ship smaller model

app.py

@@ -146,7 +146,7 @@ def init_models():
         "question-answering", model='sultan/BioM-ELECTRA-Large-SQuAD2-BioASQ8B',
         device=device
     )
-    reranker = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-6-v2', device=device)
+    reranker = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-2-v2', device=device)
     # queryexp_tokenizer = AutoTokenizer.from_pretrained("doc2query/all-with_prefix-t5-base-v1")
     # queryexp_model = AutoModelWithLMHead.from_pretrained("doc2query/all-with_prefix-t5-base-v1")
     return question_answerer, reranker, stop, device # uqeryexp_model, queryexp_tokenizer