diff --git a/python/llm/dev/benchmark/all-in-one/run.py b/python/llm/dev/benchmark/all-in-one/run.py
index f7286079060..9420609d307 100644
--- a/python/llm/dev/benchmark/all-in-one/run.py
+++ b/python/llm/dev/benchmark/all-in-one/run.py
@@ -41,6 +41,8 @@
              'decapoda-research/llama-65b-hf','lmsys/vicuna-7b-v1.5',
              'lmsys/vicuna-13b-v1.3','lmsys/vicuna-33b-v1.3','project-baize/merged-baize-30b']
 
+LLAMA3_VISION_IDS = ['meta-llama/Llama-3.2-11B-Vision-Instruct']
+
 CHATGLM_IDS = ['THUDM/chatglm-6b', 'THUDM/chatglm2-6b', 'THUDM/chatglm3-6b']
 
 LLAVA_IDS = ['liuhaotian/llava-v1.5-7b']
@@ -770,6 +772,13 @@ def run_optimize_model_gpu(repo_id,
         model = optimize_model(model, low_bit=low_bit)
         tokenizer = LlamaTokenizer.from_pretrained(model_path, trust_remote_code=True)
         model = model.to('xpu')
+    elif repo_id in LLAMA3_VISION_IDS:
+        from transformers import MllamaForConditionalGeneration
+        model = MllamaForConditionalGeneration.from_pretrained(model_path, trust_remote_code=True,
+                                                     low_cpu_mem_usage=True).eval()
+        model = optimize_model(model, low_bit=low_bit, modules_to_not_convert=["multi_modal_projector"])
+        tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
+        model = model.to('xpu')
     else:
         model = AutoModelForCausalLM.from_pretrained(model_path, torch_dtype='auto', low_cpu_mem_usage=True,
                                                      trust_remote_code=True, use_cache=True).eval()