Fix: save CLIPProcessor in Docker build

Dockerfile

@@ -21,5 +21,5 @@ RUN mkdir -p /models/clip && \
     python3 -c "from transformers import CLIPModel; CLIPModel.from_pretrained('openai/clip-vit-base-patch32').save_pretrained('/models/clip')"
 
 RUN python3 -c "from transformers import AutoTokenizer; AutoTokenizer.from_pretrained('bert-base-uncased').save_pretrained('/models/bert-tokenizer')"
-
+RUN python3 -c "from transformers import CLIPProcessor; CLIPProcessor.from_pretrained('openai/clip-vit-base-patch32').save_pretrained('/models/clip')"
 CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000"]

vit_captioning/generate.py

@@ -33,15 +33,11 @@ class CaptionGenerator:
         if model_type == "ViTEncoder":
             self.encoder = ViTEncoder().to(self.device)
             self.encoder_dim = 768
-            #self.processor = ViTImageProcessor.from_pretrained("google/vit-base-patch16-224-in21k")
-            #HF needs all model downloads to a special read-write cache dir
-            self.processor = ViTImageProcessor.from_pretrained("google/vit-base-patch16-224-in21k", cache_dir="/tmp")
+            self.processor = ViTImageProcessor.from_pretrained("google/vit-base-patch16-224-in21k")
         elif model_type == "CLIPEncoder":
             self.encoder = CLIPEncoder().to(self.device)
             self.encoder_dim = 512
             #self.processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
-            #HF needs all model downloads to a special read-write cache dir
-            #self.processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32", cache_dir="/tmp")
             self.processor = CLIPProcessor.from_pretrained("/models/clip")
         else:
             raise ValueError("Unknown model type")