Update __main__.py

__main__.py

@@ -126,13 +126,13 @@ def answer_question(
     tokenizer.eos_token = "<|eot_id|>"
 
     try:
-        inp = input("user: ")
+        q = input("user: ")
     except EOFError:
-        inp = ""
-    if not inp:
-        print("exit...")
+        q = ""
+    if not q:
+        print("no input detected. exiting.")
 
-    question = "<image>" + inp
+    question = "<image>" + q
 
     prompt = f"<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\n{question}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
 
@@ -179,6 +179,7 @@ def answer_question(
             "max_new_tokens": 2000,
             "use_cache": True,
             "streamer": streamer,
+            "pad_token_id": tokenizer.eos_token_id
         }
 
         while True:
@@ -188,11 +189,11 @@ def answer_question(
 
             generated_text = tokenizer.decode(generated_ids, skip_special_tokens=False)
             try:
-                inp = input("user: ")
+                q = input("user: ")
             except EOFError:
-                inp = ""
-            if not inp:
-                print("exit...")
+                q = ""
+            if not q:
+                print("no input detected. exiting.")
 
             new_text = (
                 generated_text