EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Clean up server_args.py to have a dedicated function for model specific adjustments (#8983)

Browse Source
This commit is contained in:
Lianmin Zheng
2025-08-08 19:56:50 -07:00
committed by GitHub
parent 23f2afb2ce
commit 706bd69cc5
24 changed files with 201 additions and 340 deletions
Download Patch File Download Diff File Expand all files Collapse all files

61
test/srt/test_mla_deepseek_v3.py
View File

@@ -149,66 +149,5 @@ class TestDeepseekV3MTP(CustomTestCase):
self.assertGreater(avg_spec_accept_length, 2.5)
# compatible with old APIs
class TestDeepseekV3MTPWithDraft(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = "lmsys/sglang-ci-dsv3-test"
cls.base_url = DEFAULT_URL_FOR_TEST
other_args = [
"--trust-remote-code",
"--cuda-graph-max-bs",
"2",
"--disable-radix",
"--enable-torch-compile",
"--torch-compile-max-bs",
"1",
"--speculative-algorithm",
"EAGLE",
"--speculative-draft",
"lmsys/sglang-ci-dsv3-test-NextN",
"--speculative-num-steps",
"2",
"--speculative-eagle-topk",
"4",
"--speculative-num-draft-tokens",
"4",
]
cls.process = popen_launch_server(
cls.model,
cls.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=other_args,
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_gsm8k(self):
requests.get(self.base_url + "/flush_cache")
args = SimpleNamespace(
num_shots=5,
data_path=None,
num_questions=200,
max_new_tokens=512,
parallel=128,
host="http://127.0.0.1",
port=int(self.base_url.split(":")[-1]),
)
metrics = run_eval_few_shot_gsm8k(args)
print(metrics)
self.assertGreater(metrics["accuracy"], 0.60)
server_info = requests.get(self.base_url + "/get_server_info")
avg_spec_accept_length = server_info.json()["internal_states"][0][
"avg_spec_accept_length"
]
print(f"{avg_spec_accept_length=}")
self.assertGreater(avg_spec_accept_length, 2.5)
if __name__ == "__main__":
unittest.main()