Training.md

Training

Basic command:

combo --mode train \
      --training_data_path your_training_path \
      --validation_data_path your_validation_path \
      --serialization_dir directory_to_store_model_output_in

Options:

combo --helpfull

Examples

For clarity, the training and validation data paths, as well as serialization directory path, are omitted.

Train on multiple accelerators (default: train on all available ones)

combo --mode train --n_cuda_devices 8

Use pretrained transformer embeddings:

combo --mode train --pretrained_transformer_name your_chosen_pretrained_transformer

Train only a dependency parser:

combo --mode train --targets head,deprel

Use additional features (e.g. part-of-speech tags) for training a dependency parser (token and char are default features)

combo --mode train --targets head,deprel --features token,char,upostag

Custom configuration

Use a custom configuration:

combo --config_path configuration.json

Discard any flags (including default flag values - without this flag, the default values will override configuration!)

combo --config_path configuration.json --use_pure_config

Finetuning

Finetune a pre-trained model:

combo --mode train --finetune \
       --finetuning_training_data_path your_training_path \
       --finetuning_validation_data_path your_validation_path \
       --model_path pretrained_model_path