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

[CI] Split test cases in CI for better load balancing (#2180)

Browse Source
This commit is contained in:
Lianmin Zheng
2024-11-25 04:58:16 -08:00
committed by GitHub
parent 538fa0ae13
commit 254fd130e2
6 changed files with 37 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
test/srt/test_chunked_prefill.py
View File

@@ -4,12 +4,7 @@ python3 -m unittest test_chunked_prefill.TestChunkedPrefill.test_mixed_chunked_p
import unittest
from sglang.test.test_utils import (
DEFAULT_MODEL_NAME_FOR_TEST,
run_bench_serving,
run_mmlu_test,
run_mulit_request_test,
)
from sglang.test.test_utils import run_mmlu_test, run_mulit_request_test
class TestChunkedPrefill(unittest.TestCase):
@@ -25,21 +20,6 @@ class TestChunkedPrefill(unittest.TestCase):
def test_mixed_chunked_prefill_without_radix_cache(self):
run_mmlu_test(disable_radix_cache=True, enable_mixed_chunk=True)
def test_no_chunked_prefill(self):
run_mmlu_test(
disable_radix_cache=False, enable_mixed_chunk=False, chunked_prefill_size=-1
)
def test_no_chunked_prefill_without_radix_cache(self):
res = run_bench_serving(
model=DEFAULT_MODEL_NAME_FOR_TEST,
num_prompts=10,
request_rate=float("inf"),
other_server_args=["--disable-radix-cache", "--chunked-prefill-size", "-1"],
)
assert res["completed"] == 10
def test_mixed_chunked_prefill_multi_requests(self):
run_mulit_request_test(
enable_mixed_chunk=True,