For text based open-source models I would start looking in the HuggingFace Models hub (https://huggingface.co/models) and filter on "Feature extraction". Here is an example (https://huggingface.co/sentence-transformers/all-MiniLM-L6-v...). What do you plan to do with the embeddings once created?

For hosting solutions that depends on the traffic you're expecting. Is this just for testing or you have a level of traffic in mind?