train_file=train_file, val_file=val_file, ) # In[ ]: device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") # In[ ]: T = TypeVar("T") TensorDict = Dict[str, Union[torch.Tensor, Dict[str, torch.Tensor]]] # In[ ]: processor = BertPreprocessor(config.model_type, config.max_seq_len) # In[ ]: DATA_ROOT = Path("../data") MODEL_SAVE_DIR = Path("../weights") # Read the model in here # In[ ]: from pytorch_pretrained_bert import BertConfig, BertForMaskedLM masked_lm = BertForMaskedLM.from_pretrained(config.model_type) masked_lm.eval() # In[ ]:
sys.path.append("../lib") # In[4]: from bert_utils import Config, BertPreprocessor # In[5]: config = Config( model_type="bert-base-uncased", max_seq_len=128, ) # In[6]: processor = BertPreprocessor(config.model_type, config.max_seq_len) # In[ ]: # In[7]: from pytorch_pretrained_bert import BertConfig, BertForMaskedLM model = BertForMaskedLM.from_pretrained(config.model_type) model.eval() # Important! Disable dropout # In[8]: def get_logits(sentence: str) -> np.ndarray: return model(processor.to_bert_model_input(sentence))[ 0, :, :].cpu().detach().numpy()
sys.path.append("../lib") # In[4]: from bert_utils import Config, BertPreprocessor # In[5]: config = Config( model_type="bert-base-uncased", max_seq_len=128, ) # In[6]: processor = BertPreprocessor(config.model_type, config.max_seq_len) # ### Prepare model # In[7]: from pytorch_pretrained_bert import BertConfig, BertForMaskedLM model = BertForMaskedLM.from_pretrained(config.model_type) model.eval() # In[8]: sequence_output, pooled_output = model.bert( processor.to_bert_model_input("hello world"), output_all_encoded_layers=False)