how to excract batch embedding?

[tomlongcool]

this is a great project! i saw the docs say this project can extract the sequence embedding by the db database file(sqlite),this is wonderful!!! but i how to generate the db file? if i have a fasta file ? I really appreciate your help！！！

[lhallee]

Thanks for the kind words and great question!

We've just added direct FASTA support to embed_dataset across all models. You can now pass a fasta_path argument instead of (or alongside) a list of sequences — no manual parsing needed. Multi-line FASTA sequences are handled automatically.

Generating a .db file from a FASTA file:

from transformers import AutoModel

model = AutoModel.from_pretrained("Synthyra/ESM2-150M", trust_remote_code=True).cuda()

model.embed_dataset(
    fasta_path='my_proteins.fasta',
    batch_size=64,
    pooling_types=['mean'],
    sql=True,
    sql_db_path='my_proteins.db',
)

That's all — when it finishes, my_proteins.db is ready to use. You can then load embeddings back out with:

embeddings = model.load_embeddings_from_db('my_proteins.db')
seq_vector = embeddings["MKTLLLTLVVVTIVCLDLGYT"]  # keyed by sequence string

You can also mix a FASTA file with an explicit list of sequences if needed — the two sources are combined before embedding:

model.embed_dataset(
    sequences=["MKTIIALSYIFCLVFA"],
    fasta_path='my_proteins.fasta',
    sql=True,
    sql_db_path='my_proteins.db',
)

The updated README has a full example under "Embedding from a FASTA File". Let us know if you run into any issues!