diff --git a/test/srt/test_large_max_new_tokens.py b/test/srt/test_large_max_new_tokens.py index 8b493f9bd39..ac0908e0cfa 100644 --- a/test/srt/test_large_max_new_tokens.py +++ b/test/srt/test_large_max_new_tokens.py @@ -1,6 +1,4 @@ -import json import os -import time import unittest from concurrent.futures import ThreadPoolExecutor @@ -16,7 +14,7 @@ ) -class TestOpenAIServer(unittest.TestCase): +class TestLargeMaxNewTokens(unittest.TestCase): @classmethod def setUpClass(cls): cls.model = DEFAULT_MODEL_NAME_FOR_TEST @@ -57,7 +55,7 @@ def test_chat_completion(self): num_requests = 4 futures = [] - with ThreadPoolExecutor(16) as executor: + with ThreadPoolExecutor(num_requests) as executor: for i in range(num_requests): futures.append(executor.submit(self.run_chat_completion))