Training your model

We are ready to train our first model with classy! Since we are doing Named Entity Recognition, and we want to go with a fast model (i.e., distilbert), let's name this experiment fast-ner:

classy train token data/ner-data/ --exp-name fast-ner --profile distilbertTraining completed

tip

Your model and experiment data will be saved in experiments/<exp-name>/YYYY-MM-DD/HH-mm-ss/.

classy automatically saves best and last checkpoints, as well as the training configuration.

info

token in the above command tells classy to train a Token Classification model. This is the only thing, besides organizing data, that classy expects you to do. We'll go back to this later on.