86 lines
2.5 KiB
Python
86 lines
2.5 KiB
Python
import unittest
|
|
from types import SimpleNamespace
|
|
|
|
from sglang.srt.utils import kill_process_tree
|
|
from sglang.test.few_shot_gsm8k import run_eval
|
|
from sglang.test.kit_matched_stop import MatchedStopMixin
|
|
from sglang.test.test_utils import (
|
|
DEFAULT_EAGLE_DRAFT_MODEL_FOR_TEST,
|
|
DEFAULT_EAGLE_TARGET_MODEL_FOR_TEST,
|
|
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
|
DEFAULT_URL_FOR_TEST,
|
|
CustomTestCase,
|
|
popen_launch_server,
|
|
)
|
|
|
|
|
|
class TestEagleServerBase(CustomTestCase, MatchedStopMixin):
|
|
max_running_requests = 64
|
|
attention_backend = "triton"
|
|
spec_steps = 5
|
|
spec_topk = 1
|
|
spec_draft_tokens = 6
|
|
page_size = 1
|
|
other_launch_args = []
|
|
model = DEFAULT_EAGLE_TARGET_MODEL_FOR_TEST
|
|
draft_model = DEFAULT_EAGLE_DRAFT_MODEL_FOR_TEST
|
|
|
|
@classmethod
|
|
def setUpClass(cls):
|
|
cls.base_url = DEFAULT_URL_FOR_TEST
|
|
launch_args = [
|
|
"--enable-beta-spec",
|
|
"--trust-remote-code",
|
|
"--attention-backend",
|
|
cls.attention_backend,
|
|
"--speculative-algorithm",
|
|
"EAGLE",
|
|
"--speculative-draft-model",
|
|
cls.draft_model,
|
|
"--speculative-num-steps",
|
|
cls.spec_steps,
|
|
"--speculative-eagle-topk",
|
|
cls.spec_topk,
|
|
"--speculative-num-draft-tokens",
|
|
cls.spec_draft_tokens,
|
|
"--page-size",
|
|
str(cls.page_size),
|
|
"--mem-fraction-static",
|
|
"0.75",
|
|
"--max-running-requests",
|
|
str(cls.max_running_requests),
|
|
"--cuda-graph-bs",
|
|
*[str(i) for i in range(1, cls.max_running_requests + 1)],
|
|
]
|
|
launch_args.extend(cls.other_launch_args)
|
|
cls.process = popen_launch_server(
|
|
cls.model,
|
|
cls.base_url,
|
|
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
|
other_args=launch_args,
|
|
)
|
|
|
|
@classmethod
|
|
def tearDownClass(cls):
|
|
kill_process_tree(cls.process.pid)
|
|
|
|
def test_gsm8k(self):
|
|
args = SimpleNamespace(
|
|
num_shots=5,
|
|
data_path=None,
|
|
num_questions=1000,
|
|
max_new_tokens=512,
|
|
parallel=128,
|
|
host="http://127.0.0.1",
|
|
port=int(self.base_url.split(":")[-1]),
|
|
)
|
|
metrics = run_eval(args)
|
|
print(f"TestEagleLargeBS -- {metrics=}")
|
|
self.assertGreater(
|
|
metrics["accuracy"], 0.23
|
|
) # 0.3333 for 60 questions; 0.234 for 1319 questions
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|