[VLM] Clean up models (#16873)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

examples/offline_inference/mistral-small.py

@@ -62,6 +62,7 @@ def run_simple_demo(args: argparse.Namespace):
         tokenizer_mode="mistral" if args.format == "mistral" else "auto",
         config_format="mistral" if args.format == "mistral" else "auto",
         load_format="mistral" if args.format == "mistral" else "auto",
+        limit_mm_per_prompt={"image": 1},
         max_model_len=4096,
         max_num_seqs=2,
         tensor_parallel_size=2,

examples/offline_inference/vision_language.py

@@ -957,7 +957,7 @@ def run_qwen2_5_omni(questions: list[str], modality: str):
             "max_pixels": 1280 * 28 * 28,
             "fps": [1],
         },
-        disable_mm_preprocessor_cache=args.disable_mm_preprocessor_cache,
+        limit_mm_per_prompt={"image": 1},
     )
 
     if modality == "image":