EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[Nightly] Optimize nightly CI (#4509)

Browse Source 
### What this PR does / why we need it?
1. Optimize multi-node waiting logic
2. Remove the `tee` pipeline for logs, which will lead to hang issue

### How was this patch tested?


- vLLM version: v0.12.0
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.12.0

Signed-off-by: wangli <wangli858794774@gmail.com>
This commit is contained in:
Li Wang
2025-12-04 22:31:07 +08:00
committed by GitHub
parent fb15fec662
commit 283bc5c7ba
8 changed files with 90 additions and 105 deletions
Download Patch File Download Diff File Expand all files Collapse all files

50
tests/e2e/nightly/multi_node/config/models/GLM-4_5.yaml
View File

@@ -1,50 +0,0 @@
test_name: "test GLM-4.5 multi-dp"
model: "ZhipuAI/GLM-4.5"
num_nodes: 2
npu_per_node: 16
env_common:
VLLM_USE_MODELSCOPE: true
OMP_PROC_BIND: false
OMP_NUM_THREADS: 100
HCCL_BUFFSIZE: 1024
SERVER_PORT: 8080
NUMEXPR_MAX_THREADS: 128
deployment:
-
server_cmd: >
vllm serve "ZhipuAI/GLM-4.5"
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-size 4
--data-parallel-size-local 2
--data-parallel-address $LOCAL_IP
--data-parallel-rpc-port 13389
--tensor-parallel-size 8
--seed 1024
--enable-expert-parallel
--max-num-seqs 16
--max-model-len 8192
--max-num-batched-tokens 8192
--trust-remote-code
--no-enable-prefix-caching
--gpu-memory-utilization 0.9
-
server_cmd: >
vllm serve "ZhipuAI/GLM-4.5"
--headless
--data-parallel-size 4
--data-parallel-size-local 2
--data-parallel-start-rank 2
--data-parallel-address $MASTER_IP
--data-parallel-rpc-port 13389
--tensor-parallel-size 8
--seed 1024
--max-num-seqs 16
--max-model-len 8192
--max-num-batched-tokens 8192
--enable-expert-parallel
--trust-remote-code
--no-enable-prefix-caching
--gpu-memory-utilization 0.9
benchmarks:

4
tests/e2e/nightly/multi_node/config/multi_node_config.py
View File

@@ -31,8 +31,8 @@ class NodeInfo:
return (f"NodeInfo:\n"
f" index={self.index}\n"
f" ip={self.ip}\n"
f" server_port={self.server_port}\n"
f" headless={self.headless}")
f" headless={self.headless}\n"
f" server_port={self.server_port}")
class MultiNodeConfig:

2
tests/e2e/nightly/multi_node/scripts/lws.yaml.jinja2
View File

@@ -87,7 +87,7 @@ spec:
- name: VLLM_ASCEND_REMOTE_URL
value: {{ vllm_ascend_remote_url | default("https://github.com/vllm-project/vllm-ascend.git") }}
- name: RESULT_FILE_PATH
value: {{ result_file_path | default("/root/.cache/tests/ret/test_result.txt") }}
value: {{ result_file_path | default("/root/.cache/tests/ret") }}
- name: FAIL_TAG
value: {{ fail_tag | default("FAIL_TAG") }}
command:

9
tests/e2e/nightly/multi_node/scripts/run.sh
View File

@@ -127,19 +127,14 @@ kill_npu_processes() {
run_tests_with_log() {
set +e
kill_npu_processes
BASENAME=$(basename "$CONFIG_YAML_PATH" .yaml)
# each worker should have log file
LOG_FILE="${RESULT_FILE_PATH}/${BASENAME}_worker_${LWS_WORKER_INDEX}.log"
mkdir -p ${RESULT_FILE_PATH}
pytest -sv tests/e2e/nightly/multi_node/test_multi_node.py 2>&1 | tee $LOG_FILE
ret=${PIPESTATUS[0]}
pytest -sv tests/e2e/nightly/multi_node/test_multi_node.py
ret=$?
set -e
if [ "$LWS_WORKER_INDEX" -eq 0 ]; then
if [ $ret -eq 0 ]; then
print_success "All tests passed!"
else
print_failure "Some tests failed!"
mv LOG_FILE error_${LOG_FILE}
fi
fi
}