Training Overview¶

Note

The CrossEncoder training approach has not been updated in v3.0 when training Sentence Transformer models was improved. Improving training CrossEncoders is planned for a future major update.

The CrossEncoder class is a wrapper around Huggingface AutoModelForSequenceClassification, but with some methods to make training and predicting scores a little bit easier. The saved models are 100% compatible with Huggingface and can also be loaded with their classes.

First, you need some sentence pair data. You can either have a continuous score, like:

from sentence_transformers import InputExample

train_samples = [
    InputExample(texts=["sentence1", "sentence2"], label=0.3),
    InputExample(texts=["Another", "pair"], label=0.8),
]

Or you have distinct classes as in the training_nli.py example:

from sentence_transformers import InputExample

label2int = {"contradiction": 0, "entailment": 1, "neutral": 2}
train_samples = [
    InputExample(texts=["sentence1", "sentence2"], label=label2int["neutral"]),
    InputExample(texts=["Another", "pair"], label=label2int["entailment"]),
]

Then, you define the base model and the number of labels. You can take any Hugging Face pre-trained model that is compatible with AutoModel:

model = CrossEncoder('distilroberta-base', num_labels=1)

For binary tasks and tasks with continuous scores (like STS), we set num_labels=1. For classification tasks, we set it to the number of labels we have.

We start the training by calling CrossEncoder.fit:

model.fit(
    train_dataloader=train_dataloader,
    evaluator=evaluator,
    epochs=num_epochs,
    warmup_steps=warmup_steps,
    output_path=model_save_path,
)