Adapted to the final model; with advanced delimiter and cutoff.

app.py

@@ -79,13 +79,13 @@ def evaluate_stream(msg, history, temperature, top_p):
 
     # concatenate context
     for h in history[:-1]:
-        context += h[0] + "\n" + h[1] + "\n"
+        context += h[0] + "||" + h[1] + "||"
 
     context += history[-1][0]
     context = context.replace(r'<br>', '')
 
     # TODO: Avoid the tokens are too long.
-    CUTOFF = 208
+    CUTOFF = 224
     while len(tokenizer.encode(context)) > CUTOFF:
         # save 15 token size for the answer
         context = context[15:]