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

use sgl custom all reduce (#4441)

Browse Source
This commit is contained in:
Yineng Zhang
2025-03-18 00:46:41 -07:00
committed by GitHub
parent 45212ce18b
commit c787298547
3 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
.github/workflows/pr-test.yml vendored
View File

@@ -221,9 +221,9 @@ jobs:
timeout-minutes: 10
run: |
cd test/srt
python3 -m unittest test_bench_one_batch.TestBenchOneBatch.test_moe_tp2_bs1
USE_VLLM_CUSTOM_ALLREDUCE=1 python3 -m unittest test_bench_one_batch.TestBenchOneBatch.test_moe_tp2_bs1
USE_VLLM_CUSTOM_ALLREDUCE=0 python3 -m unittest test_bench_one_batch.TestBenchOneBatch.test_moe_tp2_bs1
python3 -m unittest test_bench_one_batch.TestBenchOneBatch.test_moe_tp2_bs1
- name: Benchmark single latency + torch.compile (TP=2)
timeout-minutes: 10

2
python/sglang/srt/_custom_ops.py
View File

@@ -10,7 +10,7 @@ from sglang.srt.utils import get_bool_env_var, is_hip, is_hpu
logger = logging.getLogger(__name__)
use_vllm_custom_allreduce = get_bool_env_var(
"USE_VLLM_CUSTOM_ALLREDUCE", default="true"
"USE_VLLM_CUSTOM_ALLREDUCE", default="false"
)
if not is_hpu():

2
test/srt/test_bench_one_batch.py
View File

@@ -29,7 +29,7 @@ class TestBenchOneBatch(unittest.TestCase):
)
use_vllm_custom_allreduce = get_bool_env_var(
"USE_VLLM_CUSTOM_ALLREDUCE", default="true"
"USE_VLLM_CUSTOM_ALLREDUCE", default="false"
)
if is_in_ci():