disable md5 check assert for CB (#933)

llm_bench/python/benchmark.py

@@ -308,13 +308,14 @@ def run_text_generation_genai(input_text, num, model, tokenizer, args, iter_data
             log.warning(f"[{num}] Prompt[{prompt_index}]'s md5 {result_md5_list} "
                         f"is different from md5 of the {num - 1} iteration {prev_md5}")
             llm_bench_utils.metrics_print.print_generated(num, warm_up=(num == 0), generated=generated_text[0])
-            if num == 1:
-                # if the device is CPU, throw exception
-                if args['devices'].lower().startswith('cpu') is True:
+            if not args.get("use_cb", False):
+                if num == 1:
+                    # if the device is CPU, throw exception
+                    if args['devices'].lower().startswith('cpu') is True:
+                        assert (result_md5_list == prev_md5)
+                else:
+                    # throw exception
                     assert (result_md5_list == prev_md5)
-            else:
-                # throw exception
-                assert (result_md5_list == prev_md5)
     else:
         llm_bench_utils.metrics_print.print_generated(num, warm_up=(num == 0), generated=generated_text[0])
 
@@ -814,7 +815,7 @@ def get_argprser():
     llm_bench_utils.model_utils.add_stateful_model_arguments(parser)
     parser.add_argument("--genai", action="store_true", help="Use OpenVINO GenAI optimized pipelines for benchmarking")
     parser.add_argument("--use_cb", action="store_true", help="Use Continuous Batching inference mode")
-    parser.add_argument("--cb_config", required=False, default=None, help="Path to file with Continuous Batching Scheduler settings")
+    parser.add_argument("--cb_config", required=False, default=None, help="Path to file with Continuous Batching Scheduler settings or dict")
     parser.add_argument(
         '--end_token_stopping',
         action='store_true',