Upload T5/generate-dict-embeddingsT5.py with huggingface_hub

T5/generate-dict-embeddingsT5.py

@@ -33,6 +33,9 @@ def embed_from_text(text):
         print("ERROR: expected single token per word")
         print(text)
         exit(1)
+        # We can only accept single-token words, because we want our output 
+        # to be a single embedding per word, and we dont have an official
+        # way to merge multiple T5 embeddings into one, like CLIP does
 
     with torch.no_grad():
         outputs = tmodel(tokens.input_ids)