Writing logs to /p/qdata/jm8wx/research/text_attacks/textattack/outputs/training/distilbert-base-uncased-glue:wnli-2020-06-29-11:31/log.txt.
Loading nlp dataset glue, subset wnli, split train.
Loading nlp dataset glue, subset wnli, split validation.
Loaded dataset. Found: 2 labels: ([0, 1])
Loading transformers AutoModelForSequenceClassification: distilbert-base-uncased
Tokenizing training data. (len: 635)
Tokenizing eval data (len: 71)
Loaded data and tokenized in 6.015664100646973s
Training model across 4 GPUs
***** Running training *****
	Num examples = 635
	Batch size = 128
	Max sequence length = 256
	Num steps = 20
	Num epochs = 5
	Learning rate = 2e-05
Eval accuracy: 56.33802816901409%
Best acc found. Saved model to /p/qdata/jm8wx/research/text_attacks/textattack/outputs/training/distilbert-base-uncased-glue:wnli-2020-06-29-11:31/.
Eval accuracy: 40.845070422535215%
Eval accuracy: 28.169014084507044%
Eval accuracy: 26.76056338028169%
Eval accuracy: 26.76056338028169%
Saved tokenizer <textattack.models.tokenizers.auto_tokenizer.AutoTokenizer object at 0x7f50d57d5d90> to /p/qdata/jm8wx/research/text_attacks/textattack/outputs/training/distilbert-base-uncased-glue:wnli-2020-06-29-11:31/.
Wrote README to /p/qdata/jm8wx/research/text_attacks/textattack/outputs/training/distilbert-base-uncased-glue:wnli-2020-06-29-11:31/README.md.
Wrote training args to /p/qdata/jm8wx/research/text_attacks/textattack/outputs/training/distilbert-base-uncased-glue:wnli-2020-06-29-11:31/train_args.json.