Replace time.time() to time.perf_counter() for benchmarking. (#6178)

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

python/sglang/test/few_shot_gsm8k.py

@@ -90,7 +90,7 @@ def run_eval(args):
     #####################################
 
     # Run requests
-    tic = time.time()
+    tic = time.perf_counter()
     states = few_shot_gsm8k.run_batch(
         arguments,
         temperature=args.temperature if hasattr(args, "temperature") else 0,
@@ -99,7 +99,7 @@ def run_eval(args):
         return_logprob=getattr(args, "return_logprob", None),
         logprob_start_len=getattr(args, "logprob_start_len", None),
     )
-    latency = time.time() - tic
+    latency = time.perf_counter() - tic
 
     preds = []
     for i in range(len(states)):

python/sglang/test/few_shot_gsm8k_engine.py

@@ -89,7 +89,7 @@ def run_eval(args):
     }
 
     # Run requests
-    tic = time.time()
+    tic = time.perf_counter()
 
     loop = asyncio.get_event_loop()
 
@@ -98,7 +98,7 @@ def run_eval(args):
     )
 
     # End requests
-    latency = time.time() - tic
+    latency = time.perf_counter() - tic
 
     # Shutdown the engine
     engine.shutdown()

python/sglang/test/run_eval.py

@@ -71,9 +71,9 @@ def run_eval(args):
     )
 
     # Run eval
-    tic = time.time()
+    tic = time.perf_counter()
     result = eval_obj(sampler)
-    latency = time.time() - tic
+    latency = time.perf_counter() - tic
 
     # Dump reports
     metrics = result.metrics | {"score": result.score}

python/sglang/test/test_programs.py

@@ -503,7 +503,7 @@ def test_hellaswag_select():
     #####################################
 
     # Run requests
-    tic = time.time()
+    tic = time.perf_counter()
     rets = few_shot_hellaswag.run_batch(
         arguments,
         temperature=0,
@@ -514,13 +514,13 @@ def test_hellaswag_select():
     preds = []
     for i, ret in enumerate(rets):
         preds.append(choices[i].index(ret["answer"]))
-    latency = time.time() - tic
+    latency = time.perf_counter() - tic
 
     # Compute accuracy
     accuracy = np.mean(np.array(preds) == np.array(labels))
 
     # Test generator style of run_batch
-    tic = time.time()
+    tic = time.perf_counter()
     rets = few_shot_hellaswag.run_batch(
         arguments,
         temperature=0,
@@ -531,7 +531,7 @@ def test_hellaswag_select():
     preds_gen = []
     for i, ret in enumerate(rets):
         preds_gen.append(choices[i].index(ret["answer"]))
-    latency_gen = time.time() - tic
+    latency_gen = time.perf_counter() - tic
 
     # Compute accuracy
     accuracy_gen = np.mean(np.array(preds_gen) == np.array(labels))

python/sglang/test/test_utils.py

@@ -449,9 +449,9 @@ def popen_launch_server(
     else:
         process = subprocess.Popen(command, stdout=None, stderr=None, env=env)
 
-    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:
                 headers = {
                     "Content-Type": "application/json; charset=utf-8",
@@ -584,7 +584,7 @@ class TestFile:
 
 
 def run_unittest_files(files: List[TestFile], timeout_per_file: float):
-    tic = time.time()
+    tic = time.perf_counter()
     success = True
 
     for i, file in enumerate(files):
@@ -599,13 +599,13 @@ def run_unittest_files(files: List[TestFile], timeout_per_file: float):
                 f".\n.\nBegin ({i}/{len(files) - 1}):\npython3 {filename}\n.\n.\n",
                 flush=True,
             )
-            tic = time.time()
+            tic = time.perf_counter()
 
             process = subprocess.Popen(
                 ["python3", filename], stdout=None, stderr=None, env=os.environ
             )
             process.wait()
-            elapsed = time.time() - tic
+            elapsed = time.perf_counter() - tic
 
             print(
                 f".\n.\nEnd ({i}/{len(files) - 1}):\n{filename=}, {elapsed=:.0f}, {estimated_time=}\n.\n.\n",
@@ -631,9 +631,9 @@ def run_unittest_files(files: List[TestFile], timeout_per_file: float):
             break
 
     if success:
-        print(f"Success. Time elapsed: {time.time() - tic:.2f}s", flush=True)
+        print(f"Success. Time elapsed: {time.perf_counter() - tic:.2f}s", flush=True)
     else:
-        print(f"Fail. Time elapsed: {time.time() - tic:.2f}s", flush=True)
+        print(f"Fail. Time elapsed: {time.perf_counter() - tic:.2f}s", flush=True)
 
     return 0 if success else -1