SentenceTransformers Python library, offering a range of transformers for computing embeddings:
sentence_transformers

Computing embeddings

Lists or strings as input

You can either pass a list or a string, both generating embeddings - either for the each string in the list, or the single string. Thus, both cases are generating almost the same embedding for "hello world".

embeddings = model.encode(["hello world"])
embeddingsSingle = model.encode("hello world")

# embeddings[0] == embeddingsSingle

The reason for why it is only almost is the randomness in initialisation. The appraoch is not fully deterministic, thus, the embeddings vary a tiny bit - but not so much I would actually matter.

You might want to prefer providing a list, as it is faster to provide a batch.

Working with CUDA

SentenceTransformers can be accelerated using CUDA. The prequisite is installing PyTorch with CUDA:

sentences = ["This is an example sentence", "Each sentence is converted", "Another sentence"]


embeddings = model.encode(sentences, device='cuda')
print(embeddings)