Skip to content

GitLab

Switch branch/tag
  1. 30 Mar, 2023 1 commit
  3. 06 Jun, 2021 4 commits
  5. 10 Apr, 2021 1 commit
    • Reed Wanderman-Milne's avatar
      Remove dynamic_loss_scale argument to define_performance. · 3803472a
      Reed Wanderman-Milne authored 
      All models which support loss scaling support dynamic loss scaling, so the argument has no purpose. It used to be that some models scaled the loss manually instead of using a LossScaleOptimizer, and so did not support dynamic loss scaling.

PiperOrigin-RevId: 367719521
      3803472a
  7. 09 Apr, 2021 1 commit
    • Reed Wanderman-Milne's avatar
      Remove dynamic_loss_scale argument to define_performance. · e353e4e5
      Reed Wanderman-Milne authored 
      All models which support loss scaling support dynamic loss scaling, so the argument has no purpose. It used to be that some models scaled the loss manually instead of using a LossScaleOptimizer, and so did not support dynamic loss scaling.

PiperOrigin-RevId: 367719521
      e353e4e5
  9. 06 Apr, 2021 2 commits
  11. 28 Feb, 2021 4 commits
  13. 24 Jan, 2021 2 commits
  15. 12 Aug, 2020 2 commits
  17. 29 Apr, 2020 1 commit
  19. 25 Apr, 2020 1 commit
  21. 22 Apr, 2020 1 commit
  23. 17 Mar, 2020 1 commit
  25. 05 Mar, 2020 1 commit
  27. 02 Mar, 2020 1 commit
  29. 25 Feb, 2020 1 commit
  31. 27 Nov, 2019 1 commit
  33. 28 Oct, 2019 1 commit
  35. 21 Oct, 2019 1 commit
  37. 16 Oct, 2019 1 commit
    • Reed Wanderman-Milne's avatar
      Add support for the tf.keras.mixed_precision API in NCF · cb913691
      Reed Wanderman-Milne authored 
      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
  39. 07 Oct, 2019 1 commit
  41. 09 Sep, 2019 1 commit
  43. 04 Sep, 2019 1 commit
  45. 30 Aug, 2019 1 commit
  47. 26 Aug, 2019 1 commit
  49. 23 Aug, 2019 1 commit
  51. 20 Aug, 2019 2 commits
  53. 19 Aug, 2019 1 commit
    • Reed Wanderman-Milne's avatar
      Do not expose --max_train_steps in models that do not use it. · 824ff2d6
      Reed Wanderman-Milne authored 
      Only the V1 resnet model uses --max_train_steps. This unexposes the flag in the keras_application_models, mnist, keras resnet, CTL resnet Models. Before this change, such models allowed the flag to be specified, but ignored it.

I also removed the "max_train" argument from the run_synthetic function, since this only had any meaning for the V1 resnet model. Instead, the V1 resnet model now directly passes --max_train_steps=1 to run_synthetic.

PiperOrigin-RevId: 264269836
      824ff2d6
  55. 16 Aug, 2019 1 commit
    • Ayush Dubey's avatar
      Add multi-worker benchmarks to Keras ResNet model. · ff6c3b1e
      Ayush Dubey authored 
      Also add `worker_hosts` and `task_index` flags.  These flags enable running the
model over multiple hosts by passing the cluster information via command line.

Setting `TF_CONFIG` will continue to work.

PiperOrigin-RevId: 263825245
      ff6c3b1e
  57. 06 Aug, 2019 1 commit
  59. 23 Jul, 2019 1 commit