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

Add longer accuracy test on CI (#1049)

Browse Source
This commit is contained in:
Lianmin Zheng
2024-08-12 02:21:38 -07:00
committed by GitHub
parent 89f23a5178
commit 41598e0d8e
13 changed files with 385 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
test/srt/run_suite.py
View File

@@ -7,7 +7,7 @@ suites = {
"minimal": [
"test_chunked_prefill.py",
"test_embedding_openai_server.py",
"test_eval_accuracy.py",
"test_eval_accuracy_mini.py",
"test_large_max_new_tokens.py",
"test_openai_server.py",
"test_skip_tokenizer_init.py",

47
test/srt/test_chunked_prefill.py
View File

@@ -10,34 +10,41 @@ from sglang.test.test_utils import (
)
class TestAccuracy(unittest.TestCase):
class TestChunkedPrefill(unittest.TestCase):
@classmethod
def setUpClass(cls):
cls.model = DEFAULT_MODEL_NAME_FOR_TEST
cls.base_url = DEFAULT_URL_FOR_TEST
cls.process = popen_launch_server(
cls.model,
cls.base_url,
def run_mmlu(self, disable_radix_cache):
other_args = ["--chunked-prefill-size", "32"]
if disable_radix_cache:
other_args += ["--disable-radix-cache"]
model = DEFAULT_MODEL_NAME_FOR_TEST
base_url = DEFAULT_URL_FOR_TEST
process = popen_launch_server(
model,
base_url,
timeout=300,
other_args=["--chunked-prefill-size", "32"],
other_args=other_args,
)
@classmethod
def tearDownClass(cls):
kill_child_process(cls.process.pid)
def test_mmlu(self):
args = SimpleNamespace(
base_url=self.base_url,
model=self.model,
base_url=base_url,
model=model,
eval_name="mmlu",
num_examples=20,
num_threads=20,
num_examples=32,
num_threads=32,
)
metrics = run_eval(args)
assert metrics["score"] >= 0.5
try:
metrics = run_eval(args)
assert metrics["score"] >= 0.6
finally:
kill_child_process(process.pid)
def test_chunked_prefill(self):
self.run_mmlu(disable_radix_cache=False)
def test_chunked_prefill_without_radix_cache(self):
self.run_mmlu(disable_radix_cache=True)
if __name__ == "__main__":

68
test/srt/test_eval_accuracy_large.py Normal file
View File

@@ -0,0 +1,68 @@
import unittest
from types import SimpleNamespace
from sglang.srt.utils import kill_child_process
from sglang.test.run_eval import run_eval
from sglang.test.test_utils import (
DEFAULT_MODEL_NAME_FOR_TEST,
DEFAULT_URL_FOR_TEST,
popen_launch_server,
)
class TestEvalAccuracyLarge(unittest.TestCase):
@classmethod
def setUpClass(cls):
cls.model = DEFAULT_MODEL_NAME_FOR_TEST
cls.base_url = "http://127.0.0.1:7157"
cls.process = popen_launch_server(
cls.model,
cls.base_url,
timeout=300,
other_args=["--log-level-http", "warning"],
)
@classmethod
def tearDownClass(cls):
kill_child_process(cls.process.pid)
def test_mmlu(self):
args = SimpleNamespace(
base_url=self.base_url,
model=self.model,
eval_name="mmlu",
num_examples=None,
num_threads=2048,
)
metrics = run_eval(args)
assert metrics["score"] >= 0.70
def test_human_eval(self):
args = SimpleNamespace(
base_url=self.base_url,
model=self.model,
eval_name="humaneval",
num_examples=None,
num_threads=2048,
)
metrics = run_eval(args)
assert metrics["score"] >= 0.65
def test_mgsm_en(self):
args = SimpleNamespace(
base_url=self.base_url,
model=self.model,
eval_name="mgsm_en",
num_examples=None,
num_threads=2048,
)
metrics = run_eval(args)
assert metrics["score"] >= 0.85
if __name__ == "__main__":
unittest.main()

8
test/srt/test_eval_accuracy.py → test/srt/test_eval_accuracy_mini.py
View File

@@ -10,7 +10,7 @@ from sglang.test.test_utils import (
)
class TestAccuracy(unittest.TestCase):
class TestEvalAccuracyMini(unittest.TestCase):
@classmethod
def setUpClass(cls):
@@ -27,12 +27,12 @@ class TestAccuracy(unittest.TestCase):
base_url=self.base_url,
model=self.model,
eval_name="mmlu",
num_examples=20,
num_threads=20,
num_examples=32,
num_threads=32,
)
metrics = run_eval(args)
assert metrics["score"] >= 0.5
assert metrics["score"] >= 0.6
if __name__ == "__main__":

14
test/srt/test_serving_throughput.py
View File

@@ -1,3 +1,4 @@
import os
import unittest
from types import SimpleNamespace
@@ -55,21 +56,30 @@ class TestServingThroughput(unittest.TestCase):
kill_child_process(process.pid)
assert res["completed"] == num_prompts
return res
def test_default(self):
self.run_test(
res = self.run_test(
disable_radix_cache=False,
disable_flashinfer=False,
chunked_prefill_size=-1,
)
if os.getenv("SGLANG_IS_IN_CI", "false") == "true":
# A100 performance
assert res["output_throughput"] >= 1300
def test_default_without_radix_cache(self):
self.run_test(
res = self.run_test(
disable_radix_cache=True,
disable_flashinfer=False,
chunked_prefill_size=-1,
)
if os.getenv("SGLANG_IS_IN_CI", "false") == "true":
# A100 performance
assert res["output_throughput"] >= 1400
def test_default_without_flashinfer(self):
self.run_test(
disable_radix_cache=False,

8
test/srt/test_torch_compile.py
View File

@@ -10,7 +10,7 @@ from sglang.test.test_utils import (
)
class TestAccuracy(unittest.TestCase):
class TestTorchCompile(unittest.TestCase):
@classmethod
def setUpClass(cls):
@@ -29,12 +29,12 @@ class TestAccuracy(unittest.TestCase):
base_url=self.base_url,
model=self.model,
eval_name="mmlu",
num_examples=20,
num_threads=20,
num_examples=32,
num_threads=32,
)
metrics = run_eval(args)
assert metrics["score"] >= 0.5
assert metrics["score"] >= 0.6
if __name__ == "__main__":