use pretrained clip

.gitignore

@@ -0,0 +1,2 @@
+.venv/
+__pycache__/

app.py

@@ -6,8 +6,9 @@ from torchvision import transforms
 from open_clip import create_model, get_tokenizer
 from templates import openai_imagenet_template
 
-model_str = "ViT-B-16"
-pretrained = "/fs/ess/PAS2136/foundation_model/model/10m/2023_09_22-21_14_04-model_ViT-B-16-lr_0.0001-b_4096-j_8-p_amp/checkpoints/epoch_99.pt"
+model_str = "hf-hub:imageomics/bioclip"
+tokenizer_str = "ViT-B-16"
+
 
 preprocess_img = transforms.Compose(
     [
@@ -50,13 +51,13 @@ def predict(img, cls_str: str) -> dict[str, float]:
 
 if __name__ == "__main__":
     print("Starting.")
-    model = create_model(model_str, pretrained, output_dict=True)
+    model = create_model(model_str, output_dict=True, require_pretrained=True)
     print("Created model.")
 
     model = torch.compile(model)
     print("Compiled model.")
 
-    tokenizer = get_tokenizer(model_str)
+    tokenizer = get_tokenizer(tokenizer_str)
 
     demo = gr.Interface(
         fn=predict,