biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[misc] remove engine_use_ray (#8126)

Browse Source
This commit is contained in:
youkaichao
2024-09-11 18:23:36 -07:00
committed by GitHub
parent a65cb16067
commit f842a7aff1
8 changed files with 32 additions and 197 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
tests/async_engine/test_api_server.py
View File

@@ -1,4 +1,3 @@
import os
import subprocess
import sys
import time
@@ -26,8 +25,7 @@ def _query_server_long(prompt: str) -> dict:
@pytest.fixture
def api_server(tokenizer_pool_size: int, engine_use_ray: bool,
worker_use_ray: bool):
def api_server(tokenizer_pool_size: int, worker_use_ray: bool):
script_path = Path(__file__).parent.joinpath(
"api_server_async_engine.py").absolute()
commands = [
@@ -37,25 +35,17 @@ def api_server(tokenizer_pool_size: int, engine_use_ray: bool,
str(tokenizer_pool_size)
]
# Copy the environment variables and append `VLLM_ALLOW_ENGINE_USE_RAY=1`
# to prevent `--engine-use-ray` raises an exception due to it deprecation
env_vars = os.environ.copy()
env_vars["VLLM_ALLOW_ENGINE_USE_RAY"] = "1"
if engine_use_ray:
commands.append("--engine-use-ray")
if worker_use_ray:
commands.append("--worker-use-ray")
uvicorn_process = subprocess.Popen(commands, env=env_vars)
uvicorn_process = subprocess.Popen(commands)
yield
uvicorn_process.terminate()
@pytest.mark.parametrize("tokenizer_pool_size", [0, 2])
@pytest.mark.parametrize("worker_use_ray", [False, True])
@pytest.mark.parametrize("engine_use_ray", [False, True])
def test_api_server(api_server, tokenizer_pool_size: int, worker_use_ray: bool,
engine_use_ray: bool):
def test_api_server(api_server, tokenizer_pool_size: int,
worker_use_ray: bool):
"""
Run the API server and test it.