diff --git a/python/llm/dev/benchmark/all-in-one/run.py b/python/llm/dev/benchmark/all-in-one/run.py
index 58d60b14..35591f75 100644
--- a/python/llm/dev/benchmark/all-in-one/run.py
+++ b/python/llm/dev/benchmark/all-in-one/run.py
@@ -375,8 +375,11 @@ def run_transformer_int4_gpu(repo_id,
         model = model.to('xpu')
     else:
         if 'starcoder' in repo_id:
+            # Load starcoder-15.5b model in bf16 format to avoid CPU OOM. 
             model = AutoModelForCausalLM.from_pretrained(model_path, optimize_model=True, load_in_low_bit=low_bit,
                                                         trust_remote_code=True, use_cache=True, torch_dtype=torch.bfloat16).eval()
+            # Convert the low-bit model back to fp32 for performance considerations.
+            model = model.float()
         else:
             model = AutoModelForCausalLM.from_pretrained(model_path, optimize_model=True, load_in_low_bit=low_bit,
                                                         trust_remote_code=True, use_cache=True).eval()