diff --git a/generate.py b/generate.py
index 7183213e7..c1c02b448 100644
--- a/generate.py
+++ b/generate.py
@@ -46,7 +46,7 @@ class GeneratorArgs:
     def from_args(cls, args):  # -> GeneratorArgs:
         return cls(
             prompt=args.prompt,
-            chat_mode=args.subcommand == "chat",
+            chat_mode=hasattr(args, "subcommand") and args.subcommand == "chat",
             gui_mode=args.gui,
             num_samples=args.num_samples,
             max_new_tokens=args.max_new_tokens,
@@ -404,12 +404,11 @@ def _main(
         if i >= 0 and generator_args.chat_mode:
             prompt = input("What is your prompt? ")
 
-            # DEBUG DO NOT COMMIT
-            B_INST = ""
-            E_INST = ""
+            if generator_args.chat_mode:
+                # TODO Where should B_INST, E_INST come from? Model args?
+                #prompt = f"{B_INST} {prompt.strip()} {E_INST}" 
+                prompt = prompt.strip()
 
-            if chat_mode:
-                prompt = f"{B_INST} {prompt.strip()} {E_INST}"
             encoded = encode_tokens(
                 tokenizer, prompt, bos=True, device=builder_args.device
             )
@@ -492,8 +491,6 @@ def callback(x):
 
 
 def main(args):
-    is_chat = args.subcommand == "chat"
-
     # If a named model was provided and not downloaded, download it.
     if args.model and not is_model_downloaded(args.model, args.model_directory):
         download_and_convert(args.model, args.model_directory, args.hf_token)