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
飞不过沧海
Paddle
提交
bd4ce23e
P
Paddle
项目概览
飞不过沧海
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
1
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
0
列表
看板
标记
里程碑
合并请求
0
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
0
Issue
0
列表
看板
标记
里程碑
合并请求
0
合并请求
0
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
bd4ce23e
编写于
9月 08, 2022
作者:
L
Leo Chen
提交者:
GitHub
9月 08, 2022
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Increase the threshold of softmax and imperative qat UT (#45819)
上级
36046a89
变更
4
显示空白变更内容
内联
并排
Showing
4 changed file
with
11 addition
and
8 deletion
+11
-8
python/paddle/fluid/contrib/slim/tests/test_imperative_qat.py
...on/paddle/fluid/contrib/slim/tests/test_imperative_qat.py
+6
-3
python/paddle/fluid/contrib/slim/tests/test_imperative_qat_channelwise.py
...uid/contrib/slim/tests/test_imperative_qat_channelwise.py
+2
-2
python/paddle/fluid/contrib/slim/tests/test_imperative_qat_fuse.py
...ddle/fluid/contrib/slim/tests/test_imperative_qat_fuse.py
+1
-1
python/paddle/fluid/tests/unittests/ir/inference/test_trt_convert_softmax.py
.../tests/unittests/ir/inference/test_trt_convert_softmax.py
+2
-2
未找到文件。
python/paddle/fluid/contrib/slim/tests/test_imperative_qat.py
浏览文件 @
bd4ce23e
...
...
@@ -56,7 +56,10 @@ class TestImperativeQat(unittest.TestCase):
self
.
activation_quantize_type
=
'moving_average_abs_max'
self
.
onnx_format
=
False
self
.
check_export_model_accuracy
=
True
self
.
diff_threshold
=
0.01
# The original model and quantized model may have different prediction.
# There are 32 test data and we allow at most one is different.
# Hence, the diff_threshold is 1 / 32 = 0.03125
self
.
diff_threshold
=
0.03125
self
.
fuse_conv_bn
=
False
def
func_qat
(
self
):
...
...
@@ -207,7 +210,7 @@ class TestImperativeQat(unittest.TestCase):
quant_acc
=
fluid
.
layers
.
accuracy
(
quant_out
,
label
).
numpy
()
paddle
.
enable_static
()
delta_value
=
fp32_acc
-
quant_acc
self
.
assertLess
(
delta_value
,
self
.
diff_threshold
)
self
.
assertLess
Equal
(
delta_value
,
self
.
diff_threshold
)
def
test_qat
(
self
):
with
_test_eager_guard
():
...
...
@@ -221,7 +224,7 @@ class TestImperativeQatONNXFormat(unittest.TestCase):
self
.
weight_quantize_type
=
'abs_max'
self
.
activation_quantize_type
=
'moving_average_abs_max'
self
.
onnx_format
=
True
self
.
diff_threshold
=
0.025
self
.
diff_threshold
=
0.0
31
25
self
.
fuse_conv_bn
=
False
...
...
python/paddle/fluid/contrib/slim/tests/test_imperative_qat_channelwise.py
浏览文件 @
bd4ce23e
...
...
@@ -43,7 +43,7 @@ class TestImperativeQatChannelWise(TestImperativeQat):
def
set_vars
(
self
):
self
.
weight_quantize_type
=
'channel_wise_abs_max'
self
.
activation_quantize_type
=
'moving_average_abs_max'
self
.
diff_threshold
=
0.0
1
self
.
diff_threshold
=
0.0
3125
self
.
onnx_format
=
False
self
.
fuse_conv_bn
=
False
print
(
'weight_quantize_type'
,
self
.
weight_quantize_type
)
...
...
@@ -55,7 +55,7 @@ class TestImperativeQatChannelWiseONNXFormat(TestImperativeQat):
self
.
weight_quantize_type
=
'channel_wise_abs_max'
self
.
activation_quantize_type
=
'moving_average_abs_max'
self
.
onnx_format
=
True
self
.
diff_threshold
=
0.025
self
.
diff_threshold
=
0.0
31
25
self
.
fuse_conv_bn
=
False
print
(
'weight_quantize_type'
,
self
.
weight_quantize_type
)
...
...
python/paddle/fluid/contrib/slim/tests/test_imperative_qat_fuse.py
浏览文件 @
bd4ce23e
...
...
@@ -43,7 +43,7 @@ class TestImperativeQatfuseBN(TestImperativeQat):
def
set_vars
(
self
):
self
.
weight_quantize_type
=
'abs_max'
self
.
activation_quantize_type
=
'moving_average_abs_max'
self
.
diff_threshold
=
0.0
1
self
.
diff_threshold
=
0.0
3125
self
.
onnx_format
=
False
self
.
fuse_conv_bn
=
True
...
...
python/paddle/fluid/tests/unittests/ir/inference/test_trt_convert_softmax.py
浏览文件 @
bd4ce23e
...
...
@@ -126,7 +126,7 @@ class TrtConvertSoftmaxTest(TrtLayerAutoScanTest):
attrs
,
False
),
1e-5
self
.
trt_param
.
precision
=
paddle_infer
.
PrecisionType
.
Half
yield
self
.
create_inference_config
(),
generate_trt_nodes_num
(
attrs
,
False
),
1e-
5
attrs
,
False
),
1e-
3
# for dynamic_shape
generate_dynamic_shape
(
attrs
)
...
...
@@ -135,7 +135,7 @@ class TrtConvertSoftmaxTest(TrtLayerAutoScanTest):
attrs
,
True
),
1e-5
self
.
trt_param
.
precision
=
paddle_infer
.
PrecisionType
.
Half
yield
self
.
create_inference_config
(),
generate_trt_nodes_num
(
attrs
,
True
),
1e-
5
attrs
,
True
),
1e-
3
def
test
(
self
):
self
.
run_test
()
...
...
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