mirror of
https://github.com/vllm-project/vllm-ascend.git
synced 2025-10-20 13:43:53 +08:00
4c4a8458a5
### What this PR does / why we need it? Refactor the multi-machine CI use case. The purpose of this PR is to increase the ease of adding multi-machine CI use cases, allowing developers to add multi-machine cluster model testing use cases (including PD separation) by simply adding a new YAML configuration file. ### Does this PR introduce _any_ user-facing change? ### How was this patch tested? - vLLM version: v0.11.0rc3 - vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0 --------- Signed-off-by: wangli <wangli858794774@gmail.com>
31 lines
1.2 KiB
Python
31 lines
1.2 KiB
Python
from tests.e2e.conftest import RemoteOpenAIServer
|
|
from tests.e2e.nightly.multi_node.config.multi_node_config import (
|
|
DISAGGREGATED_PREFILL_PROXY_SCRIPT, MultiNodeConfig)
|
|
|
|
|
|
def test_multi_node() -> None:
|
|
config = MultiNodeConfig.from_yaml()
|
|
env_dict = config.envs
|
|
# perf_cmd = config.perf_cmd
|
|
# acc_cmd = config.acc_cmd
|
|
server_port = config.server_port if not config.disaggregated_prefill else config.proxy_port
|
|
server_host = config.cluster_ips[0]
|
|
with config.launch_server_proxy(DISAGGREGATED_PREFILL_PROXY_SCRIPT):
|
|
with RemoteOpenAIServer(
|
|
model=config.model,
|
|
vllm_serve_args=config.server_cmd,
|
|
server_port=server_port,
|
|
server_host=server_host,
|
|
env_dict=env_dict,
|
|
auto_port=False,
|
|
max_wait_seconds=2000,
|
|
) as remote_server:
|
|
# base_url = remote_server.url_root
|
|
if config.is_master:
|
|
pass
|
|
# TODO: enable perf and acc test
|
|
# subprocess.run(perf_cmd, check=True)
|
|
# subprocess.run(acc_cmd, check=True)
|
|
else:
|
|
remote_server.hang_until_terminated()
|