Update answerer.py

answerer.py

@@ -1,4 +1,4 @@
-from typing import Dict, Generator, List
+from typing import Dict, AsyncGenerator, List
 import os, gc
 from huggingface_hub import hf_hub_download
 from rwkv.model import RWKV
@@ -13,7 +13,7 @@ class Answerer:
     self.__pipeline = PIPELINE(self.__model, vocab)
     self.ctx_limit = ctx_limit
 
-  def __call__(
+  async def __call__(
     self,
     input: str,
     max_output_length_tk: int,
@@ -21,7 +21,7 @@ class Answerer:
     repetitiveness = .3,
     diversity = 0,
     _count_penalty = 1,
-  ) -> Generator[str, None, None]:
+  ) -> AsyncGenerator[str, None]:
     args = PIPELINE_ARGS(
       temperature=chaos,
       top_p=repetitiveness,