Update app.py

app.py

@@ -7,7 +7,6 @@ from typing import Iterator
 from transformers import (
     AutoModelForCausalLM,
     BitsAndBytesConfig,
-    GemmaTokenizerFast,
     TextIteratorStreamer,
 )
 
@@ -37,7 +36,7 @@ if True:
 
 MAX_MAX_NEW_TOKENS = 2048
 DEFAULT_MAX_NEW_TOKENS = 1024
-MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
+MAX_INPUT_TOKEN_LENGTH = 4096 #int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
 
 
 model.config.sliding_window = 4096