Update answerer.py

answerer.py

@@ -1,7 +1,7 @@
-from typing import Dict, List
 import os, gc
 from rwkv.model import RWKV
 from rwkv.utils import PIPELINE, PIPELINE_ARGS
+from asyncio import sleep
 
 class Answerer:
   def __init__(self, model: str, vocab: str, strategy: str, ctx_limit: int):
@@ -12,7 +12,7 @@ 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,
@@ -34,8 +34,8 @@ class Answerer:
 
     result: str = ""
     
-    occurrences: Dict[int, int] = {}
-    tokens: List[int] = []
+    occurrences: dict[int, int] = {}
+    tokens: list[int] = []
     current_token = None
     state = None
     for _ in range(max_output_length_tk):
@@ -68,6 +68,7 @@ class Answerer:
         tokens.clear()
         result += tmp
         yield result
+        await sleep(.2)
 
     tokens.clear()
     occurrences.clear()