis=1, keepdims=True)
|
|
var = np.var(x, axis=1, keepdims=True)
|
|
output = (x - mean) / np.sqrt(var + epsilon)
|
|
output = output.reshape((N, C, L)) * scale.reshape(
|
|
(-1, 1)
|
|
) + bias.reshape((-1, 1))
|
|
if data_layout == "NHWC":
|
|
output = np.transpose(output, (0, 2, 1)) # NCL => NLC
|
|
return [output, mean.reshape((N, G)), var.reshape((N, G))]
|
|
elif dim == 4:
|
|
if data_layout == "NHWC":
|
|
x = np.transpose(x, (0, 3, 1, 2)) # NHWC => NCHW
|
|
N, C, H, W = x.shape
|
|
G = groups
|
|
x = x.reshape((N * G, -1))
|
|
mean = np.mean(x, axis=1, keepdims=True)
|
|
var = np.var(x, axis=1, keepdims=True)
|
|
output = (x - mean) / np.sqrt(var + epsilon)
|
|
output = output.reshape((N, C, H, W)) * scale.reshape(
|
|
(-1, 1, 1)
|
|
) + bias.reshape((-1, 1, 1))
|
|
if data_layout == "NHWC":
|
|
output = np.transpose(output, (0, 2, 3, 1)) # NCHW => NHWC
|
|
return [output, mean.reshape((N, G)), var.reshape((N, G))]
|
|
else:
|
|
if data_layout == "NHWC":
|
|
x = np.transpose(x, (0, 4, 1, 2, 3)) # NDHWC => NCDHW
|
|
N, C, D, H, W = x.shape
|
|
G = groups
|
|
x = x.reshape((N * G, -1))
|
|
mean = np.mean(x, axis=1, keepdims=True)
|
|
var = np.var(x, axis=1, keepdims=True)
|
|
output = (x - mean) / np.sqrt(var + epsilon)
|
|
output = output.reshape((N, C, D, H, W)) * scale.reshape(
|
|
(-1, 1, 1, 1)
|
|
) + bias.reshape((-1, 1, 1, 1))
|
|
if data_layout == "NHWC":
|
|
output = np.transpose(output, (0, 2, 3, 4, 1)) # NCDHW => NDHWC
|
|
return [output, mean.reshape((N, G)), var.reshape((N, G))]
|
|
|
|
|
|
def naive_residual_biasadd_layer_norm(
|
|
x, residual, scale, bias, epsilon, groups, data_layout, activation
|
|
):
|
|
x = x + residual
|
|
out = naive_group_norm(x, scale, bias, epsilon, groups, data_layout)
|
|
if activation == "silu":
|
|
out[0] = F.silu(paddle.to_tensor(out[0])).numpy()
|
|
return out
|
|
|
|
|
|
def add_group_norm_silu_static_wrapper(
|
|
x, residual, scale, bias, epsilon, groups, data_layout="NHWC", activation=""
|
|
):
|
|
helper = LayerHelper('add_group_norm_silu', **locals())
|
|
mean_out = helper.create_variable_for_type_inference(
|
|
dtype=x.dtype, stop_gradient=True
|
|
)
|
|
variance_out = helper.create_variable_for_type_inference(
|
|
dtype=x.dtype, stop_gradient=True
|
|
)
|
|
|
|
inputs = {'x': x}
|
|
if bias is not None:
|
|
inputs['bias'] = bias
|
|
if scale is not None:
|
|
inputs['scale'] = scale
|
|
if residual is not None:
|
|
inputs['residual'] = residual
|
|
|
|
# create output
|
|
group_norm_out = helper.create_variable_for_type_inference(dtype=x.dtype)
|
|
residual_out = helper.create_variable_for_type_inference(dtype=x.dtype)
|
|
helper.append_op(
|
|
type="add_group_norm_silu",
|
|
inputs=inputs,
|
|
outputs={
|
|
"y": group_norm_out,
|
|
"residual_out": residual_out,
|
|
"mean": mean_out,
|
|
"variance": variance_out,
|
|
},
|
|
attrs={
|
|
"epsilon": epsilon,
|
|
"groups": groups,
|
|
"data_format": data_layout,
|
|
"activation": activation,
|
|
},
|
|
)
|
|
|
|
return group_norm_out, residual_out
|
|
|
|
|
|
@unittest.skipIf(
|
|
not (core.is_compiled_with_cuda() or is_custom_device())
|
|
or not core.is_float16_supported(get_device_place()),
|
|
"core is not compiled with CUDA or not support the bfloat16",
|
|
)
|
|
class TestGroupNormNHWC_StaticOp(unittest.TestCase):
|
|
def setUp(self):
|
|
np.random.seed(20)
|
|
self.shape = (2, 4, 2, 6)
|
|
self.r_shape = (1, 1, 1, 6)
|
|
self.x_np = np.random.uniform(-0.05, 0.05, self.shape)
|
|
self.residual_np = np.random.uniform(-0.05, 0.05, self.r_shape)
|
|
self.scale_np = np.random.uniform(-0.05, 0.05, [self.shape[-1]])
|
|
self.bias_np = np.random.uniform(-0.05, 0.05, [self.shape[-1]])
|
|
self.epsilon = 1e-5
|
|
self.groups = 2
|
|
self.data_layout = 'NHWC'
|
|
self.activation = ''
|
|
self.place = get_device_place()
|
|
|
|
def check_residual_add_groupnorm(
|
|
self, x_np, scale_np, bias_np, residual_np, activation, dtype
|
|
):
|
|
paddle.disable_static()
|
|
naive_groupnorm_out = naive_residual_biasadd_layer_norm(
|
|
x_np,
|
|
residual_np,
|
|
scale_np,
|
|
bias_np,
|
|
self.epsilon,
|
|
self.groups,
|
|
self.data_layout,
|
|
self.activation,
|
|
)
|
|
naive_residual_out = naive_residual_add(x_np, residual_np)
|
|
paddle.enable_static()
|
|
|
|
with (
|
|
paddle.pir_utils.OldIrGuard(),
|
|
paddle.static.program_guard(paddle.static.Program()),
|
|
):
|
|
x_static = paddle.static.data(
|
|
name="x_static", shape=self.shape, dtype=dtype
|
|
)
|
|
residual_static = paddle.static.data(
|
|
name="residual_static",
|
|
shape=self.r_shape,
|
|
dtype=dtype,
|
|
)
|
|
|
|
scale_static = paddle.static.data(
|
|
name="scale_static", shape=[self.shape[-1]], dtype=dtype
|
|
)
|
|
bias_static = paddle.static.data(
|
|
name="bias_static", shape=[self.shape[-1]], dtype=dtype
|
|
)
|
|
outs = add_group_norm_silu_static_wrapper(
|
|
x_static,
|
|
residual_static,
|
|
scale_static,
|
|
bias_static,
|
|
self.epsilon,
|
|
self.groups,
|
|
self.data_layout,
|
|
activation,
|
|
)
|
|
|
|
exe = base.Executor(self.place)
|
|
out_s = exe.run(
|
|
feed={
|
|
"x_static": x_np.astype(dtype),
|
|
"scale_static": scale_np.astype(dtype),
|
|
"residual_static": residual_np.astype(dtype),
|
|
"bias_static": bias_np.astype(dtype),
|
|
},
|
|
fetch_list=[outs],
|
|
)
|
|
return (out_s[0], out_s[1]), naive_groupnorm_out, naive_residual_out
|
|
|
|
def test_residual_add_groupnorm_fp16(self):
|
|
if not (paddle.is_compiled_with_cuda() or is_custom_device()):
|
|
return
|
|
self.dtype = np.float16
|
|
(
|
|
paddle_group_list,
|
|
paddle_naive_group_out,
|
|
paddle_naive_group_residual,
|
|
) = self.check_residual_add_groupnorm(
|
|
self.x_np.astype(self.dtype),
|
|
self.scale_np.astype(self.dtype),
|
|
self.bias_np.astype(self.dtype),
|
|
self.residual_np.astype(self.dtype),
|
|
self.activation,
|
|
self.dtype,
|
|
)
|
|
np.testing.assert_allclose(
|
|
paddle_group_list[1],
|
|
paddle_naive_group_residual,
|
|
rtol=1e-5,
|
|
atol=1e-5,
|
|
)
|
|
np.testing.assert_allclose(
|
|
paddle_group_list[0],
|
|
paddle_naive_group_out[0],
|
|
rtol=1e-4,
|
|
atol=1e-4,
|
|
)
|
|
|
|
|
|
@unittest.skipIf(
|
|
not (core.is_compiled_with_cuda() or is_custom_device())
|
|
or not core.is_float16_supported(get_device_place()),
|
|
"core is not compiled with CUDA or not support the bfloat16",
|
|
)
|
|
class TestGroupNormNHWCSilu_StaticOp(TestGroupNormNHWC_StaticOp):
|
|
def setUp(self):
|
|
np.random.seed(20)
|
|
self.shape = (2, 4, 2, 6)
|
|
self.r_shape = (1, 1, 1, 6)
|
|
self.x_np = np.random.uniform(-0.05, 0.05, self.shape)
|
|
self.residual_np = np.random.uniform(-0.05, 0.05, self.r_shape)
|
|
self.scale_np = np.random.uniform(-0.05, 0.05, [self.shape[-1]])
|
|
self.bias_np = np.random.uniform(-0.05, 0.05, [self.shape[-1]])
|
|
self.epsilon = 1e-5
|
|
self.groups = 2
|
|
self.data_layout = 'NHWC'
|
|
self.activation = 'silu'
|
|
self.place = get_device_place()
|
|
|
|
|
|
@unittest.skipIf(
|
|
not (core.is_compiled_with_cuda() or is_custom_device())
|
|
or not core.is_float16_supported(get_device_place()),
|
|
"core is not compiled with CUDA or not support the bfloat16",
|
|
)
|
|
class TestGroupNormNHWC_StaticOp_1(TestGroupNormNHWC_StaticOp):
|
|
def setUp(self):
|
|
np.random.seed(20)
|
|
self.shape = (2, 4, 2, 6)
|
|
self.r_shape = (2, 4, 2, 6)
|
|
self.x_np = np.random.uniform(-0.05, 0.05, self.shape)
|
|
self.residual_np = np.random.uniform(-0.05, 0.05, self.r_shape)
|
|
self.scale_np = np.random.uniform(-0.05, 0.05, [self.shape[-1]])
|
|
self.bias_np = np.random.uniform(-0.05, 0.05, [self.shape[-1]])
|
|
self.epsilon = 1e-5
|
|
self.groups = 2
|
|
self.data_layout = 'NHWC'
|
|
self.activation = 'silu'
|
|
self.place = get_device_place()
|
|
|
|
|
|
@unittest.skipIf(
|
|
not (core.is_compiled_with_cuda() or is_custom_device())
|
|
or not core.is_float16_supported(get_device_place()),
|
|
"core is not compiled with CUDA or not support the bfloat16",
|
|
)
|
|
class TestGroupNormNHWCSilu_StaticOp_1(TestGroupNormNHWC_StaticOp):
|
|
def setUp(self):
|
|
np.random.seed(20)
|
|
self.shape = (2, 4, 2, 6)
|
|
self.r_shape = (2, 4, 2, 6)
|
|
self.x_np = np.random.uniform(-0.05, 0.05, self.shape)
|
|
self.residual_np = np.random.uniform(-0.05, 0.05, self.r_shape)
|
|
self.scale_np = np.random.uniform(-0.05, 0.05, [self.shape[-1]])
|
|
self.bias_np = np.random.uniform(-0.05, 0.05, [self.shape[-1]])
|
|
self.epsilon = 1e-5
|
|
self.groups = 2
|
|
self.data_layout = 'NHWC'
|
|
self.activation = 'silu'
|
|
self.place = get_device_place()
|
|
|
|
|
|
@unittest.skipIf(
|
|
not (core.is_compiled_with_cuda() or is_custom_device())
|
|
or not core.is_float16_supported(get_device_place()),
|
|
"core is not compiled with CUDA or not support the bfloat16",
|
|
)
|
|
class TestGroupNormNHWCSingleC_StaticOp(TestGroupNormNHWC_StaticOp):
|
|
def setUp(self):
|
|
np.random.seed(20)
|
|
self.shape = (2, 4, 2, 6)
|
|
self.r_shape = (2, 4, 2, 6)
|
|
self.x_np = np.random.uniform(-0.05, 0.05, self.shape)
|
|
self.residual_np = np.random.uniform(-0.05, 0.05, self.r_shape)
|
|
self.scale_np = np.random.uniform(-0.05, 0.05, [self.shape[-1]])
|
|
self.bias_np = np.random.uniform(-0.05, 0.05, [self.shape[-1]])
|
|
self.epsilon = 1e-5
|
|
self.groups = 6
|
|
self.data_layout = 'NHWC'
|
|
self.activation = ''
|
|
self.place = get_device_place()
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|