Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
PaddlePaddle
Paddle
提交
7eeff7b1
P
Paddle
项目概览
PaddlePaddle
/
Paddle
12 个月 前同步成功
通知
2292
Star
20931
Fork
5422
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
1423
列表
看板
标记
里程碑
合并请求
543
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
Paddle
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
1,423
Issue
1,423
列表
看板
标记
里程碑
合并请求
543
合并请求
543
Pages
分析
分析
仓库分析
DevOps
Wiki
0
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
提交
Issue看板
体验新版 GitCode,发现更多精彩内容 >>
未验证
提交
7eeff7b1
编写于
7月 20, 2023
作者:
L
Leo Chen
提交者:
GitHub
7月 20, 2023
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Fix UT failure (#55360)
* Fix TRT multihead matmul UT failure
上级
fa084e5e
变更
7
显示空白变更内容
内联
并排
Showing
7 changed file
with
123 addition
and
481 deletion
+123
-481
paddle/fluid/inference/api/CMakeLists.txt
paddle/fluid/inference/api/CMakeLists.txt
+6
-3
paddle/fluid/inference/api/paddle_pass_builder.cc
paddle/fluid/inference/api/paddle_pass_builder.cc
+13
-6
test/ir/inference/CMakeLists.txt
test/ir/inference/CMakeLists.txt
+0
-1
test/ir/inference/test_trt_activation_pass.py
test/ir/inference/test_trt_activation_pass.py
+0
-399
test/ir/inference/test_trt_convert_multihead_matmul.py
test/ir/inference/test_trt_convert_multihead_matmul.py
+101
-71
test/legacy_test/test_math_op_patch.py
test/legacy_test/test_math_op_patch.py
+2
-0
test/quantization/test_post_training_quantization_mobilenetv1.py
...antization/test_post_training_quantization_mobilenetv1.py
+1
-1
未找到文件。
paddle/fluid/inference/api/CMakeLists.txt
浏览文件 @
7eeff7b1
...
...
@@ -20,7 +20,10 @@ cc_library(
paddle_infer_contrib
SRCS paddle_infer_contrib.cc
DEPS zero_copy_tensor
)
cc_library
(
paddle_pass_builder SRCS paddle_pass_builder.cc
)
cc_library
(
paddle_pass_builder
SRCS paddle_pass_builder.cc
DEPS framework_proto
)
set
(
paddle_inference_api_deps
reset_tensor_array
...
...
@@ -54,8 +57,8 @@ endif()
cc_library
(
analysis_config
SRCS analysis_config.cc
DEPS
${
mkldnn_quantizer_cfg
}
paddle_inference_api
lod_tenso
r
paddle_pass_builder
table_printer utf8proc
)
DEPS
${
mkldnn_quantizer_cfg
}
paddle_inference_api
paddle_pass_builde
r
table_printer utf8proc
)
if
(
WIN32
)
target_link_libraries
(
paddle_inference_api phi
)
...
...
paddle/fluid/inference/api/paddle_pass_builder.cc
浏览文件 @
7eeff7b1
...
...
@@ -19,6 +19,9 @@
#ifdef PADDLE_WITH_HIP
#include <miopen/miopen.h>
#endif
#ifdef PADDLE_WITH_TENSORRT
#include "paddle/fluid/inference/tensorrt/helper.h"
#endif
#include <glog/logging.h>
...
...
@@ -103,8 +106,12 @@ const std::vector<std::string> kTRTSubgraphPasses({
"trt_multihead_matmul_fuse_pass_v3"
,
//
"multihead_matmul_roformer_fuse_pass"
,
//
"constant_folding_pass"
,
//
#ifdef PADDLE_WITH_TENSORRT
#if !IS_TRT_VERSION_GE(8610)
"trt_flash_multihead_matmul_fuse_pass"
,
//
"trt_cross_multihead_matmul_fuse_pass"
,
//
#endif
#endif
"vit_attention_fuse_pass"
,
//
"trt_qk_multihead_matmul_fuse_pass"
,
//
"layernorm_shift_partition_fuse_pass"
,
//
...
...
test/ir/inference/CMakeLists.txt
浏览文件 @
7eeff7b1
...
...
@@ -168,7 +168,6 @@ endif()
if
(
WITH_GPU AND TENSORRT_FOUND
)
set_tests_properties
(
test_trt_subgraph_pass PROPERTIES TIMEOUT 120
)
# longer timeout for trt_activation_pass for longer trt optimization time in trt 8
set_tests_properties
(
test_trt_activation_pass PROPERTIES TIMEOUT 250
)
set_tests_properties
(
test_trt_conv_pass PROPERTIES TIMEOUT 120
)
#set_tests_properties(test_trt_multiclass_nms_op PROPERTIES TIMEOUT 200)
set_tests_properties
(
test_trt_dynamic_shape PROPERTIES TIMEOUT 120
)
...
...
test/ir/inference/test_trt_activation_pass.py
已删除
100644 → 0
浏览文件 @
fa084e5e
# Copyright (c) 2020 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
os
import
shutil
import
unittest
import
numpy
as
np
from
inference_pass_test
import
InferencePassTest
import
paddle
import
paddle.nn.functional
as
F
from
paddle
import
fluid
from
paddle.fluid
import
core
from
paddle.fluid.core
import
AnalysisConfig
,
PassVersionChecker
from
paddle.static
import
nn
class
TensorRTSubgraphPassActivationTest
(
InferencePassTest
):
def
setUpTensorRTParam
(
self
):
self
.
enable_trt
=
True
self
.
trt_parameters
=
TensorRTSubgraphPassActivationTest
.
TensorRTParam
(
1
<<
30
,
32
,
0
,
AnalysisConfig
.
Precision
.
Float32
,
False
,
False
)
def
setUp
(
self
):
self
.
setUpTensorRTParam
()
with
fluid
.
program_guard
(
self
.
main_program
,
self
.
startup_program
):
data
=
paddle
.
static
.
data
(
name
=
"data"
,
shape
=
[
-
1
,
6
,
32
,
32
],
dtype
=
"float32"
)
act_out
=
self
.
append_act
(
data
)
out
=
nn
.
batch_norm
(
act_out
,
is_test
=
True
)
self
.
feeds
=
{
"data"
:
np
.
random
.
random
([
1
,
6
,
32
,
32
]).
astype
(
"float32"
),
}
self
.
fetch_list
=
[
out
]
def
append_act
(
self
,
x
):
return
F
.
relu
(
x
)
def
test_check_output
(
self
):
if
core
.
is_compiled_with_cuda
():
use_gpu
=
True
opt_path
=
os
.
path
.
join
(
self
.
path
,
'_opt_cache'
)
if
os
.
path
.
exists
(
opt_path
):
shutil
.
rmtree
(
opt_path
)
if
(
self
.
trt_parameters
.
precision
==
AnalysisConfig
.
Precision
.
Float32
):
self
.
check_output_with_option
(
use_gpu
)
else
:
self
.
check_output_with_option
(
use_gpu
,
1e-3
)
self
.
assertTrue
(
PassVersionChecker
.
IsCompatible
(
'tensorrt_subgraph_pass'
)
)
class
TensorRTSubgraphPassLeakyReluTest
(
TensorRTSubgraphPassActivationTest
):
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
leaky_relu
(
x
)
class
TensorRTSubgraphPassRelu6Test
(
TensorRTSubgraphPassActivationTest
):
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
relu6
(
x
)
class
TensorRTSubgraphPassSoftMaxTest
(
TensorRTSubgraphPassActivationTest
):
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
softmax
(
x
)
class
TensorRTSubgraphPassSigmoidTest
(
TensorRTSubgraphPassActivationTest
):
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
sigmoid
(
x
)
class
TensorRTSubgraphPassHardSwishTest
(
TensorRTSubgraphPassActivationTest
):
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
hardswish
(
x
)
class
TensorRTSubgraphPassHardSigmoidTest
(
TensorRTSubgraphPassActivationTest
):
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
hardsigmoid
(
x
)
class
TensorRTSubgraphPassHardSwishPluginTest
(
TensorRTSubgraphPassActivationTest
):
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
hardswish
(
x
)
class
TensorRTSubgraphPassClipTest
(
TensorRTSubgraphPassActivationTest
):
def
append_act
(
self
,
x
):
return
paddle
.
clip
(
x
,
0
,
1
)
class
TensorRTSubgraphPassTanhTest
(
TensorRTSubgraphPassActivationTest
):
def
append_act
(
self
,
x
):
return
paddle
.
tanh
(
x
)
class
TensorRTSubgraphPassSwishTest
(
TensorRTSubgraphPassActivationTest
):
def
setUpTensorRTParam
(
self
):
self
.
enable_trt
=
True
self
.
trt_parameters
=
TensorRTSubgraphPassActivationTest
.
TensorRTParam
(
1
<<
30
,
32
,
0
,
AnalysisConfig
.
Precision
.
Float32
,
True
,
False
)
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
swish
(
x
)
class
TensorRTSubgraphPassSwishFp16SerializeTest
(
TensorRTSubgraphPassActivationTest
):
def
setUpTensorRTParam
(
self
):
self
.
enable_trt
=
True
self
.
trt_parameters
=
TensorRTSubgraphPassActivationTest
.
TensorRTParam
(
1
<<
30
,
32
,
0
,
AnalysisConfig
.
Precision
.
Half
,
True
,
False
)
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
swish
(
x
)
class
TensorRTSubgraphPassDynamicSwishFp16SerializeTest
(
TensorRTSubgraphPassActivationTest
):
def
setUpTensorRTParam
(
self
):
self
.
enable_trt
=
True
self
.
trt_parameters
=
TensorRTSubgraphPassActivationTest
.
TensorRTParam
(
1
<<
30
,
32
,
0
,
AnalysisConfig
.
Precision
.
Half
,
True
,
False
)
self
.
dynamic_shape_params
=
(
TensorRTSubgraphPassActivationTest
.
DynamicShapeParam
(
{
'data'
:
[
1
,
6
,
8
,
8
]},
{
'data'
:
[
1
,
6
,
128
,
128
]},
{
'data'
:
[
1
,
6
,
64
,
64
]},
False
,
)
)
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
swish
(
x
)
class
TensorRTSubgraphPassMishTest
(
TensorRTSubgraphPassActivationTest
):
def
setUpTensorRTParam
(
self
):
self
.
enable_trt
=
True
self
.
trt_parameters
=
TensorRTSubgraphPassActivationTest
.
TensorRTParam
(
1
<<
30
,
32
,
0
,
AnalysisConfig
.
Precision
.
Float32
,
True
,
False
)
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
mish
(
x
)
class
TensorRTSubgraphPassMishFp16SerializeTest
(
TensorRTSubgraphPassActivationTest
):
def
setUpTensorRTParam
(
self
):
self
.
enable_trt
=
True
self
.
trt_parameters
=
TensorRTSubgraphPassActivationTest
.
TensorRTParam
(
1
<<
30
,
32
,
0
,
AnalysisConfig
.
Precision
.
Half
,
True
,
False
)
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
mish
(
x
)
class
TensorRTSubgraphPassDynamicMishFp16SerializeTest
(
TensorRTSubgraphPassActivationTest
):
def
setUpTensorRTParam
(
self
):
self
.
enable_trt
=
True
self
.
trt_parameters
=
TensorRTSubgraphPassActivationTest
.
TensorRTParam
(
1
<<
30
,
32
,
0
,
AnalysisConfig
.
Precision
.
Half
,
False
,
False
)
self
.
dynamic_shape_params
=
(
TensorRTSubgraphPassActivationTest
.
DynamicShapeParam
(
{
'data'
:
[
1
,
6
,
8
,
8
]},
{
'data'
:
[
1
,
6
,
128
,
128
]},
{
'data'
:
[
1
,
6
,
64
,
64
]},
False
,
)
)
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
mish
(
x
)
class
TensorRTSubgraphPassPreluAllTest
(
TensorRTSubgraphPassActivationTest
):
def
append_act
(
self
,
x
):
return
paddle
.
static
.
nn
.
prelu
(
x
,
mode
=
'all'
)
class
TensorRTSubgraphPassPreluChannelTest
(
TensorRTSubgraphPassActivationTest
):
def
append_act
(
self
,
x
):
return
paddle
.
static
.
nn
.
prelu
(
x
,
mode
=
'channel'
)
class
TensorRTSubgraphPassPreluElementTest
(
TensorRTSubgraphPassActivationTest
):
def
append_act
(
self
,
x
):
return
paddle
.
static
.
nn
.
prelu
(
x
,
mode
=
'element'
)
class
TensorRTSubgraphPassPreluDynamicTest
(
TensorRTSubgraphPassActivationTest
):
def
setUpTensorRTParam
(
self
):
self
.
enable_trt
=
True
self
.
trt_parameters
=
TensorRTSubgraphPassActivationTest
.
TensorRTParam
(
1
<<
30
,
32
,
0
,
AnalysisConfig
.
Precision
.
Float32
,
False
,
False
)
self
.
dynamic_shape_params
=
(
TensorRTSubgraphPassActivationTest
.
DynamicShapeParam
(
{
'data'
:
[
1
,
6
,
8
,
8
]},
{
'data'
:
[
1
,
6
,
128
,
128
]},
{
'data'
:
[
1
,
6
,
64
,
64
]},
False
,
)
)
def
append_act
(
self
,
x
):
return
paddle
.
static
.
nn
.
prelu
(
x
,
mode
=
'all'
)
class
TensorRTSubgraphPassPreluFp16Test
(
TensorRTSubgraphPassActivationTest
):
def
setUpTensorRTParam
(
self
):
self
.
enable_trt
=
True
self
.
trt_parameters
=
TensorRTSubgraphPassActivationTest
.
TensorRTParam
(
1
<<
30
,
32
,
0
,
AnalysisConfig
.
Precision
.
Half
,
False
,
False
)
def
append_act
(
self
,
x
):
return
paddle
.
static
.
nn
.
prelu
(
x
,
mode
=
'all'
)
class
TensorRTSubgraphPassPreluFp16SerializeTest
(
TensorRTSubgraphPassActivationTest
):
def
setUpTensorRTParam
(
self
):
self
.
enable_trt
=
True
self
.
trt_parameters
=
TensorRTSubgraphPassActivationTest
.
TensorRTParam
(
1
<<
30
,
32
,
0
,
AnalysisConfig
.
Precision
.
Half
,
True
,
False
)
def
append_act
(
self
,
x
):
return
paddle
.
static
.
nn
.
prelu
(
x
,
mode
=
'all'
)
class
TensorRTSubgraphPassPreluFp16DynamicTest
(
TensorRTSubgraphPassActivationTest
):
def
setUpTensorRTParam
(
self
):
self
.
enable_trt
=
True
self
.
trt_parameters
=
TensorRTSubgraphPassActivationTest
.
TensorRTParam
(
1
<<
30
,
32
,
0
,
AnalysisConfig
.
Precision
.
Half
,
False
,
False
)
self
.
dynamic_shape_params
=
(
TensorRTSubgraphPassActivationTest
.
DynamicShapeParam
(
{
'data'
:
[
1
,
6
,
8
,
8
]},
{
'data'
:
[
1
,
6
,
128
,
128
]},
{
'data'
:
[
1
,
6
,
64
,
64
]},
False
,
)
)
def
append_act
(
self
,
x
):
return
paddle
.
static
.
nn
.
prelu
(
x
,
mode
=
'all'
)
class
TensorRTSubgraphPassPreluFp16DynamicSerializeTest
(
TensorRTSubgraphPassActivationTest
):
def
setUpTensorRTParam
(
self
):
self
.
enable_trt
=
True
self
.
trt_parameters
=
TensorRTSubgraphPassActivationTest
.
TensorRTParam
(
1
<<
30
,
32
,
0
,
AnalysisConfig
.
Precision
.
Half
,
True
,
False
)
self
.
dynamic_shape_params
=
(
TensorRTSubgraphPassActivationTest
.
DynamicShapeParam
(
{
'data'
:
[
1
,
6
,
8
,
8
]},
{
'data'
:
[
1
,
6
,
128
,
128
]},
{
'data'
:
[
1
,
6
,
64
,
64
]},
False
,
)
)
def
append_act
(
self
,
x
):
return
paddle
.
static
.
nn
.
prelu
(
x
,
mode
=
'all'
)
class
TensorRTSubgraphPassGeluTest
(
TensorRTSubgraphPassActivationTest
):
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
gelu
(
x
)
class
TensorRTSubgraphPassGeluDynamicTest
(
TensorRTSubgraphPassActivationTest
):
def
setUpTensorRTParam
(
self
):
self
.
enable_trt
=
True
self
.
trt_parameters
=
TensorRTSubgraphPassActivationTest
.
TensorRTParam
(
1
<<
30
,
32
,
0
,
AnalysisConfig
.
Precision
.
Float32
,
False
,
False
)
self
.
dynamic_shape_params
=
(
TensorRTSubgraphPassActivationTest
.
DynamicShapeParam
(
{
'data'
:
[
1
,
6
,
8
,
8
]},
{
'data'
:
[
1
,
6
,
128
,
128
]},
{
'data'
:
[
1
,
6
,
64
,
64
]},
False
,
)
)
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
gelu
(
x
)
class
TensorRTSubgraphPassGeluFp16Test
(
TensorRTSubgraphPassActivationTest
):
def
setUpTensorRTParam
(
self
):
self
.
enable_trt
=
True
self
.
trt_parameters
=
TensorRTSubgraphPassActivationTest
.
TensorRTParam
(
1
<<
30
,
32
,
0
,
AnalysisConfig
.
Precision
.
Half
,
False
,
False
)
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
gelu
(
x
)
class
TensorRTSubgraphPassGeluFp16SerializeTest
(
TensorRTSubgraphPassActivationTest
):
def
setUpTensorRTParam
(
self
):
self
.
enable_trt
=
True
self
.
trt_parameters
=
TensorRTSubgraphPassActivationTest
.
TensorRTParam
(
1
<<
30
,
32
,
0
,
AnalysisConfig
.
Precision
.
Half
,
True
,
False
)
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
gelu
(
x
)
class
TensorRTSubgraphPassGeluFp16DynamicTest
(
TensorRTSubgraphPassActivationTest
):
def
setUpTensorRTParam
(
self
):
self
.
enable_trt
=
True
self
.
trt_parameters
=
TensorRTSubgraphPassActivationTest
.
TensorRTParam
(
1
<<
30
,
32
,
0
,
AnalysisConfig
.
Precision
.
Half
,
False
,
False
)
self
.
dynamic_shape_params
=
(
TensorRTSubgraphPassActivationTest
.
DynamicShapeParam
(
{
'data'
:
[
1
,
6
,
8
,
8
]},
{
'data'
:
[
1
,
6
,
128
,
128
]},
{
'data'
:
[
1
,
6
,
64
,
64
]},
False
,
)
)
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
gelu
(
x
)
class
TensorRTSubgraphPassGeluFp16DynamicSerializeTest
(
TensorRTSubgraphPassActivationTest
):
def
setUpTensorRTParam
(
self
):
self
.
enable_trt
=
True
self
.
trt_parameters
=
TensorRTSubgraphPassActivationTest
.
TensorRTParam
(
1
<<
30
,
32
,
0
,
AnalysisConfig
.
Precision
.
Half
,
True
,
False
)
self
.
dynamic_shape_params
=
(
TensorRTSubgraphPassActivationTest
.
DynamicShapeParam
(
{
'data'
:
[
1
,
6
,
8
,
8
]},
{
'data'
:
[
1
,
6
,
128
,
128
]},
{
'data'
:
[
1
,
6
,
64
,
64
]},
False
,
)
)
def
append_act
(
self
,
x
):
return
paddle
.
nn
.
functional
.
gelu
(
x
)
if
__name__
==
"__main__"
:
unittest
.
main
()
test/ir/inference/test_trt_convert_multihead_matmul.py
浏览文件 @
7eeff7b1
...
...
@@ -28,17 +28,14 @@ class TrtConvertMultiHeadMatmulTest(TrtLayerAutoScanTest):
return
True
def
sample_program_configs
(
self
):
def
generate_input
1
(
batch
,
dim1
):
return
np
.
full
(
(
batch
,
dim1
,
768
),
1
).
astype
(
np
.
float32
)
def
generate_input
(
shape
):
return
np
.
full
(
shape
,
0.
1
).
astype
(
np
.
float32
)
def
generate_input2
(
shape
):
return
np
.
full
(
shape
,
1
).
astype
(
np
.
float32
)
def
generate_weight1
():
return
np
.
full
((
768
,
768
),
0.1
).
astype
(
np
.
float32
)
def
generate_weight2
():
return
np
.
full
((
768
),
0.1
).
astype
(
np
.
float32
)
def
generate_weight
(
shape
):
return
(
np
.
random
.
rand
(
*
shape
).
astype
(
np
.
float32
).
round
(
decimals
=
1
)
/
5
-
0.1
)
for
batch
in
[
1
,
4
]:
self
.
batch
=
batch
...
...
@@ -303,36 +300,50 @@ class TrtConvertMultiHeadMatmulTest(TrtLayerAutoScanTest):
ops
=
ops
,
weights
=
{
"mul1_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight1
)
data_gen
=
partial
(
generate_weight
,
(
768
,
768
)
)
),
"mul2_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight1
)
data_gen
=
partial
(
generate_weight
,
(
768
,
768
)
)
),
"mul3_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight1
)
data_gen
=
partial
(
generate_weight
,
(
768
,
768
)
)
),
"mul4_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight1
)
data_gen
=
partial
(
generate_weight
,
(
768
,
768
)
)
),
"elementwise_add1_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight2
)
data_gen
=
partial
(
generate_weight
,
(
768
,)
)
),
"elementwise_add2_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight2
)
data_gen
=
partial
(
generate_weight
,
(
768
,)
)
),
"elementwise_add3_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight2
)
data_gen
=
partial
(
generate_weight
,
(
768
,)
)
),
},
inputs
=
{
"input_data1"
:
TensorConfig
(
data_gen
=
partial
(
generate_input
1
,
batch
,
dim1
generate_input
,
(
batch
,
dim1
,
768
)
)
),
"input_data2"
:
TensorConfig
(
data_gen
=
partial
(
generate_input
2
,
input2_shape
generate_input
,
input2_shape
)
),
},
...
...
@@ -375,9 +386,9 @@ class TrtConvertMultiHeadMatmulTest(TrtLayerAutoScanTest):
generate_dynamic_shape
(
attrs
)
self
.
trt_param
.
precision
=
paddle_infer
.
PrecisionType
.
Float32
self
.
trt_param
.
workspace_size
=
2013265920
yield
self
.
create_inference_config
(),
(
1
,
3
),
(
1e-5
,
1e-
4
)
yield
self
.
create_inference_config
(),
(
1
,
3
),
(
1e-5
,
1e-
5
)
self
.
trt_param
.
precision
=
paddle_infer
.
PrecisionType
.
Half
yield
self
.
create_inference_config
(),
(
1
,
3
),
(
1e-
3
,
1e-2
)
yield
self
.
create_inference_config
(),
(
1
,
3
),
(
1e-
2
,
1e-2
)
def
test
(
self
):
self
.
run_test
()
...
...
@@ -385,17 +396,14 @@ class TrtConvertMultiHeadMatmulTest(TrtLayerAutoScanTest):
class
TrtConvertMultiHeadMatmulTestInt8
(
TrtConvertMultiHeadMatmulTest
):
def
sample_program_configs
(
self
):
def
generate_input1
(
batch
,
dim1
):
return
np
.
full
((
batch
,
dim1
,
768
),
1
).
astype
(
np
.
float32
)
def
generate_input2
(
shape
):
return
np
.
full
(
shape
,
1
).
astype
(
np
.
float32
)
def
generate_input
(
shape
):
return
np
.
full
(
shape
,
0.1
).
astype
(
np
.
float32
)
def
generate_weight
1
(
):
return
np
.
full
((
768
,
768
),
0.1
).
astype
(
np
.
float32
)
def
generate_weight2
():
return
np
.
full
((
768
),
0.1
).
astype
(
np
.
float32
)
def
generate_weight
(
shape
):
return
(
np
.
random
.
rand
(
*
shape
).
astype
(
np
.
float32
).
round
(
decimals
=
1
)
/
5
-
0.1
)
for
batch
in
[
4
]:
self
.
batch
=
batch
...
...
@@ -684,36 +692,50 @@ class TrtConvertMultiHeadMatmulTestInt8(TrtConvertMultiHeadMatmulTest):
ops
=
ops
,
weights
=
{
"mul1_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight1
)
data_gen
=
partial
(
generate_weight
,
(
768
,
768
)
)
),
"mul2_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight1
)
data_gen
=
partial
(
generate_weight
,
(
768
,
768
)
)
),
"mul3_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight1
)
data_gen
=
partial
(
generate_weight
,
(
768
,
768
)
)
),
"mul4_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight1
)
data_gen
=
partial
(
generate_weight
,
(
768
,
768
)
)
),
"elementwise_add1_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight2
)
data_gen
=
partial
(
generate_weight
,
(
768
,)
)
),
"elementwise_add2_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight2
)
data_gen
=
partial
(
generate_weight
,
(
768
,)
)
),
"elementwise_add3_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight2
)
data_gen
=
partial
(
generate_weight
,
(
768
,)
)
),
},
inputs
=
{
"input_data1"
:
TensorConfig
(
data_gen
=
partial
(
generate_input
1
,
batch
,
dim1
generate_input
,
(
batch
,
dim1
,
768
)
)
),
"input_data2"
:
TensorConfig
(
data_gen
=
partial
(
generate_input
2
,
input2_shape
generate_input
,
input2_shape
)
),
},
...
...
@@ -731,11 +753,11 @@ class TrtConvertVitToMultiHeadMatmulTest(TrtLayerAutoScanTest):
def
generate_input1
(
batch
,
length
):
return
np
.
full
((
batch
,
length
,
768
),
0.1
).
astype
(
np
.
float32
)
def
generate_weight
1
(
):
return
np
.
full
((
768
,
2304
),
0.1
).
astype
(
np
.
float32
)
def
generate_weight2
():
return
np
.
full
((
2304
),
0.1
).
astype
(
np
.
float32
)
def
generate_weight
(
shape
):
return
(
np
.
random
.
rand
(
*
shape
).
astype
(
np
.
float32
).
round
(
decimals
=
1
)
/
5
-
0.1
)
for
batch
in
[
4
]:
self
.
batch
=
batch
...
...
@@ -911,10 +933,10 @@ class TrtConvertVitToMultiHeadMatmulTest(TrtLayerAutoScanTest):
ops
=
ops
,
weights
=
{
"matmul1_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight
1
)
data_gen
=
partial
(
generate_weight
,
(
768
,
2304
)
)
),
"elementwise_add1_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight
2
)
data_gen
=
partial
(
generate_weight
,
(
2304
,)
)
),
},
inputs
=
{
...
...
@@ -985,20 +1007,14 @@ class TrtConvertMultiHeadMatmulTest_biasqk_seqseq(TrtLayerAutoScanTest):
return
True
def
sample_program_configs
(
self
):
def
generate_input1
(
batch
,
dim1
):
return
np
.
full
((
batch
,
dim1
,
768
),
1
).
astype
(
np
.
float32
)
def
generate_input2
(
shape
):
return
np
.
full
(
shape
,
1
).
astype
(
np
.
float32
)
def
generate_weight1
():
return
np
.
full
((
768
,
768
),
0.1
).
astype
(
np
.
float32
)
def
generate_input
(
shape
):
return
np
.
full
(
shape
,
0.1
).
astype
(
np
.
float32
)
def
generate_weight
2
(
):
return
np
.
full
((
768
),
0.1
).
astype
(
np
.
float32
)
def
generate_weight3
():
return
np
.
full
((
768
,
768
),
0.1
).
astype
(
np
.
float32
)
def
generate_weight
(
shape
):
return
(
np
.
random
.
rand
(
*
shape
).
astype
(
np
.
float32
).
round
(
decimals
=
1
)
/
5
-
0.1
)
for
batch
in
[
2
]:
self
.
batch
=
batch
...
...
@@ -1263,36 +1279,50 @@ class TrtConvertMultiHeadMatmulTest_biasqk_seqseq(TrtLayerAutoScanTest):
ops
=
ops
,
weights
=
{
"mul1_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight1
)
data_gen
=
partial
(
generate_weight
,
(
768
,
768
)
)
),
"mul2_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight1
)
data_gen
=
partial
(
generate_weight
,
(
768
,
768
)
)
),
"mul3_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight1
)
data_gen
=
partial
(
generate_weight
,
(
768
,
768
)
)
),
"mul4_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight1
)
data_gen
=
partial
(
generate_weight
,
(
768
,
768
)
)
),
"elementwise_add1_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight2
)
data_gen
=
partial
(
generate_weight
,
(
768
,)
)
),
"elementwise_add2_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight3
)
data_gen
=
partial
(
generate_weight
,
(
768
,
768
)
)
),
"elementwise_add3_weight"
:
TensorConfig
(
data_gen
=
partial
(
generate_weight2
)
data_gen
=
partial
(
generate_weight
,
(
768
,)
)
),
},
inputs
=
{
"input_data1"
:
TensorConfig
(
data_gen
=
partial
(
generate_input
1
,
batch
,
dim1
generate_input
,
(
batch
,
dim1
,
768
)
)
),
"input_data2"
:
TensorConfig
(
data_gen
=
partial
(
generate_input
2
,
input2_shape
generate_input
,
input2_shape
)
),
},
...
...
@@ -1335,9 +1365,9 @@ class TrtConvertMultiHeadMatmulTest_biasqk_seqseq(TrtLayerAutoScanTest):
generate_dynamic_shape
(
attrs
)
self
.
trt_param
.
precision
=
paddle_infer
.
PrecisionType
.
Float32
self
.
trt_param
.
workspace_size
=
2013265920
yield
self
.
create_inference_config
(),
(
1
,
3
),
(
1e-5
,
1e-
4
)
yield
self
.
create_inference_config
(),
(
1
,
3
),
(
1e-5
,
1e-
5
)
self
.
trt_param
.
precision
=
paddle_infer
.
PrecisionType
.
Half
yield
self
.
create_inference_config
(),
(
1
,
3
),
(
1e-
3
,
1e-2
)
yield
self
.
create_inference_config
(),
(
1
,
3
),
(
1e-
2
,
1e-2
)
def
test
(
self
):
self
.
run_test
()
...
...
test/legacy_test/test_math_op_patch.py
浏览文件 @
7eeff7b1
...
...
@@ -23,7 +23,9 @@ from paddle import fluid
class
TestMathOpPatches
(
unittest
.
TestCase
):
@
classmethod
def
setUp
(
self
):
np
.
random
.
seed
(
1024
)
paddle
.
enable_static
()
@
prog_scope
()
...
...
test/quantization/test_post_training_quantization_mobilenetv1.py
浏览文件 @
7eeff7b1
...
...
@@ -591,7 +591,7 @@ class TestPostTrainingAvgONNXFormatForMobilenetv1TensorRT(
):
def
test_post_training_onnx_format_mobilenetv1_tensorrt
(
self
):
model
=
"MobileNet-V1"
algo
=
"
avg
"
algo
=
"
KL
"
round_type
=
"round"
data_urls
=
[
'https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/inference/MobileNetV1_infer.tar'
...
...
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录