I'm currently working on implementing Embeddings for typesense search. To speed things up I am thinking about generating the embeddings before hand and then syncing them over to my typesense server (so there is no need to generate the embeddings). Now I am wondering is this a good idea? I'm just thinking that if the generation is handled differently by typesense on search this could make the search inaccurate

Should be ok to do this and I have heard of users doing this already. If you have auto-embedding enabled and you provide pre-generated embeddings, then Typesense will store those in the index. Then auto-embedding will only run at query time. You just want to make sure you use the exact same model as Typesense to generate the embeddings.