Commits · 9691ef7a2a64c39b3096c34a8fba0e168175cb82 · ModelZoo / ResNet50_tensorflow

"tests/models/segformer/__init__.py" did not exist on "81d6841b4be25a164235975e5ebdcf99d7a26633"

17 Oct, 2019 1 commit
- Add compat.v1 to support TF 2.0 in mnist · 9691ef7a
  minoring authored Oct 16, 2019
  
  9691ef7a
16 Oct, 2019 4 commits

Internal change · 06412123
David Chen authored Oct 16, 2019
```
PiperOrigin-RevId: 275103426
```
06412123
Enables timer callback and disables checkpoint saving in retinanet benchmark test. · 1b77cd80
Yeqing Li authored Oct 16, 2019
```
PiperOrigin-RevId: 275080469
```
1b77cd80

Add support for the tf.keras.mixed_precision API in NCF · cb913691

Reed Wanderman-Milne authored Oct 16, 2019

To test, I did 50 fp32 runs and 50 fp16 runs. I used the following command:

python ncf_keras_main.py --dataset=ml-20m --num_gpus=1 --train_epochs=10 --clean --batch_size=99000 --learning_rate=0.00382059 --beta1=0.783529 --beta2=0.909003 --epsilon=1.45439e-7 --layers=256,256,128,64 --num_factors=64 --hr_threshold=0.635 --ml_perf --nouse_synthetic_data --data_dir ~/ncf_data_dir_python3 --model_dir ~/tmp_model_dir --keras_use_ctl

For the fp16 runs, I added --dtype=fp16. The average hit-rate for both fp16 and fp32 was 0.6365. I also did 50 runs with the mixed precision graph rewrite, and the average hit-rate was 0.6363. The difference is likely due to noise.

PiperOrigin-RevId: 275059871

cb913691

Internal change · e97e22df
Yeqing Li authored Oct 15, 2019
```
PiperOrigin-RevId: 274921478
```
e97e22df

15 Oct, 2019 6 commits
- Internal change · 32229eae
  Yeqing Li authored Oct 15, 2019
```
PiperOrigin-RevId: 274918820
```
  32229eae
- Update resnet hub export · 63e20270
  Hongkun Yu authored Oct 15, 2019
```
PiperOrigin-RevId: 274917111
```
  63e20270
- Export the pretrained transformer-xl model for finetuning stage. · 048f5a95
  Jing Li authored Oct 15, 2019
```
Add to option to init checkpoint from transformer-xl model.

PiperOrigin-RevId: 274875006
```
  048f5a95
- Adds assets for bert export · 188536e7
  Hongkun Yu authored Oct 15, 2019
```
PiperOrigin-RevId: 274844449
```
  188536e7
- Internal change · 9af441a4
  Yeqing Li authored Oct 15, 2019
```
PiperOrigin-RevId: 274807747
```
  9af441a4
- Fix variable mismatch between XLNet pretrain and finetune tasks. · ddd45b81
  Jing Li authored Oct 14, 2019
```
PiperOrigin-RevId: 274699918
```
  ddd45b81
14 Oct, 2019 1 commit
- Adds reference section to README. · bb7d95ab
  Yeqing Li authored Oct 14, 2019
```
PiperOrigin-RevId: 274642627
```
  bb7d95ab
13 Oct, 2019 2 commits
- Internal change · dc4e7c4f
  A. Unique TensorFlower authored Oct 13, 2019
```
PiperOrigin-RevId: 274460885
```
  dc4e7c4f
- Internal change · 9a833e2c
  Hongkun Yu authored Oct 12, 2019
```
PiperOrigin-RevId: 274386468
```
  9a833e2c
12 Oct, 2019 2 commits
- Internal change · d0600d40
  A. Unique TensorFlower authored Oct 12, 2019
```
PiperOrigin-RevId: 274347990
```
  d0600d40
- Internal change · dccc21dd
  Rajagopal Ananthanarayanan authored Oct 11, 2019
```
PiperOrigin-RevId: 274281911
```
  dccc21dd
11 Oct, 2019 8 commits
- Revert "Revert "Update usage of tf.contrib.data to tf.data.experimental" (#7654)" · 35daa566
  Hongkun Yu authored Oct 11, 2019
```
This reverts commit b4e560dc.
```
  35daa566
- Revert "Update usage of tf.contrib.data to tf.data.experimental" (#7654) · b4e560dc
  Hongkun Yu authored Oct 11, 2019
```
* Revert "Update tf.contrib.data to tf.data.experimental. (#7650)"

This reverts commit faf4bbb3.

* revert research
```
  b4e560dc
- Update tf.contrib.data to tf.data.experimental. (#7650) · faf4bbb3
  Derek Murray authored Oct 11, 2019
  
  faf4bbb3
- Internal change · cb136c62
  Yeqing Li authored Oct 11, 2019
```
PiperOrigin-RevId: 274241934
```
  cb136c62
- Change summary directory and model checkpoint directory so that training via... · e67a2064
  A. Unique TensorFlower authored Oct 11, 2019
```
Change summary directory and model checkpoint directory so that training via Keras Compile/Fit() and custom training loop is consistent.

PiperOrigin-RevId: 274202793
```
  e67a2064
- Internal changes. · ad1a37c9
  Hongkun Yu authored Oct 11, 2019
```
PiperOrigin-RevId: 274201399
```
  ad1a37c9
- Bert Hub usage for squad · 223f5148
  Hongkun Yu authored Oct 10, 2019
```
PiperOrigin-RevId: 274090672
```
  223f5148
- Do not show the percent symbol for accuracy, as the value is between 0 and 1, not 0 and 100 · 9f8e0646
  Reed Wanderman-Milne authored Oct 10, 2019
```
PiperOrigin-RevId: 274090348
```
  9f8e0646
10 Oct, 2019 8 commits
- change benchmark's log verbosity to logging.INFO. it seems to me that DEBUG... · 1f9efe10
  A. Unique TensorFlower authored Oct 10, 2019
```
change benchmark's log verbosity to logging.INFO. it seems to me that DEBUG map to ---v=1 internally, which is way to verbose for the purpose of benchmarking.

PiperOrigin-RevId: 274040907
```
  1f9efe10
- Perfzero RetinaLNet detection COCO AP test on 8 GPUs. · b63030a3
  Yeqing Li authored Oct 10, 2019
```
PiperOrigin-RevId: 274035928
```
  b63030a3
- Internal change · adc3e31f
  A. Unique TensorFlower authored Oct 10, 2019
```
PiperOrigin-RevId: 274028786
```
  adc3e31f
- Add __init__.py. · 26317fe3
  Yeqing Li authored Oct 10, 2019
```
PiperOrigin-RevId: 274023277
```
  26317fe3
- experimental_allow_stateful is not necessary with make dataset fn · 399b778f
  Hongkun Yu authored Oct 10, 2019
```
PiperOrigin-RevId: 274015143
```
  399b778f
- Move retinanet keras model to tensorflow_models/official · bcb231f0
  Yeqing Li authored Oct 10, 2019
```
PiperOrigin-RevId: 274010788
```
  bcb231f0
- Internal change · cf4cae61
  Hongkun Yu authored Oct 10, 2019
```
PiperOrigin-RevId: 273966871
```
  cf4cae61
- experimental_allow_stateful is not necessary with make dataset fn · 8110bb64
  Hongkun Yu authored Oct 09, 2019
```
PiperOrigin-RevId: 273861263
```
  8110bb64
09 Oct, 2019 3 commits

Simply LayerNorm mixed precision logic. · 0257b276

Reed Wanderman-Milne authored Oct 09, 2019

Instead of needing to ensure variables are float32, casting inputs to float32, etc, instead dtype="float32" is passed to the layer constructor, which will do all that logic automatically.

The only difference is the output of LayerNorm is now float32 instead of float16, so an extra cast is needed elsewhere.

PiperOrigin-RevId: 273833286

0257b276

Fix BERT benchmark test. · f5e2211f
A. Unique TensorFlower authored Oct 09, 2019
```
PiperOrigin-RevId: 273795511
```
f5e2211f
Internal change · 7207422d
A. Unique TensorFlower authored Oct 08, 2019
```
PiperOrigin-RevId: 273653001
```
7207422d

08 Oct, 2019 2 commits
- Internal change · dc93d9e5
  A. Unique TensorFlower authored Oct 08, 2019
```
PiperOrigin-RevId: 273562498
```
  dc93d9e5
- Port shakespeare benchmark to TF2 · d0f21f25
  George Karpenkov authored Oct 08, 2019
```
PiperOrigin-RevId: 273527676
```
  d0f21f25
07 Oct, 2019 3 commits
- Bert fp16 perf improvements, do the matmul in intermediate later in fp16, and... · 3fa65606
  A. Unique TensorFlower authored Oct 07, 2019
```
Bert fp16 perf improvements, do the matmul in intermediate later in fp16, and also remove explicit casting to fp32 for layerNorm.

PiperOrigin-RevId: 273379063
```
  3fa65606
- Internal change · 41293260
  A. Unique TensorFlower authored Oct 07, 2019
```
PiperOrigin-RevId: 273371605
```
  41293260
- support TPU on resnet50 using keras compile/fit · fa5b66e5
  Jing Li authored Oct 07, 2019
```
PiperOrigin-RevId: 273358759
```
  fa5b66e5