chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[AMD] add testcases for Qwen3 235b Instruct 2507 models (#18805)

Browse Source 
Co-authored-by: bingxche <Bingxu.Chen@amd.com>
This commit is contained in:
mqhc2020
2026-02-25 02:17:52 -08:00
committed by GitHub
co-authored by bingxche
parent 60eeef7370
commit 307a2b73ed
5 changed files with 413 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/amd/test_qwen3_instruct.py
+92
View File
@@ -0,0 +1,92 @@
import os
import unittest
from types import SimpleNamespace
import requests
from sglang.srt.utils import kill_process_tree
from sglang.test.ci.ci_register import register_amd_ci
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.send_one import BenchArgs, send_one_prompt
from sglang.test.test_utils import (
DEFAULT_URL_FOR_TEST,
CustomTestCase,
is_in_amd_ci,
is_in_ci,
popen_launch_server,
write_github_step_summary,
)
register_amd_ci(est_time=3600, suite="nightly-8-gpu-qwen3-235b", nightly=True)
QWEN3_MODEL_PATH = "Qwen/Qwen3-235B-A22B-Instruct-2507"
SERVER_LAUNCH_TIMEOUT = 3600
class TestQwen3Instruct2507(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = QWEN3_MODEL_PATH
cls.base_url = DEFAULT_URL_FOR_TEST
other_args = [
"--tp",
"8",
"--trust-remote-code",
"--attention-backend",
"aiter",
]
env = os.environ.copy()
env["SGLANG_USE_AITER"] = "1"
cls.process = popen_launch_server(
cls.model,
cls.base_url,
timeout=SERVER_LAUNCH_TIMEOUT,
other_args=other_args,
env=env,
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_a_gsm8k(
self,
): # Append an "a" to make this test run first (alphabetically) to warm up the server
requests.get(self.base_url + "/flush_cache")
args = SimpleNamespace(
num_shots=8,
data_path=None,
num_questions=1319,
parallel=1319,
max_new_tokens=512,
host="http://127.0.0.1",
port=int(self.base_url.split(":")[-1]),
)
metrics = run_eval_few_shot_gsm8k(args)
print(f"{metrics=}")
if is_in_ci():
write_github_step_summary(
f"### test_gsm8k ({self.model})\n" f'{metrics["accuracy"]=:.3f}\n'
)
self.assertGreater(metrics["accuracy"], 0.95)
def test_bs_1_speed(self):
args = BenchArgs(port=int(self.base_url.split(":")[-1]), max_new_tokens=2048)
_, speed = send_one_prompt(args)
print(f"{speed=:.2f}")
if is_in_ci():
write_github_step_summary(
f"### test_bs_1_speed ({self.model})\n" f"{speed=:.2f} token/s\n"
)
if is_in_amd_ci():
self.assertGreater(speed, 50)
else:
self.assertGreater(speed, 75)
if __name__ == "__main__":
unittest.main()
test/registered/amd/test_qwen3_instruct_fp8.py
+92
View File
@@ -0,0 +1,92 @@
import os
import unittest
from types import SimpleNamespace
import requests
from sglang.srt.utils import kill_process_tree
from sglang.test.ci.ci_register import register_amd_ci
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.send_one import BenchArgs, send_one_prompt
from sglang.test.test_utils import (
DEFAULT_URL_FOR_TEST,
CustomTestCase,
is_in_amd_ci,
is_in_ci,
popen_launch_server,
write_github_step_summary,
)
register_amd_ci(est_time=3600, suite="nightly-8-gpu-qwen3-235b", nightly=True)
QWEN3_MODEL_PATH = "Qwen/Qwen3-235B-A22B-Instruct-2507-FP8"
SERVER_LAUNCH_TIMEOUT = 3600
class TestQwen3Instruct2507FP8(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = QWEN3_MODEL_PATH
cls.base_url = DEFAULT_URL_FOR_TEST
other_args = [
"--tp",
"4",
"--trust-remote-code",
"--attention-backend",
"aiter",
]
env = os.environ.copy()
env["SGLANG_USE_AITER"] = "1"
cls.process = popen_launch_server(
cls.model,
cls.base_url,
timeout=SERVER_LAUNCH_TIMEOUT,
other_args=other_args,
env=env,
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_a_gsm8k(
self,
): # Append an "a" to make this test run first (alphabetically) to warm up the server
requests.get(self.base_url + "/flush_cache")
args = SimpleNamespace(
num_shots=8,
data_path=None,
num_questions=1319,
parallel=1319,
max_new_tokens=512,
host="http://127.0.0.1",
port=int(self.base_url.split(":")[-1]),
)
metrics = run_eval_few_shot_gsm8k(args)
print(f"{metrics=}")
if is_in_ci():
write_github_step_summary(
f"### test_gsm8k ({self.model})\n" f'{metrics["accuracy"]=:.3f}\n'
)
self.assertGreater(metrics["accuracy"], 0.95)
def test_bs_1_speed(self):
args = BenchArgs(port=int(self.base_url.split(":")[-1]), max_new_tokens=2048)
_, speed = send_one_prompt(args)
print(f"{speed=:.2f}")
if is_in_ci():
write_github_step_summary(
f"### test_bs_1_speed ({self.model})\n" f"{speed=:.2f} token/s\n"
)
if is_in_amd_ci():
self.assertGreater(speed, 40)
else:
self.assertGreater(speed, 60)
if __name__ == "__main__":
unittest.main()
test/registered/amd/test_qwen3_instruct_mxfp4.py
+96
View File
@@ -0,0 +1,96 @@
import os
import unittest
from types import SimpleNamespace
import requests
from sglang.srt.utils import kill_process_tree
from sglang.test.ci.ci_register import register_amd_ci
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.send_one import BenchArgs, send_one_prompt
from sglang.test.test_utils import (
DEFAULT_URL_FOR_TEST,
CustomTestCase,
is_in_amd_ci,
is_in_ci,
popen_launch_server,
write_github_step_summary,
)
register_amd_ci(
est_time=3600, suite="nightly-8-gpu-mi35x-qwen3-235b-mxfp4", nightly=True
)
QWEN3_MODEL_PATH = "amd/Qwen3-235B-A22B-Instruct-2507-mxfp4"
SERVER_LAUNCH_TIMEOUT = 3600
class TestQwen3Instruct2507MXFP4(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = QWEN3_MODEL_PATH
cls.base_url = DEFAULT_URL_FOR_TEST
other_args = [
"--tp",
"4",
"--ep",
"2",
"--trust-remote-code",
"--attention-backend",
"aiter",
]
env = os.environ.copy()
env["SGLANG_USE_AITER"] = "1"
cls.process = popen_launch_server(
cls.model,
cls.base_url,
timeout=SERVER_LAUNCH_TIMEOUT,
other_args=other_args,
env=env,
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_a_gsm8k(
self,
): # Append an "a" to make this test run first (alphabetically) to warm up the server
requests.get(self.base_url + "/flush_cache")
args = SimpleNamespace(
num_shots=8,
data_path=None,
num_questions=1319,
parallel=1319,
max_new_tokens=512,
host="http://127.0.0.1",
port=int(self.base_url.split(":")[-1]),
)
metrics = run_eval_few_shot_gsm8k(args)
print(f"{metrics=}")
if is_in_ci():
write_github_step_summary(
f"### test_gsm8k ({self.model})\n" f'{metrics["accuracy"]=:.3f}\n'
)
self.assertGreater(metrics["accuracy"], 0.93)
def test_bs_1_speed(self):
args = BenchArgs(port=int(self.base_url.split(":")[-1]), max_new_tokens=2048)
_, speed = send_one_prompt(args)
print(f"{speed=:.2f}")
if is_in_ci():
write_github_step_summary(
f"### test_bs_1_speed ({self.model})\n" f"{speed=:.2f} token/s\n"
)
if is_in_amd_ci():
self.assertGreater(speed, 60)
else:
self.assertGreater(speed, 90)
if __name__ == "__main__":
unittest.main()