Update app.py

app.py

@@ -80,9 +80,9 @@ def generate(prompt, temperature=0.9, max_new_tokens=256, top_p=0.95, repetition
     if FIM_INDICATOR in prompt:
         fim_mode = True
         try:
-            prefix, suffix = prompt.split("<FILL-HERE>")
+            prefix, suffix = prompt.split(FIM_INDICATOR)
         except:
-            ValueError("Only one <FILL-HERE> allowed in prompt!")
+            ValueError(f"Only one {FIM_INDICATOR} allowed in prompt!")
         prompt = f"{FIM_PREFIX}{prefix}{FIM_SUFFIX}{suffix}{FIM_MIDDLE}"
 
     stream = client.generate_stream(prompt, **generate_kwargs)