configurable delay (#1963)

CVS-164856

Author: eaidova

tools/llm_bench/benchmark.py

@@ -85,6 +85,14 @@ def get_argprser():
         help='if the value is 1, output the maximum memory consumption in warm-up iterations. If the value is 2,'
         ' output the maximum memory consumption in all iterations.',
     )
+    parser.add_argument(
+        "--memory_consumption_delay",
+        default=0.5,
+        required=False,
+        type=float,
+        help="delay for memory consumption check in seconds, smaller value will lead to more precised memory consumption, but may affects performance."
+        "It is not recommended to run memory consumption and performance benchmarking in the same time"
+    )
     parser.add_argument('-bs', '--batch_size', type=int, default=1, required=False, help='Batch size value')
     parser.add_argument('--num_beams', type=int, default=1, help='Number of beams in the decoding strategy, activates beam_search if greater than 1')
     parser.add_argument(
@@ -230,6 +238,7 @@ def main():
                      f'{original_torch_thread_nums} to {torch.get_num_threads()}, avoid to use the CPU cores for OpenVINO inference.')
     log.info(out_str)
     if args.memory_consumption:
+        mem_consumption.delay = args.memory_consumption_delay
         mem_consumption.start_collect_mem_consumption_thread()
     try:
         if model_args['use_case'] in ['text_gen', 'code_gen']:

tools/llm_bench/llm_bench_utils/memory_profile.py

@@ -18,6 +18,7 @@ def __init__(self):
         self.g_max_shared_mem_consumption = -1
         self.g_event = Event()
         self.g_data_event = Event()
+        self.delay = 0.5
 
     def collect_memory_consumption(self):
         """Collect the data."""
@@ -54,7 +55,7 @@ def collect_memory_consumption(self):
                     self.g_event.clear()
                     self.g_end_collect_mem = False
                     break
-                time.sleep(0.0001)
+                time.sleep(self.delay)
 
     def start_collect_memory_consumption(self):
         """Start collect."""