added truncation to clip tokenizer. TODO: only use the question, not the context.

clip_for_ppts.py

@@ -74,7 +74,7 @@ class ClipImage:
 
     elif self.mode == 'text':
       # Preprocess the text
-      input_arr = torch.cat([clip.tokenize(f"{input_text_or_img}")]).to(self.device)
+      input_arr = torch.cat([clip.tokenize(f"{input_text_or_img}", truncate=True)]).to(self.device)
 
     # Encode the image or text
     with torch.no_grad():