[NPU][CI] Fix sglang.test.ascend import failure in multi-node e2e pods (#39403)

This commit is contained in:
pllimax
2026-09-14 22:23:47 +08:00
committed by GitHub
parent 242d8a70c0
commit 433c999dd0
3 changed files with 15 additions and 3 deletions
@@ -6,6 +6,10 @@ if [ ! -f "${sglang_source_path}/${test_case}" ];then
exit 0 exit 0
fi fi
# Leave the image WORKDIR (/sgl-workspace): the sglang repo cloned there can
# shadow the installed package on sys.path.
cd "${sglang_source_path}"
echo "NPU info:" echo "NPU info:"
npu-smi info npu-smi info
@@ -83,6 +83,8 @@ GLM_5_2_W8A8_16P_TWO_NODE_OTHER_ARGS = [
1, 1,
"--speculative-num-draft-tokens", "--speculative-num-draft-tokens",
4, 4,
"--speculative-draft-model-quantization",
"unquant",
] ]
GLM_5_2_W8A8_16P_TWO_NODE_MODEL_CONFIG = { GLM_5_2_W8A8_16P_TWO_NODE_MODEL_CONFIG = {
@@ -103,9 +105,11 @@ class TestNPUGLM_5_2_W8A8_16P_GPQA(TestNpuAccuracyMultiNodePdMixTestCaseBase):
# generation_config = {"max_tokens": 131072, "temperature": 1.0} # generation_config = {"max_tokens": 131072, "temperature": 1.0}
eval_batch_size = 32 eval_batch_size = 32
generation_config = { generation_config = {
"max_tokens": 65536, "max_tokens": 131072,
"top_p": 0.95,
"temperature": 1.0, "temperature": 1.0,
"timeout": 1200, "timeout": 7200,
"retries": 2,
"stream": True, "stream": True,
} }
@@ -23,7 +23,11 @@ from sglang.test.test_utils import (
popen_launch_server, popen_launch_server,
) )
register_npu_ci(est_time=400, suite="base-b-test-4-npu-a3") register_npu_ci(
est_time=400,
suite="base-b-test-4-npu-a3",
disabled="Temporarily disable this test case due to DTS2026091464350",
)
register_npu_ci(est_time=2400, suite="nightly-4-npu-a3", nightly=True) register_npu_ci(est_time=2400, suite="nightly-4-npu-a3", nightly=True)