clip-as-service

Why we included this project

CLIP-as-service wraps OpenAI's CLIP models behind a network API, so you can embed images and text into the same vector space without writing the serving layer yourself. It runs PyTorch, ONNX Runtime, or TensorRT backends and can spread several CLIP models across GPUs with load balancing, which is what you want once embedding demand outgrows a single in-process call. Client and server talk over gRPC, HTTP, or WebSocket with streaming, so the service slots into an existing neural-search pipeline rather than staying stuck in a notebook. The client ships as a separate lightweight package, a practical detail when you want embedding logic on a laptop while the heavier inference runs on a GPU box.