Search

huggingface-pytorch-tpuv3x8

train steps
900
batch-size
16
samples/sec
Empty
train-time(sec)
237.898
x faster
Empty
python xla_spawn.py --num_cores 8 \ language-modeling/run_mlm.py \ --model_name_or_path roberta-base\ --dataset_name wikitext\ --dataset_config_name wikitext-2-raw-v1\ --do_train\ --do_eval\ --output_dir ../test-result\ --per_device_train_batch_size=2
Shell
[INFO|trainer.py:837] 2021-02-20 16:21:42,479 >> ***** Running training ***** [INFO|trainer.py:838] 2021-02-20 16:21:42,480 >> Num examples = 4798 [INFO|trainer.py:839] 2021-02-20 16:21:42,480 >> Num Epochs = 3 [INFO|trainer.py:840] 2021-02-20 16:21:42,480 >> Instantaneous batch size per device = 2 [INFO|trainer.py:841] 2021-02-20 16:21:42,480 >> Total train batch size (w. parallel, distributed & accumulation) = 16 [INFO|trainer.py:842] 2021-02-20 16:21:42,480 >> Gradient Accumulation steps = 1 [INFO|trainer.py:843] 2021-02-20 16:21:42,480 >> Total optimization steps = 900 02/20/2021 16:25:44 - INFO - run_mlm - ***** Train results ***** 02/20/2021 16:25:44 - INFO - run_mlm - epoch = 3.0 02/20/2021 16:25:44 - INFO - run_mlm - train_runtime = 237.898 02/20/2021 16:25:44 - INFO - run_mlm - train_samples_per_second = 3.783 02/20/2021 16:25:44 - INFO - run_mlm - *** Evaluate *** [INFO|trainer.py:1600] 2021-02-20 16:25:44,839 >> ***** Running Evaluation ***** [INFO|trainer.py:1601] 2021-02-20 16:25:44,839 >> Num examples = 496 [INFO|trainer.py:1602] 2021-02-20 16:25:44,841 >> Batch size = 8 100%|###########################################################################################################################| 8/8 [00:01<00:00, 6.61it/s] 02/20/2021 16:25:46 - INFO - run_mlm - ***** Eval results ***** 02/20/2021 16:25:46 - INFO - run_mlm - perplexity = 3.5175859654699786
Shell