Files
sglang/test/srt/test_gpt_oss_4gpu.py
2025-08-09 00:34:23 -07:00

43 lines
1.1 KiB
Python

import unittest
from test_gpt_oss_common import BaseTestGptOss
class TestGptOss4Gpu(BaseTestGptOss):
def test_bf16_120b(self):
self.run_test(
model_variant="120b",
quantization="bf16",
expected_score_of_reasoning_effort={
"low": 0.61,
# remove to speed up
# "medium": 0.61,
# "high": 0.61,
},
other_args=["--tp", "4", "--cuda-graph-max-bs", "200"],
)
def test_mxfp4_120b(self):
self.run_test(
model_variant="120b",
quantization="mxfp4",
expected_score_of_reasoning_effort={
"low": 0.61,
# remove to speed up
# "medium": 0.61,
# "high": 0.61,
},
other_args=[
"--tp",
"4",
"--cuda-graph-max-bs",
"200",
"--mem-fraction-static",
"0.93",
],
)
if __name__ == "__main__":
unittest.main()