Encode Sentences

After loading the model as model, you can encode sentences by

model.encode(sentences, device=None, return_numpy=False, normalize_to_unit=True, keepdim=False, batch_size=64, max_length=128)

Inputs

• sentences: a string or a list of strings.
• device: cuda or cpu.
• return_numpy: whether to return numpy arrays (True) or return PyTorch tensors (False by default).
• normalize_to_unit: whether to normalize the output embeddings as unit vectors.
• keepdim: if the input is a single sentence, whether to keep the batch-size dimension of the output embedding.
• batch_size: if the input is a list of sentences, the batch size for encoding. Usually larger batch sizes lead to higher efficiency (as long as it can fit into your computing devices).
• max_length: truncate the sentences if they exceed the maximum length.

Outputs

• A numpy array or a PyTorch tensor with size (n, dim), where n is # sentences.