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

sglangv0.5.2 & support Qwen3-Next-80B-A3B-Instruct

Browse Source
This commit is contained in:
maxiao1
2025-09-13 17:00:20 +08:00
commit 118f1fc726
2037 changed files with 515371 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

121
test/srt/ascend/test_ascend_deepep.py Normal file
View File

@@ -0,0 +1,121 @@
import os
import unittest
from types import SimpleNamespace
from urllib.parse import urlparse
from sglang.srt.utils import kill_process_tree
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.test_utils import (
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
is_in_ci,
popen_launch_server,
run_bench_offline_throughput,
)
TEST_MODEL_MATRIX = {
"/root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-R1-0528-W8A8": {
"accuracy": 0.95,
"latency": 1000,
"output_throughput": 6,
},
}
class TestAscendDeepEP(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.models = TEST_MODEL_MATRIX.keys()
cls.base_url = DEFAULT_URL_FOR_TEST
cls.url = urlparse(DEFAULT_URL_FOR_TEST)
cls.common_args = [
"--trust-remote-code",
"--attention-backend",
"ascend",
"--quantization",
"w8a8_int8",
"--mem-fraction-static",
0.9,
"--max-running-requests",
32,
"--disable-radix-cache",
"--chunked-prefill-size",
32768,
"--disable-cuda-graph",
"--tp-size",
16,
"--dp-size",
1,
"--ep-size",
16,
"--moe-a2a-backend",
"deepep",
"--deepep-mode",
"auto",
]
cls.extra_envs = {
"HCCL_BUFFSIZE": "500",
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "32",
}
os.environ.update(cls.extra_envs)
def test_a_gsm8k(self):
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing accuracy: {model} ===##")
process = popen_launch_server(
model,
self.base_url,
timeout=1500,
other_args=[
*self.common_args,
],
)
try:
args = SimpleNamespace(
num_shots=5,
data_path=None,
num_questions=1319,
max_new_tokens=512,
parallel=128,
host=f"http://{self.url.hostname}",
port=int(self.url.port),
)
metrics = run_eval_few_shot_gsm8k(args)
self.assertGreaterEqual(
metrics["accuracy"],
TEST_MODEL_MATRIX[model]["accuracy"],
)
finally:
kill_process_tree(process.pid)
def test_b_throughput(self):
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing throughput: {model} ===##")
output_throughput = run_bench_offline_throughput(
model,
[
*self.common_args,
],
)
print(f"##=== {model} throughput: {output_throughput} ===##")
if is_in_ci():
self.assertGreater(
output_throughput,
TEST_MODEL_MATRIX[model]["output_throughput"],
)
if __name__ == "__main__":
unittest.main()

95
test/srt/ascend/test_ascend_graph_tp1_bf16.py Normal file
View File

@@ -0,0 +1,95 @@
import unittest
from types import SimpleNamespace
from urllib.parse import urlparse
from sglang.srt.utils import kill_process_tree
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.test_utils import (
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
is_in_ci,
popen_launch_server,
run_bench_offline_throughput,
)
TEST_MODEL_MATRIX = {
"Qwen/Qwen2.5-7B-Instruct": {
"accuracy": 0.85,
"latency": 150,
"output_throughput": 30,
},
}
class TestAscendGraphTp1Bf16(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.models = TEST_MODEL_MATRIX.keys()
cls.base_url = DEFAULT_URL_FOR_TEST
cls.url = urlparse(DEFAULT_URL_FOR_TEST)
cls.common_args = [
"--trust-remote-code",
"--mem-fraction-static",
0.8,
"--attention-backend",
"ascend",
]
def test_a_gsm8k(self):
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing accuracy: {model} ===##")
process = popen_launch_server(
model,
self.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=[
*self.common_args,
],
)
try:
args = SimpleNamespace(
num_shots=5,
data_path=None,
num_questions=1319,
max_new_tokens=512,
parallel=128,
host=f"http://{self.url.hostname}",
port=int(self.url.port),
)
metrics = run_eval_few_shot_gsm8k(args)
self.assertGreaterEqual(
metrics["accuracy"],
TEST_MODEL_MATRIX[model]["accuracy"],
)
finally:
kill_process_tree(process.pid)
def test_b_throughput(self):
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing throughput: {model} ===##")
output_throughput = run_bench_offline_throughput(
model,
[
*self.common_args,
],
)
print(f"##=== {model} throughput: {output_throughput} ===##")
if is_in_ci():
self.assertGreater(
output_throughput,
TEST_MODEL_MATRIX[model]["output_throughput"],
)
if __name__ == "__main__":
unittest.main()

97
test/srt/ascend/test_ascend_graph_tp2_bf16.py Normal file
View File

@@ -0,0 +1,97 @@
import unittest
from types import SimpleNamespace
from urllib.parse import urlparse
from sglang.srt.utils import kill_process_tree
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.test_utils import (
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
is_in_ci,
popen_launch_server,
run_bench_offline_throughput,
)
TEST_MODEL_MATRIX = {
"Qwen/Qwen2.5-7B-Instruct": {
"accuracy": 0.85,
"latency": 180,
"output_throughput": 20,
},
}
class TestAscendGraphTp2Bf16(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.models = TEST_MODEL_MATRIX.keys()
cls.base_url = DEFAULT_URL_FOR_TEST
cls.url = urlparse(DEFAULT_URL_FOR_TEST)
cls.common_args = [
"--trust-remote-code",
"--mem-fraction-static",
0.8,
"--attention-backend",
"ascend",
"--tp-size",
2,
]
def test_a_gsm8k(self):
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing accuracy: {model} ===##")
process = popen_launch_server(
model,
self.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=[
*self.common_args,
],
)
try:
args = SimpleNamespace(
num_shots=5,
data_path=None,
num_questions=1319,
max_new_tokens=512,
parallel=128,
host=f"http://{self.url.hostname}",
port=int(self.url.port),
)
metrics = run_eval_few_shot_gsm8k(args)
self.assertGreaterEqual(
metrics["accuracy"],
TEST_MODEL_MATRIX[model]["accuracy"],
)
finally:
kill_process_tree(process.pid)
def test_b_throughput(self):
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing throughput: {model} ===##")
output_throughput = run_bench_offline_throughput(
model,
[
*self.common_args,
],
)
print(f"##=== {model} throughput: {output_throughput} ===##")
if is_in_ci():
self.assertGreater(
output_throughput,
TEST_MODEL_MATRIX[model]["output_throughput"],
)
if __name__ == "__main__":
unittest.main()

103
test/srt/ascend/test_ascend_mla_fia_w8a8int8.py Normal file
View File

@@ -0,0 +1,103 @@
import os
import unittest
from types import SimpleNamespace
from urllib.parse import urlparse
from sglang.srt.utils import kill_process_tree
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.test_utils import (
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
is_in_ci,
popen_launch_server,
run_bench_offline_throughput,
)
TEST_MODEL_MATRIX = {
"/root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V2-Lite-W8A8": {
"accuracy": 0.34,
"latency": 1000,
"output_throughput": 6,
},
}
class TestAscendMlaW8A8Int8(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.models = TEST_MODEL_MATRIX.keys()
cls.base_url = DEFAULT_URL_FOR_TEST
cls.url = urlparse(DEFAULT_URL_FOR_TEST)
cls.common_args = [
"--trust-remote-code",
"--disable-cuda-graph",
"--mem-fraction-static",
0.8,
"--attention-backend",
"ascend",
"--quantization",
"w8a8_int8",
"--tp-size",
2,
"--disable-radix-cache",
]
def test_a_gsm8k(self):
os.environ["ASCEND_USE_FIA"] = "true"
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing accuracy: {model} ===##")
process = popen_launch_server(
model,
self.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=[
*self.common_args,
],
)
try:
args = SimpleNamespace(
num_shots=5,
data_path=None,
num_questions=1319,
max_new_tokens=512,
parallel=128,
host=f"http://{self.url.hostname}",
port=int(self.url.port),
)
metrics = run_eval_few_shot_gsm8k(args)
self.assertGreaterEqual(
metrics["accuracy"],
TEST_MODEL_MATRIX[model]["accuracy"],
)
finally:
kill_process_tree(process.pid)
def test_b_throughput(self):
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing throughput: {model} ===##")
output_throughput = run_bench_offline_throughput(
model,
[
*self.common_args,
],
)
print(f"##=== {model} throughput: {output_throughput} ===##")
if is_in_ci():
self.assertGreater(
output_throughput,
TEST_MODEL_MATRIX[model]["output_throughput"],
)
if __name__ == "__main__":
unittest.main()

101
test/srt/ascend/test_ascend_mla_w8a8int8.py Normal file
View File

@@ -0,0 +1,101 @@
import unittest
from types import SimpleNamespace
from urllib.parse import urlparse
from sglang.srt.utils import kill_process_tree
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.test_utils import (
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
is_in_ci,
popen_launch_server,
run_bench_offline_throughput,
)
TEST_MODEL_MATRIX = {
"/root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V2-Lite-W8A8": {
"accuracy": 0.34,
"latency": 1000,
"output_throughput": 6,
},
}
class TestAscendMlaW8A8Int8(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.models = TEST_MODEL_MATRIX.keys()
cls.base_url = DEFAULT_URL_FOR_TEST
cls.url = urlparse(DEFAULT_URL_FOR_TEST)
cls.common_args = [
"--trust-remote-code",
"--disable-cuda-graph",
"--mem-fraction-static",
0.8,
"--attention-backend",
"ascend",
"--quantization",
"w8a8_int8",
"--tp-size",
4,
"--disable-radix-cache",
]
def test_a_gsm8k(self):
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing accuracy: {model} ===##")
process = popen_launch_server(
model,
self.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=[
*self.common_args,
],
)
try:
args = SimpleNamespace(
num_shots=5,
data_path=None,
num_questions=1319,
max_new_tokens=512,
parallel=128,
host=f"http://{self.url.hostname}",
port=int(self.url.port),
)
metrics = run_eval_few_shot_gsm8k(args)
self.assertGreaterEqual(
metrics["accuracy"],
TEST_MODEL_MATRIX[model]["accuracy"],
)
finally:
kill_process_tree(process.pid)
def test_b_throughput(self):
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing throughput: {model} ===##")
output_throughput = run_bench_offline_throughput(
model,
[
*self.common_args,
],
)
print(f"##=== {model} throughput: {output_throughput} ===##")
if is_in_ci():
self.assertGreater(
output_throughput,
TEST_MODEL_MATRIX[model]["output_throughput"],
)
if __name__ == "__main__":
unittest.main()

96
test/srt/ascend/test_ascend_tp1_bf16.py Normal file
View File

@@ -0,0 +1,96 @@
import unittest
from types import SimpleNamespace
from urllib.parse import urlparse
from sglang.srt.utils import kill_process_tree
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.test_utils import (
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
is_in_ci,
popen_launch_server,
run_bench_offline_throughput,
)
TEST_MODEL_MATRIX = {
"Qwen/Qwen2.5-7B-Instruct": {
"accuracy": 0.84,
"latency": 150,
"output_throughput": 30,
},
}
class TestAscendTp1Bf16(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.models = TEST_MODEL_MATRIX.keys()
cls.base_url = DEFAULT_URL_FOR_TEST
cls.url = urlparse(DEFAULT_URL_FOR_TEST)
cls.common_args = [
"--trust-remote-code",
"--disable-cuda-graph",
"--mem-fraction-static",
0.8,
"--attention-backend",
"ascend",
]
def test_a_gsm8k(self):
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing accuracy: {model} ===##")
process = popen_launch_server(
model,
self.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=[
*self.common_args,
],
)
try:
args = SimpleNamespace(
num_shots=5,
data_path=None,
num_questions=1319,
max_new_tokens=512,
parallel=128,
host=f"http://{self.url.hostname}",
port=int(self.url.port),
)
metrics = run_eval_few_shot_gsm8k(args)
self.assertGreaterEqual(
metrics["accuracy"],
TEST_MODEL_MATRIX[model]["accuracy"],
)
finally:
kill_process_tree(process.pid)
def test_b_throughput(self):
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing throughput: {model} ===##")
output_throughput = run_bench_offline_throughput(
model,
[
*self.common_args,
],
)
print(f"##=== {model} throughput: {output_throughput} ===##")
if is_in_ci():
self.assertGreater(
output_throughput,
TEST_MODEL_MATRIX[model]["output_throughput"],
)
if __name__ == "__main__":
unittest.main()

98
test/srt/ascend/test_ascend_tp2_bf16.py Normal file
View File

@@ -0,0 +1,98 @@
import unittest
from types import SimpleNamespace
from urllib.parse import urlparse
from sglang.srt.utils import kill_process_tree
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.test_utils import (
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
is_in_ci,
popen_launch_server,
run_bench_offline_throughput,
)
TEST_MODEL_MATRIX = {
"Qwen/Qwen2.5-7B-Instruct": {
"accuracy": 0.85,
"latency": 180,
"output_throughput": 20,
},
}
class TestAscendTp2Bf16(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.models = TEST_MODEL_MATRIX.keys()
cls.base_url = DEFAULT_URL_FOR_TEST
cls.url = urlparse(DEFAULT_URL_FOR_TEST)
cls.common_args = [
"--trust-remote-code",
"--disable-cuda-graph",
"--mem-fraction-static",
0.8,
"--attention-backend",
"ascend",
"--tp-size",
2,
]
def test_a_gsm8k(self):
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing accuracy: {model} ===##")
process = popen_launch_server(
model,
self.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=[
*self.common_args,
],
)
try:
args = SimpleNamespace(
num_shots=5,
data_path=None,
num_questions=1319,
max_new_tokens=512,
parallel=128,
host=f"http://{self.url.hostname}",
port=int(self.url.port),
)
metrics = run_eval_few_shot_gsm8k(args)
self.assertGreaterEqual(
metrics["accuracy"],
TEST_MODEL_MATRIX[model]["accuracy"],
)
finally:
kill_process_tree(process.pid)
def test_b_throughput(self):
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing throughput: {model} ===##")
output_throughput = run_bench_offline_throughput(
model,
[
*self.common_args,
],
)
print(f"##=== {model} throughput: {output_throughput} ===##")
if is_in_ci():
self.assertGreater(
output_throughput,
TEST_MODEL_MATRIX[model]["output_throughput"],
)
if __name__ == "__main__":
unittest.main()

101
test/srt/ascend/test_ascend_tp2_fia_bf16.py Normal file
View File

@@ -0,0 +1,101 @@
import os
import unittest
from types import SimpleNamespace
from urllib.parse import urlparse
from sglang.srt.utils import kill_process_tree
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.test_utils import (
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
is_in_ci,
popen_launch_server,
run_bench_offline_throughput,
)
TEST_MODEL_MATRIX = {
"Qwen/Qwen2.5-7B-Instruct": {
"accuracy": 0.85,
"latency": 180,
"output_throughput": 20,
},
}
class TestAscendTp2Bf16(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.models = TEST_MODEL_MATRIX.keys()
cls.base_url = DEFAULT_URL_FOR_TEST
cls.url = urlparse(DEFAULT_URL_FOR_TEST)
cls.common_args = [
"--trust-remote-code",
"--disable-cuda-graph",
"--mem-fraction-static",
0.8,
"--attention-backend",
"ascend",
"--tp-size",
2,
"--disable-radix-cache",
]
def test_a_gsm8k(self):
os.environ["ASCEND_USE_FIA"] = "true"
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing accuracy: {model} ===##")
process = popen_launch_server(
model,
self.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=[
*self.common_args,
],
)
try:
args = SimpleNamespace(
num_shots=5,
data_path=None,
num_questions=1319,
max_new_tokens=512,
parallel=128,
host=f"http://{self.url.hostname}",
port=int(self.url.port),
)
metrics = run_eval_few_shot_gsm8k(args)
self.assertGreaterEqual(
metrics["accuracy"],
TEST_MODEL_MATRIX[model]["accuracy"],
)
finally:
kill_process_tree(process.pid)
def test_b_throughput(self):
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing throughput: {model} ===##")
output_throughput = run_bench_offline_throughput(
model,
[
*self.common_args,
],
)
print(f"##=== {model} throughput: {output_throughput} ===##")
if is_in_ci():
self.assertGreater(
output_throughput,
TEST_MODEL_MATRIX[model]["output_throughput"],
)
if __name__ == "__main__":
unittest.main()

101
test/srt/ascend/test_ascend_tp4_bf16.py Normal file
View File

@@ -0,0 +1,101 @@
import unittest
from types import SimpleNamespace
from urllib.parse import urlparse
from sglang.srt.utils import kill_process_tree
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.test_utils import (
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
is_in_ci,
popen_launch_server,
run_bench_offline_throughput,
)
TEST_MODEL_MATRIX = {
"Qwen/Qwen3-30B-A3B-Instruct-2507": {
"accuracy": 0.90,
"latency": 180,
"output_throughput": 20,
},
}
class TestAscendTp4Bf16(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.models = TEST_MODEL_MATRIX.keys()
cls.base_url = DEFAULT_URL_FOR_TEST
cls.url = urlparse(DEFAULT_URL_FOR_TEST)
cls.common_args = [
"--trust-remote-code",
"--mem-fraction-static",
0.7,
"--max-running-requests",
32,
"--attention-backend",
"ascend",
"--cuda-graph-max-bs",
32,
"--tp-size",
4,
]
def test_a_gsm8k(self):
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing accuracy: {model} ===##")
process = popen_launch_server(
model,
self.base_url,
timeout=1800,
other_args=[
*self.common_args,
],
)
try:
args = SimpleNamespace(
num_shots=5,
data_path=None,
num_questions=1319,
max_new_tokens=512,
parallel=128,
host=f"http://{self.url.hostname}",
port=int(self.url.port),
)
metrics = run_eval_few_shot_gsm8k(args)
self.assertGreaterEqual(
metrics["accuracy"],
TEST_MODEL_MATRIX[model]["accuracy"],
)
finally:
kill_process_tree(process.pid)
def test_b_throughput(self):
for model in self.models:
with self.subTest(model=model):
print(f"##=== Testing throughput: {model} ===##")
output_throughput = run_bench_offline_throughput(
model,
[
*self.common_args,
],
)
print(f"##=== {model} throughput: {output_throughput} ===##")
if is_in_ci():
self.assertGreater(
output_throughput,
TEST_MODEL_MATRIX[model]["output_throughput"],
)
if __name__ == "__main__":
unittest.main()

104
test/srt/ascend/test_ascend_w8a8_quantization.py Normal file
View File

@@ -0,0 +1,104 @@
"""
Usage:
python3 -m unittest test_ascend_w8a8_quantization.TestAscendW8A8.test_gsm8k
"""
import os
import time
import unittest
from types import SimpleNamespace
from urllib.parse import urlparse
import requests
from sglang.srt.utils import kill_process_tree
from sglang.test.few_shot_gsm8k import run_eval
from sglang.test.test_utils import (
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
CustomTestCase,
is_in_ci,
popen_launch_server,
)
if "ASCEND_RT_VISIBLE_DEVICES" not in os.environ:
os.environ["ASCEND_RT_VISIBLE_DEVICES"] = "0,1"
DEFAULT_PORT_FOR_SRT_TEST_RUNNER = (
7000 + int(os.environ.get("ASCEND_RT_VISIBLE_DEVICES", "0")[0]) * 100
)
DEFAULT_URL_FOR_TEST = f"http://127.0.0.1:{DEFAULT_PORT_FOR_SRT_TEST_RUNNER + 1000}"
class TestAscendW8A8(CustomTestCase):
@classmethod
def setUpClass(cls):
cls.model = "vllm-ascend/Qwen2.5-0.5B-Instruct-w8a8"
cls.base_url = DEFAULT_URL_FOR_TEST
cls.process = popen_launch_server(
cls.model,
cls.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
other_args=[
"--trust-remote-code",
"--disable-cuda-graph",
"--device",
"npu",
"--attention-backend",
"ascend",
"--quantization",
"w8a8_int8",
],
)
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.process.pid)
def test_gsm8k(self):
base_url = DEFAULT_URL_FOR_TEST
url = urlparse(base_url)
args = SimpleNamespace(
num_shots=5,
data_path=None,
num_questions=200,
max_new_tokens=512,
parallel=128,
host=f"http://{url.hostname}",
port=int(url.port),
)
metrics = run_eval(args)
print(metrics)
self.assertGreaterEqual(metrics["accuracy"], 0.25)
self.assertGreaterEqual(metrics["output_throughput"], 1000)
def run_decode(self, max_new_tokens):
response = requests.post(
self.base_url + "/generate",
json={
"text": "The capital of France is",
"sampling_params": {
"temperature": 0,
"max_new_tokens": max_new_tokens,
},
"ignore_eos": True,
},
)
return response.json()
def test_throughput(self):
max_tokens = 256
tic = time.perf_counter()
res = self.run_decode(max_tokens)
tok = time.perf_counter()
print(res["text"])
throughput = max_tokens / (tok - tic)
print(f"Throughput: {throughput} tokens/s")
if is_in_ci():
self.assertGreaterEqual(throughput, 25)
if __name__ == "__main__":
unittest.main()