Huge fan of Typesense here 🤟 hoping to get some advice on implementing semantic search. Typesense Cloud is powering our ecommerce site search. It's a small dataset of about 4.5k documents for a total of 50MB of JSON. The smallest cluster size (0.5GB, 2 vCPU with HA) runs this just fine, and I can re-import the whole dataset in seconds. I spun up a larger cluster (4GB, 2 vCPU 4hr burst, no HA) to experiment with enabling semantic search. I don't know much about embedding models but chose the

ts/gte-small

model since it seems popular. Unfortunately importing the documents is slowwww - as best I can tell from the analytics something like 10-20s per document! For reference I think each document averages about 200 tokens. I could add more CPUs or even GPUs to the cluster but that could take a $50/month service to $1,000+/month. I'm after some guidance or real-world experience with implementing text embeddings. Am I doing something wrong? Should I be using a different model? Or is that just the computation that's required and I have to stump up for faster infrastructure? Also, while the cluster was importing my data it became totally unresponsive. Obviously a problem if it's going to take hours. Is that just because I didn't have HA enabled? Would a HA cluster still be able to serve search queries while ingesting new documents? Thanks again for your hard work on Typesense, it's a brilliant product and the progress of development blows me away.