clip-models

Welcome to CLIP-as-service!#

GitHub: clip-as-service

Docs: clip-as-service

CLIP-as-service is a low-latency high-scalability service for embedding images and text. It can be easily integrated as a microservice into neural search solutions.

⚡ Fast: Serve CLIP models with TensorRT, ONNX runtime and PyTorch w/o JIT with 800QPS[*]. Non-blocking duplex streaming on requests and responses, designed for large data and long-running tasks.

🫐 Elastic: Horizontally scale up and down multiple CLIP models on single GPU, with automatic load balancing.

🐥 Easy-to-use: No learning curve, minimalist design on client and server. Intuitive and consistent API for image and sentence embedding.

👒 Modern: Async client support. Easily switch between gRPC, HTTP, WebSocket protocols with TLS and compression.

🍱 Integration: Smooth integration with neural search ecosystem including Jina and DocArray. Build cross-modal and multi-modal solutions in no time.

[*] with default config (single replica, PyTorch no JIT) on GeForce RTX 3090.

Try it!#

Install#

PyPI is the latest version.

Make sure you are using Python 3.7+. You can install the client and server independently. It is not required to install both: e.g. you can install clip_server on a GPU machine and clip_client on a local laptop.

Client

pip install clip-client

Server (PyTorch)

pip install clip-server

Server (ONNX)

pip install "clip_server[onnx]"

Server (TensorRT)

pip install nvidia-pyindex 
pip install "clip_server[tensorrt]"

Server on Google Colab

Quick check#

After installing, you can run the following commands for a quick connectivity check.

Start the server#

Start PyTorch Server

python -m clip_server

Start ONNX Server

python -m clip_server onnx-flow.yml

Start TensorRT Server

python -m clip_server tensorrt-flow.yml

At the first time starting the server, it will download the default pretrained model, which may take a while depending on your network speed. Then you will get the address information similar to the following:

╭────────────── 🔗 Endpoint ───────────────╮
│  🔗     Protocol                   GRPC  │
│  🏠        Local          0.0.0.0:51000  │
│  🔒      Private    192.168.31.62:51000  │
|  🌍       Public   87.105.159.191:51000  |
╰──────────────────────────────────────────╯  

This means the server is ready to serve. Note down the three addresses shown above, you will need them later.

Connect from client#

Depending on the location of the client and server. You may use different IP addresses:
- Client and server are on the same machine: use local address, e.g. `0.0.0.0`
- Client and server are connected to the same router: use private network address, e.g. `192.168.3.62`
- Server is in public network: use public network address, e.g. `87.105.159.191`

Run the following Python script:

from clip_client import Client

c = Client('grpc://0.0.0.0:51000')
c.profile()

will give you:

 Roundtrip  16ms  100%
├──  Client-server network  8ms  49%
└──  Server  8ms  51%
    ├──  Gateway-CLIP network  2ms  25%
    └──  CLIP model  6ms  75%
{'Roundtrip': 15.684750003856607, 'Client-server network': 7.684750003856607, 'Server': 8, 'Gateway-CLIP network': 2, 'CLIP model': 6}

It means the client and the server are now connected. Well done!