fix falcon initialization during benchmarking (#362)

llm_bench/python/utils/ov_utils.py

@@ -141,7 +141,7 @@ def create_text_gen_model(model_path, device, **kwargs):
     if not model_path_existed:
         raise RuntimeError(f'==Failure ==: model path:{model_path} does not exist')
     else:
-        if model_type in ['mpt', 'falcon', 'replit', 'codegen2', 'chatglm']:
+        if model_type in ['replit', 'codegen2', 'chatglm', 'mpt']:
             start = time.perf_counter()
             ov_model = model_class.from_pretrained(
                 model_path,
@@ -151,6 +151,15 @@ def create_text_gen_model(model_path, device, **kwargs):
                 stateful=kwargs.get("stateful", None)
             )
             end = time.perf_counter()
+        elif model_type in ['falcon']:
+            ov_model = model_class.from_pretrained(
+                model_path,
+                device=device,
+                ov_config=ov_config,
+                stateful=kwargs.get("stateful", None),
+                trust_remote_code=False
+            )
+            end = time.perf_counter()
         else:
             start = time.perf_counter()
             config = AutoConfig.from_pretrained(model_path, trust_remote_code=True)