Use monotonic clock for interval measurement (#6211)

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

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)