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

[NPU]Optimize the PR pipeline to reduce E2E runtime (#18767)

Browse Source 
Co-authored-by: Sugar920 <121632458+Sugar920@users.noreply.github.com>
This commit is contained in:
Cherry_ming
2026-02-28 17:06:03 +08:00
committed by GitHub
co-authored by Sugar920
parent 366574b2b8
commit 1eb40d8d45
12 changed files with 17 additions and 238 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/srt/ascend/test_ascend_compile_graph_tp1_bf16.py
-22
View File
@@ -9,9 +9,7 @@ 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 = {
@@ -77,26 +75,6 @@ class TestAscendTp1Bf16(CustomTestCase):
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()
test/srt/ascend/test_ascend_deepep.py
+1 -23
View File
@@ -8,9 +8,7 @@ from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.test_utils import (
DEFAULT_URL_FOR_TEST,
CustomTestCase,
is_in_ci,
popen_launch_server,
run_bench_offline_throughput,
)
TEST_MODEL_MATRIX = {
@@ -78,7 +76,7 @@ class TestAscendDeepEP(CustomTestCase):
args = SimpleNamespace(
num_shots=5,
data_path=None,
num_questions=1319,
num_questions=500,
max_new_tokens=512,
parallel=128,
host=f"http://{self.url.hostname}",
@@ -93,26 +91,6 @@ class TestAscendDeepEP(CustomTestCase):
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()
test/srt/ascend/test_ascend_graph_tp1_bf16.py
-22
View File
@@ -8,9 +8,7 @@ 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 = {
@@ -70,26 +68,6 @@ class TestAscendGraphTp1Bf16(CustomTestCase):
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()
test/srt/ascend/test_ascend_graph_tp2_bf16.py
-22
View File
@@ -8,9 +8,7 @@ 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 = {
@@ -72,26 +70,6 @@ class TestAscendGraphTp2Bf16(CustomTestCase):
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()
test/srt/ascend/test_ascend_hicache_mha.py
-22
View File
@@ -8,9 +8,7 @@ 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 = {
@@ -73,26 +71,6 @@ class TestAscendMhaHicache(CustomTestCase):
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()
test/srt/ascend/test_ascend_hicache_mla.py
-22
View File
@@ -8,9 +8,7 @@ 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 = {
@@ -75,26 +73,6 @@ class TestAscendMlaHicache(CustomTestCase):
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()
test/srt/ascend/test_ascend_mla_fia_w8a8int8.py
-22
View File
@@ -9,9 +9,7 @@ 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 = {
@@ -76,26 +74,6 @@ class TestAscendMlaW8A8Int8(CustomTestCase):
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()
test/srt/ascend/test_ascend_mla_w8a8int8.py
-22
View File
@@ -8,9 +8,7 @@ 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 = {
@@ -74,26 +72,6 @@ class TestAscendMlaW8A8Int8(CustomTestCase):
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()
test/srt/ascend/test_ascend_tp2_bf16.py
-22
View File
@@ -8,9 +8,7 @@ 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 = {
@@ -73,26 +71,6 @@ class TestAscendTp2Bf16(CustomTestCase):
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()
test/srt/ascend/test_ascend_tp4_bf16.py
-22
View File
@@ -7,9 +7,7 @@ from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.test_utils import (
DEFAULT_URL_FOR_TEST,
CustomTestCase,
is_in_ci,
popen_launch_server,
run_bench_offline_throughput,
)
TEST_MODEL_MATRIX = {
@@ -76,26 +74,6 @@ class TestAscendTp4Bf16(CustomTestCase):
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()