Create README.md

README.md

@@ -0,0 +1,45 @@
+---
+license: apache-2.0
+datasets:
+- UCSC-VLAA/Recap-DataComp-1B
+---
+# Model Card for ViT-H-14-CLIPS-224-Recap-DataComp-1B
+
+## Model Details
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** https://github.com/UCSC-VLAA/CLIPS
+- **Paper:** https://arxiv.org/abs/2411.16828
+- **Project Page:** https://ucsc-vlaa.github.io/CLIPS/
+
+## Model Usage
+### With OpenCLIP
+#### Note: Due to differences in the default epsilon values for LayerNorm initialization between JAX and PyTorch, we made some modifications in open_clip/transformer.py to align the model's behavior. Refer to https://github.com/UCSC-VLAA/CLIPS for more details.
+```
+import torch
+import torch.nn.functional as F
+from urllib.request import urlopen
+from PIL import Image
+from open_clip import create_model_from_pretrained, get_tokenizer
+
+model, preprocess = create_model_from_pretrained('hf-hub:UCSC-VLAA/ViT-H-14-CLIPS-224-Recap-DataComp-1B')
+tokenizer = get_tokenizer('hf-hub:UCSC-VLAA/ViT-H-14-CLIPS-224-Recap-DataComp-1B')
+
+image = Image.open(urlopen(
+    'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
+))
+image = preprocess(image).unsqueeze(0)
+
+text = tokenizer(["a diagram", "a dog", "a cat", "a beignet"], context_length=model.context_length)
+
+with torch.no_grad(), torch.cuda.amp.autocast():
+    image_features = model.encode_image(image)
+    text_features = model.encode_text(text)
+    image_features = F.normalize(image_features, dim=-1)
+    text_features = F.normalize(text_features, dim=-1)
+
+    text_probs = (100.0 * image_features @ text_features.T).softmax(dim=-1)
+
+print("Label probs:", text_probs)  # prints: [[0., 0., 0., 1.0]]
+```