Use monotonic clock for interval measurement (#6211)

Signed-off-by: Lifu Huang <lifu.hlf@gmail.com>

examples/frontend_language/usage/llava_video/srt_example_llava_v.py

@@ -109,9 +109,9 @@ def batch(video_dir, save_dir, cur_chunk, num_chunks, num_frames=16, batch_size=
             for video_path in batch_video_files
         ]
 
-        start_time = time.time()
+        start_time = time.perf_counter()
         states = video_qa.run_batch(batch_input, max_new_tokens=512, temperature=0.2)
-        total_time = time.time() - start_time
+        total_time = time.perf_counter() - start_time
         average_time = total_time / len(batch_video_files)
         print(
             f"Number of videos in batch: {len(batch_video_files)}. Average processing time per video: {average_time:.2f} seconds. Total time for this batch: {total_time:.2f} seconds"
@@ -240,11 +240,11 @@ if __name__ == "__main__":
             for f in os.listdir(root)
             if f.endswith((".mp4", ".avi", ".mov"))
         ]  # Add more extensions if needed
-    start_time = time.time()  # Start time for processing a single video
+    start_time = time.perf_counter()  # Start time for processing a single video
     for cur_video in video_files[:1]:
         print(cur_video)
         single(cur_video, num_frames)
-    end_time = time.time()  # End time for processing a single video
+    end_time = time.perf_counter()  # End time for processing a single video
     total_time = end_time - start_time
     average_time = total_time / len(
         video_files

examples/runtime/engine/fastapi_engine_inference.py

@@ -89,9 +89,9 @@ def start_server(args, timeout=60):
 
     process = subprocess.Popen(command, stdout=None, stderr=None)
 
-    start_time = time.time()
+    start_time = time.perf_counter()
     with requests.Session() as session:
-        while time.time() - start_time < timeout:
+        while time.perf_counter() - start_time < timeout:
             try:
                 # Check the /docs endpoint which FastAPI provides by default
                 response = session.get(

examples/runtime/multimodal/llava_onevision_server.py

@@ -150,7 +150,7 @@ def video_stream_request_test(client, video_path):
 
 def image_speed_test(client):
     print("----------------------Image Speed Test----------------------")
-    start_time = time.time()
+    start_time = time.perf_counter()
     request = client.chat.completions.create(
         model="default",
         messages=[
@@ -173,7 +173,7 @@ def image_speed_test(client):
         temperature=0,
         max_tokens=1024,
     )
-    end_time = time.time()
+    end_time = time.perf_counter()
     response = request.choices[0].message.content
     print(response)
     print("-" * 30)
@@ -184,14 +184,14 @@ def video_speed_test(client, video_path):
     print("------------------------Video Speed Test------------------------")
     messages = prepare_video_messages(video_path)
 
-    start_time = time.time()
+    start_time = time.perf_counter()
     video_request = client.chat.completions.create(
         model="default",
         messages=messages,
         temperature=0,
         max_tokens=1024,
     )
-    end_time = time.time()
+    end_time = time.perf_counter()
     video_response = video_request.choices[0].message.content
     print(video_response)
     print("-" * 30)