Basic inference endpoints

Models in the embedding module can be selected with the session.embedding.get_model(model_id) function and expose up to four basic inference methods:

• model.embed([sequences], reduction='mean')

• model.logits([sequences])

• model.attn([sequences])

• model.fit_svd([sequences])

Sequences are expected to be amino acid sequences using the standard single character amino acid alphabet, plus X to indicate masked/unknown positions. More information about sequences can be retrieved from the model metadata (model.metadata) which also includes sequence length constraints and information about tokens.

Available models can be listed with session.embedding.list_models().

Next steps

For more information, visit the Embeddings API reference.