Commits · f4dd2cfffa934093f41e9ea5b2e6f00fc41a3021 · ModelZoo / ResNet50_tensorflow

12 Dec, 2019 1 commit
- Internal change · f4dd2cff
  Yeqing Li authored Dec 12, 2019
```
PiperOrigin-RevId: 285223338
```
  f4dd2cff
06 Dec, 2019 1 commit
- Internal cleanup. · 8a8baded
  Yeqing Li authored Dec 06, 2019
```
PiperOrigin-RevId: 284259997
```
  8a8baded
05 Dec, 2019 1 commit

Remove tweaked benchmark for no_dist_strat in function mode. · ef8aed79

Haoyu Zhang authored Dec 04, 2019

No need to test explicit_gpu_placement + no_dist_strat + function case, since the performance issue is already fixed by unifying the Keras codepaths. However, there is still performance gap using the explicit_gpu_placement flag when run_eagerly.

PiperOrigin-RevId: 283890008

ef8aed79

04 Dec, 2019 1 commit
- Internal change · 6a50c338
  A. Unique TensorFlower authored Dec 03, 2019
```
PiperOrigin-RevId: 283648734
```
  6a50c338
03 Dec, 2019 1 commit
- Fix bug introduced. · 281d1757
  Hongkun Yu authored Dec 02, 2019
```
PiperOrigin-RevId: 283447458
```
  281d1757
27 Nov, 2019 1 commit
- Remove 'default' in get_distribution_strategy which is complex and error-prone · 04256053
  Hongkun Yu authored Nov 26, 2019
```
PiperOrigin-RevId: 282669615
```
  04256053
25 Nov, 2019 2 commits

Inject enable_runtime_flags into benchmarks. · bcce419a

Sai Ganesh Bandiatmakuri authored Nov 25, 2019

This will help general debugging by enabling custom execution with --benchmark_method_steps.

E.g --benchmark_method_steps=train_steps=7 will run the benchmark for only 7 steps without modifying benchmark code.

PiperOrigin-RevId: 282396875

bcce419a

Completely remove "use_keras_bert_for_squad" FLAG. · 9c50e961
Chen Chen authored Nov 25, 2019
```
PiperOrigin-RevId: 282384787
```
9c50e961

22 Nov, 2019 2 commits
- Internal change · 082bb878
  David Chen authored Nov 22, 2019
```
PiperOrigin-RevId: 281977686
```
  082bb878
- Internal change · a23beab1
  David Chen authored Nov 21, 2019
```
PiperOrigin-RevId: 281908068
```
  a23beab1
20 Nov, 2019 1 commit
- Internal change · e16594d1
  Chen Chen authored Nov 20, 2019
```
PiperOrigin-RevId: 281473612
```
  e16594d1
12 Nov, 2019 1 commit
- Internal change · a76237da
  Rajagopal Ananthanarayanan authored Nov 12, 2019
```
PiperOrigin-RevId: 280019807
```
  a76237da
11 Nov, 2019 2 commits
- Fix checkpoint path in bert_benchmark.py. · 0faa2ff3
  Hongkun Yu authored Nov 11, 2019
```
PiperOrigin-RevId: 279768972
```
  0faa2ff3
- Release keras bert: · f1d35b4e
  Hongkun Yu authored Nov 11, 2019
```
- Update classifier example.
- Add new converted checkpoints.
- Update benchmark,

PiperOrigin-RevId: 279762797
```
  f1d35b4e
08 Nov, 2019 1 commit
- Internal change · 4972c084
  A. Unique TensorFlower authored Nov 08, 2019
```
PiperOrigin-RevId: 279361336
```
  4972c084
06 Nov, 2019 2 commits
- Internal change · bf0dc049
  Ayush Dubey authored Nov 06, 2019
```
PiperOrigin-RevId: 278902210
```
  bf0dc049
- Internal change · 94833324
  Yeqing Li authored Nov 05, 2019
```
PiperOrigin-RevId: 278756245
```
  94833324
02 Nov, 2019 1 commit
- Internal change · b3e74d15
  Hongkun Yu authored Nov 01, 2019
```
PiperOrigin-RevId: 278028560
```
  b3e74d15
01 Nov, 2019 2 commits

Support fp16 using tf.keras.mixed_precision in CTL resnet. · ffa522ea

Reed Wanderman-Milne authored Nov 01, 2019

To test, I ran the following command:

python resnet_ctl_imagenet_main.py --batch_size=2048 --data_dir ~/imagenet --datasets_num_private_threads=14 --epochs_between_evals=10 --model_dir ~/tmp_model_dir --clean --num_gpus=8 --train_epochs=90 --dtype=fp16

I got 76.15% final evaluation accuracy.

PiperOrigin-RevId: 278010061

ffa522ea

Add squad xlnet accuracy test · 9df6a3d6
Hongkun Yu authored Nov 01, 2019
```
PiperOrigin-RevId: 277992916
```
9df6a3d6

29 Oct, 2019 1 commit
- Internal change · 9f3b0269
  David Chen authored Oct 29, 2019
```
PiperOrigin-RevId: 277332242
```
  9f3b0269
28 Oct, 2019 1 commit
- Add Resnet50 benchmark suite that read training data from remote storage · 06f22a59
  Zongwei Zhou authored Oct 28, 2019
```
PiperOrigin-RevId: 277082247
```
  06f22a59
22 Oct, 2019 1 commit
- train_data_size flag is deleted. · 128326ca
  Hongkun Yu authored Oct 21, 2019
```
PiperOrigin-RevId: 275986990
```
  128326ca
18 Oct, 2019 1 commit
- Move CTL resnet example. · f2c61881
  Hongkun Yu authored Oct 18, 2019
```
PiperOrigin-RevId: 275417626
```
  f2c61881
17 Oct, 2019 1 commit
- Move ResNet CTL benchmark to benchmark folder. · d2f86342
  Hongkun Yu authored Oct 17, 2019
```
PiperOrigin-RevId: 275288636
```
  d2f86342
16 Oct, 2019 3 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
- Internal change · e97e22df
  Yeqing Li authored Oct 15, 2019
```
PiperOrigin-RevId: 274921478
```
  e97e22df
11 Oct, 2019 1 commit

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

10 Oct, 2019 1 commit
- Perfzero RetinaLNet detection COCO AP test on 8 GPUs. · b63030a3
  Yeqing Li authored Oct 10, 2019
```
PiperOrigin-RevId: 274035928
```
  b63030a3
09 Oct, 2019 1 commit
- Fix BERT benchmark test. · f5e2211f
  A. Unique TensorFlower authored Oct 09, 2019
```
PiperOrigin-RevId: 273795511
```
  f5e2211f
01 Oct, 2019 1 commit
- Internal change · 6bbc45dd
  David Chen authored Sep 30, 2019
```
PiperOrigin-RevId: 272121528
```
  6bbc45dd
26 Sep, 2019 1 commit
- Internal change · d735ecf8
  David Chen authored Sep 26, 2019
```
PiperOrigin-RevId: 271422100
```
  d735ecf8
24 Sep, 2019 1 commit
- Perfzero XLNet classifier Imdb accuracy test on 8 GPUs. · a52564cb
  Hongkun Yu authored Sep 23, 2019
```
PiperOrigin-RevId: 270817869
```
  a52564cb
19 Sep, 2019 2 commits
- Fix for resnet data_dir in the keras_imagenet_benchmark suite. · f28059c0
  A. Unique TensorFlower authored Sep 19, 2019
```
PiperOrigin-RevId: 270122397
```
  f28059c0
- Fix for resnet data_dir name in copybara. · 0a79175b
  A. Unique TensorFlower authored Sep 19, 2019
```
PiperOrigin-RevId: 270079001
```
  0a79175b
18 Sep, 2019 1 commit
- Move bert benchmark stuff to a central place. · 7bd795db
  Hongkun Yu authored Sep 18, 2019
```
PiperOrigin-RevId: 269835917
```
  7bd795db
13 Sep, 2019 1 commit
- Set `datasets_num_private_threads` flag for multi-worker tuned benchmarks. · 956f295f
  Ayush Dubey authored Sep 12, 2019
```
PiperOrigin-RevId: 268836696
```
  956f295f
12 Sep, 2019 1 commit
- Add Keras-based ResNet50 multi-worker accuracy tests. · cbf29854
  Ayush Dubey authored Sep 12, 2019
```
PiperOrigin-RevId: 268703439
```
  cbf29854
29 Aug, 2019 1 commit

Use new mixed_float16 policy for resnet. · dcd0e7ad

Reed Wanderman-Milne authored Aug 29, 2019

The old infer_float32_policies policy will be removed from TensorFlow soon.

To test convergence, I ran the Resnet50KerasAccuracy.benchmark_8_gpu_fp16 benchmark. I got an accuracy of 0.76037 and an exp_per_second of 6908.

PiperOrigin-RevId: 266191126

dcd0e7ad