diff --git a/examples/unsupervised_learning/MLM/train_mlm.py b/examples/unsupervised_learning/MLM/train_mlm.py index 20287bdf0..cebc72013 100644 --- a/examples/unsupervised_learning/MLM/train_mlm.py +++ b/examples/unsupervised_learning/MLM/train_mlm.py @@ -27,7 +27,7 @@ use_fp16 = False #Set to True, if your GPU supports FP16 operations max_length = 100 #Max length for a text input do_whole_word_mask = True #If set to true, whole words are masked -mlm_prob = 15 #Probability that a word is replaced by a [MASK] token +mlm_prob = 0.15 #Probability that a word is replaced by a [MASK] token # Load the model model = AutoModelForMaskedLM.from_pretrained(model_name) @@ -121,4 +121,4 @@ def __len__(self): print("Save model to:", output_dir) model.save_pretrained(output_dir) -print("Training done") \ No newline at end of file +print("Training done")