Update llm.py

llm.py

@@ -5,7 +5,7 @@ import torch
 import asyncio
 from functools import partial
 import time
-from transformers import AutoProcessor, AutoModelForImageTextToText #, BitsAndBytesConfig
+from transformers import AutoProcessor, AutoModelForVision2Seq #, BitsAndBytesConfig
 
 # Quantization config
 # quant_config = BitsAndBytesConfig(load_in_8bit=True)
@@ -14,7 +14,7 @@ model_name="Qwen/Qwen3.5-9B-Base"
 processor = AutoProcessor.from_pretrained(model_name)
 
 # Load model (auto device mapping)
-model = AutoModelForImageTextToText.from_pretrained(
+model = AutoModelForVision2Seq.from_pretrained(
     model_name,
     # quantization_config=quant_config,
     device_map="auto",