提交 · 12bffdc08637cacb88bace6a5f276ac9b2550138 · BaiXuePrincess / Paddle

02 6月, 2020 1 次提交
- W
  
  add option to exclude ops by id from quantization (#24689) · 78d4f0cc
  由 Wojciech Uss 提交于 6月 02, 2020
  
  78d4f0cc
27 5月, 2020 1 次提交
- C
  
  Update sigmoid output from Y to out, test=develop (#24765) · dbcd7c69
  由 cc 提交于 5月 27, 2020
  
  dbcd7c69
26 5月, 2020 2 次提交
- C
  
  Collecting concat output threshold, test=develop (#24742) · 88e9d74a
  由 cc 提交于 5月 26, 2020
  
  88e9d74a
- C
  
  Add output threshold for ops that have several output activations, test=develop (#24726) · 6c89ca21
  由 cc 提交于 5月 26, 2020
  
  6c89ca21
14 5月, 2020 1 次提交
- L
  Update DNNL QAT document 2.0-alpha (#24494) · 8ef3c02e
  由 lidanqing 提交于 5月 14, 2020
```
Update DNNL QAT document 2.0-alpha
```
  8ef3c02e
13 5月, 2020 1 次提交

[Fix bug] Init scale node in OutScaleForTrainingPass and enable... · 4d351122

由 cc 提交于 5月 13, 2020

[Fix bug] Init scale node in OutScaleForTrainingPass and enable test_quantization_scale_pass UT (#24393)

* Init scale node in OutScaleForTrainingPass, test=develop
* Enable test_quantization_scale, test=develop

4d351122

12 5月, 2020 1 次提交
- J
  
  Model converter to dot file (#23169) · 53125c2f
  由 joanna.wozna.intel 提交于 5月 12, 2020
  
  53125c2f
08 5月, 2020 1 次提交
- W
  Enabled quantize all and skip missing in QAT (#24281) · db052009
  由 Wojciech Uss 提交于 5月 08, 2020
```
* Enabled quantize all and skip missing in QAT
```
  db052009
28 4月, 2020 2 次提交

Update QAT INT8 2.0 doc (#24127) · 61ec30f0

由 lidanqing 提交于 4月 28, 2020

* update local data preprocess doc

* update for 2.0 QAT

test=develop

test=document_fix

* update benchmark data

test=develop

test=document_fix
Co-authored-by: NWojciech Uss <wojciech.uss@intel.com>

61ec30f0

S

added reshape transpose matmul fuse pass (#23754) · e1a7a880
由 Sylwester Fraczek 提交于 4月 28, 2020

e1a7a880

24 4月, 2020 1 次提交
- A
  
  added fusing matmul-transpose-reshape pass (#23866) · d31a174f
  由 arlesniak 提交于 4月 24, 2020
  
  d31a174f
23 4月, 2020 1 次提交

QAT: support for new models (#23928) · 3d744162

由 Wojciech Uss 提交于 4月 23, 2020

* QAT: support range-based quantization and scales from attribute

* added support for channelwise

3d744162

17 4月, 2020 1 次提交
- C
  Weight quantization support channel_wise_abs_max method to achieve higher accuracy (#23629) · 40aa14ec
  由 cc 提交于 4月 17, 2020
```
* Weight quantization support channel_wise_abs_max method to achieve higher accuracy
```
  40aa14ec
13 4月, 2020 1 次提交
- J
  
  Add scale-matmul fuse pass (#23734) · 12ba05ce
  由 joanna.wozna.intel 提交于 4月 13, 2020
  
  12ba05ce
11 4月, 2020 1 次提交
- W
  [Doc update] Update for QAT INT8 MKL-DNN document (#23361) · 2383a9f7
  由 Wojciech Uss 提交于 4月 11, 2020
```
* Update for QAT INT8 MKL-DNN document, added info on VNNI in Windows, benchmark results added and updated
```
  2383a9f7
10 4月, 2020 1 次提交
- W
  
  Enable matmul and cleanup in QAT2 (#23657) · 1753860d
  由 Wojciech Uss 提交于 4月 10, 2020
  
  1753860d
07 4月, 2020 2 次提交
- C
  Collect output scale for quantized op and fused op (#23369) · 25628587
  由 cc 提交于 4月 07, 2020
```
* Collect output scale for quantized op and fused op
* Post_training_quantizaion sets batch_generator to support lod tensor
```
  25628587
- B
  
  fix test_graph_wrapper failure on cudnnv7, test=develop (#23451) · 9bc223c8
  由 Bai Yifan 提交于 4月 07, 2020
  
  9bc223c8
03 4月, 2020 3 次提交
- C
  
  Set fuse_all_reduce_ops=false for quantization test, test=develop (#23413) · 3ea7c59f
  由 cc 提交于 4月 03, 2020
  
  3ea7c59f
- C
  
  Disable test_quantization_scale_pass unittest for random error, test=develop (#23441) · 7c55a94d
  由 cc 提交于 4月 03, 2020
  
  7c55a94d
- Y
  
  Disable test_code_generator and test_post_training_quantization_mobilenetv1 (#23440) · bc2981e9
  由 Yiqun Liu 提交于 4月 03, 2020
  
  bc2981e9
02 4月, 2020 1 次提交
- W
  
  handle conv2d activations in older QAT models (#23202) · 9fd90674
  由 Wojciech Uss 提交于 4月 01, 2020
  
  9fd90674
29 3月, 2020 1 次提交
- W
  
  separated QAT1 and QAT2 (#23284) · be2ac9cc
  由 Wojciech Uss 提交于 3月 29, 2020
  
  be2ac9cc
28 3月, 2020 2 次提交
- L
  
  Update QAT INT8 related code (#23104) · c524b930
  由 lidanqing 提交于 3月 28, 2020
  
  c524b930
- W
  
  add check for scales and a message (#23119) · f836c8aa
  由 Wojciech Uss 提交于 3月 28, 2020
  
  f836c8aa
26 3月, 2020 1 次提交
- C
  
  Add activation_type in AddQuantDequantPass to be compatible with paddleslim, test=develop (#23221) · bd809033
  由 cc 提交于 3月 26, 2020
  
  bd809033
24 3月, 2020 1 次提交
- C
  Post_training_quantizaion supports min_max methon (#23078) · 589cd878
  由 cc 提交于 3月 24, 2020
```
* Post_training_quantizaion supports min_max methon
```
  589cd878
11 3月, 2020 1 次提交
- L
  
  Changes QAT MKL-DNN documents (#22840) · 432a4b27
  由 lidanqing 提交于 3月 11, 2020
  
  432a4b27
05 3月, 2020 1 次提交
- C
  Added an option to use external FP32 model in QAT comparison test (#22858) · b6717faf
  由 cc 提交于 3月 05, 2020
```
* Added an option to use external FP32 model in QAT comparison test
```
  b6717faf
04 3月, 2020 1 次提交
- S
  
  fixed save qat2 model resnet50 and ernie (#22822) · 5ff2439f
  由 Sylwester Fraczek 提交于 3月 04, 2020
  
  5ff2439f
23 2月, 2020 1 次提交
- T
  
  fix typo words (#22653) · d2ba91aa
  由 tianshuo78520a 提交于 2月 23, 2020
  
  d2ba91aa
20 2月, 2020 1 次提交
- W
  
  Fix batch_num default value (#22682) · 405bb94b
  由 Wojciech Uss 提交于 2月 20, 2020
  
  405bb94b
19 2月, 2020 1 次提交
- B
  
  fix ernie accuracy doc (#22671) · 6abeb5c5
  由 bingyanghuang 提交于 2月 19, 2020
  
  6abeb5c5
18 2月, 2020 1 次提交

Doc update with Ernie QAT INT8 benchmarking (#22519) · fce37bc5

由 Wojciech Uss 提交于 2月 18, 2020

* Doc update with Ernie QAT INT8 benchmarking

test=develop

* fixes after review

test=develop

* remove ernie part, test=develop test=document_fix

* Fix model name for qatv2

test=develop test=document_fix

* Add Ernie data

test=develop test=document_fix

* update ERNIE benchmark with baidu QA results, test=develop test=document_fix
Co-authored-by: Nbingyanghuang <33643817+bingyanghuang@users.noreply.github.com>
Co-authored-by: NMichał Gallus <sand3r@interia.eu>

fce37bc5

12 2月, 2020 1 次提交

Add support for Ernie NLP model to the Slim QAT (#22506) · 4cddb43c

由 Wojciech Uss 提交于 2月 12, 2020

* a test for Ernie QAT INT8 accuracy check

test=develop

* Remove NLP comparison test to split PRs

test=develop

* Fix typo and tabs, delete commented lines

test=develop

* re-combine the 2 PRs, test=develop
Co-authored-by: NMichał Gallus <sand3r@interia.eu>
Co-authored-by: Nbingyanghuang <33643817+bingyanghuang@users.noreply.github.com>

4cddb43c

10 2月, 2020 1 次提交
- C
  Post_training_quantization support set quant 8/16 bits (#22492) · d143f70a
  由 cc 提交于 2月 10, 2020
```
* post_training_quantization support set bits, test=develop

* up, test=develop
```
  d143f70a
07 2月, 2020 1 次提交

Add weight quantization in post_training_quanzitaion (#22445) · 197913eb

由 cc 提交于 2月 07, 2020

* support weight quantization in post_training_quanzitaion, test=develop
* add test for weight quantization, test=develop

197913eb

25 1月, 2020 1 次提交
- J
  
  Restore requantize squash (#22399) · 3099d9d4
  由 joanna.wozna.intel 提交于 1月 25, 2020
  
  3099d9d4
16 1月, 2020 1 次提交
- J
  
  fix use cache file, test=develop (#22240) · b339dff2
  由 juncaipeng 提交于 1月 16, 2020
  
  b339dff2
15 1月, 2020 1 次提交
- J
  add mul and matmul quantization, test=develop (#22054) · 8f7372ca
  由 juncaipeng 提交于 1月 15, 2020
```
* add mul and matmul quantization, test=develop
* add test for matmul, test=develop
```
  8f7372ca

BaiXuePrincess / Paddle 与 Fork 源项目一致

BaiXuePrincess / Paddle
与 Fork 源项目一致