Skip to content

pranavajitnair/CoLA

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

CoLA

A Trandformer style and a BiLSTM with ELMo to evaluate on Corpus of Linguistic Acceptibility(CoLA dataset)

Dataset

The dataset can be downloaded here

Training

To train and validate the model run

python train.py

Optional Arguments

--batch_size              Batch size to train the model           
--lr                      Learning rate for the model
--hidden_size             Hidden size, either for the BiLSTM for the Transformer
--embed_size              Embedding size, either for the BiLSTM for the Transformer
--n_heads                 Number of heads for Multi Head Attention for the Traansformer
--n_layer                 Number of layers in the Transformer
--per_layer               Iterations per layer of the Transformer
--inter_size              Intermediate size for the Transformer
--train_path              Path to the training file
--dev_path                Path to the development file
--epochs                  Number of epochs for training
--min_word_count          Minimum word count to include a word in the vocabulary
--eval_every              Evaluate after how many epochs
--dropout_prob_classifier Dropout rate for the Classifier in the Transformer
--dropout_prob_attn       Dropout rate for self attention 
--dropout_prob_hidden     Dropout rate for the self feed forward network of the Transformer
--num_rep                 Number of ELMo representations
--elmo_drop               Dropout rate for ELMo
--use_elmo                Whether to use ELMo or not
--BiLSTM                  Whether to use BiLSTM model or not
--transformer             Whether to use Transformer model or not

About

A transformer and BiLSTM architecture with ELMo to evaluate performance on CoLA (Corpus of Linguistic Acceptability) dataset. PyTorch used

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages