Async requests

app.py

@@ -39,7 +39,7 @@ async def model_gen(
     
     chat_history = [] if no_context else chat_history
     
-    return model_response(
+    res = await model_response(
         content, 
         chat_history,
         S3_SESSION, 
@@ -49,6 +49,8 @@ async def model_gen(
         {"top_p": top_p, "temperature": temp, "max_tokens": max_tokens}
         )
 
+    return res
+
 async def model_regen(
         content,
         chat_history,
@@ -67,7 +69,7 @@ async def model_regen(
     chat_history = chat_history[-1] if no_context else chat_history
     content = copy(chat_history[-1][0])
     
-    return model_response(
+    res = await model_response(
         content, 
         chat_history[:-1],
         S3_SESSION, 
@@ -77,6 +79,8 @@ async def model_regen(
         {"top_p": top_p, "temperature": temp, "max_tokens": max_tokens}
         )
 
+    return res
+
 def model_manager(
         add_model,
         model_lib,

utils.py

@@ -7,7 +7,7 @@ from copy import copy
 def clear_chat():
     return '', []
 
-def model_response(
+async def model_response(
         content,
         chat_history,
         s3_session,