()
| 11 | |
| 12 | |
| 13 | def test_inference(): |
| 14 | for model_name in ["ViT-B-32", "ViT-B-32-quickgelu", "ViT-B-16", "ViT-L-14"]: |
| 15 | for pretrained in ["metaclip400m", "metaclip2_5b"]: |
| 16 | model, _, preprocess = open_clip.create_model_and_transforms( |
| 17 | model_name, pretrained=pretrained |
| 18 | ) |
| 19 | |
| 20 | current_dir = os.path.dirname(os.path.realpath(__file__)) |
| 21 | |
| 22 | image = preprocess(Image.open(current_dir + "/../docs/CLIP.png")).unsqueeze( |
| 23 | 0 |
| 24 | ) |
| 25 | text = tokenizer.tokenize(["a diagram", "a dog", "a cat"]) |
| 26 | |
| 27 | with torch.no_grad(): |
| 28 | image_features = model.encode_image(image) |
| 29 | text_features = model.encode_text(text) |
| 30 | |
| 31 | text_probs = (100.0 * image_features @ text_features.T).softmax(dim=-1) |
| 32 | |
| 33 | assert text_probs.cpu().numpy()[0].tolist() == [1.0, 0.0, 0.0] |
nothing calls this directly
no test coverage detected