all-clip 1.3.0

Load any clip model with a standardized interface

all_clip

Load any clip model with a standardized interface

Install

pip install all_clip

Python examples

from all_clip import load_clip
import torch
from PIL import Image
import pathlib


model, preprocess, tokenizer = load_clip("open_clip:ViT-B-32/laion2b_s34b_b79k", device="cpu", use_jit=False)


image = preprocess(Image.open(str(pathlib.Path(__file__).parent.resolve()) + "/CLIP.png")).unsqueeze(0)
text = tokenizer(["a diagram", "a dog", "a cat"])

with torch.no_grad(), torch.cuda.amp.autocast():
    image_features = model.encode_image(image)
    text_features = model.encode_text(text)
    image_features /= image_features.norm(dim=-1, keepdim=True)
    text_features /= text_features.norm(dim=-1, keepdim=True)

    text_probs = (100.0 * image_features @ text_features.T).softmax(dim=-1)

print("Label probs:", text_probs)  # prints: [[1., 0., 0.]]

Checkout these examples to call this as a lib:

• example.py

API

This module exposes a single function load_clip:

• clip_model CLIP model to load (default ViT-B/32). See below supported models section.
• use_jit uses jit for the clip model (default True)
• warmup_batch_size warmup batch size (default 1)
• clip_cache_path cache path for clip (default None)
• device device (default None)

Related projects

• clip-retrieval to use clip for inference, and retrieval
• open_clip to train clip models
• CLIP_benchmark to evaluate clip models

Supported models

OpenAI

Specify the model as "ViT-B-32"

Openclip

"open_clip:ViT-B-32/laion2b_s34b_b79k" to use the open_clip

HF CLIP

"hf_clip:patrickjohncyh/fashion-clip" to use the hugging face

Deepsparse backend

DeepSparse is an inference runtime for fast sparse model inference on CPUs. There is a backend available within clip-retrieval by installing it with pip install deepsparse-nightly[clip], and specifying a clip_model with a prepended "nm:", such as "nm:neuralmagic/CLIP-ViT-B-32-256x256-DataComp-s34B-b86K-quant-ds" or "nm:mgoin/CLIP-ViT-B-32-laion2b_s34b_b79k-ds".

Japanese clip

japanese-clip provides some models for japanese. For example one is ja_clip:rinna/japanese-clip-vit-b-16

How to add a model type

Please follow these steps:

1. Add a file to load model in all_clip/
2. Define a loading function, that returns a tuple (model, transform, tokenizer). Please see all_clip/open_clip.py as an example.
3. Add the function into TYPE2FUNC in all_clip/main.py
4. Add the model type in test_main.py and ci.yml

Remarks:

• The new tokenizer/model must enable to do the following things as https://github.com/openai/CLIP#usage
  • tokenizer(texts).to(device) ... texts is a list of string
  • model.encode_text(tokenized_texts) ... tokenized_texts is a output from tokenizer(texts).to(device)
  • model.encode_image(images) ... images is a image tensor by the transform

For development

Setup a virtualenv:

python3 -m venv .env
source .env/bin/activate
pip install -e .

to run tests:

pip install -r requirements-test.txt

then

make lint
make test

You can use make black to reformat the code

python -m pytest -x -s -v tests -k "ja_clip" to run a specific test