[V1][Core] Generic mechanism for handling engine utility (#13060)

Signed-off-by: Nick Hill <nhill@redhat.com>
This commit is contained in:
Nick Hill
2025-02-19 17:09:22 +08:00
committed by GitHub
parent f525c0be8b
commit caf7ff4456
5 changed files with 197 additions and 56 deletions
+1 -1
View File
@@ -41,7 +41,7 @@ def download_and_prepare_lora_module():
]
for tokenizer_file in tokenizer_files:
del_path = Path(LORA_MODULE_DOWNLOAD_PATH) / tokenizer_file
del_path.unlink()
del_path.unlink(missing_ok=True)
@pytest.fixture(autouse=True)
+49 -8
View File
@@ -3,7 +3,8 @@
import asyncio
import time
import uuid
from typing import Dict, List
from contextlib import ExitStack
from typing import Dict, List, Optional
import pytest
from transformers import AutoTokenizer
@@ -14,7 +15,9 @@ from vllm.engine.arg_utils import EngineArgs
from vllm.platforms import current_platform
from vllm.usage.usage_lib import UsageContext
from vllm.v1.engine import EngineCoreRequest
from vllm.v1.engine.core_client import EngineCoreClient
from vllm.v1.engine.core import EngineCore
from vllm.v1.engine.core_client import (AsyncMPClient, EngineCoreClient,
SyncMPClient)
from vllm.v1.executor.abstract import Executor
if not current_platform.is_cuda():
@@ -63,7 +66,7 @@ def loop_until_done(client: EngineCoreClient, outputs: Dict):
async def loop_until_done_async(client: EngineCoreClient, outputs: Dict):
while True:
engine_core_outputs = await client.get_output_async().outputs
engine_core_outputs = (await client.get_output_async()).outputs
if len(engine_core_outputs) == 0:
break
@@ -78,6 +81,14 @@ async def loop_until_done_async(client: EngineCoreClient, outputs: Dict):
break
# Dummy utility function to monkey-patch into engine core.
def echo(self, msg: str, err_msg: Optional[str] = None) -> str:
print(f"echo util function called: {msg}, {err_msg}")
if err_msg is not None:
raise ValueError(err_msg)
return msg
@fork_new_process_for_each_test
@pytest.mark.parametrize("multiprocessing_mode", [True, False])
def test_engine_core_client(monkeypatch, multiprocessing_mode: bool):
@@ -85,7 +96,10 @@ def test_engine_core_client(monkeypatch, multiprocessing_mode: bool):
with monkeypatch.context() as m:
m.setenv("VLLM_USE_V1", "1")
engine_args = EngineArgs(model=MODEL_NAME, compilation_config=3)
# Monkey-patch core engine utility function to test.
m.setattr(EngineCore, "echo", echo, raising=False)
engine_args = EngineArgs(model=MODEL_NAME, enforce_eager=True)
vllm_config = engine_args.create_engine_config(
UsageContext.UNKNOWN_CONTEXT)
executor_class = Executor.get_class(vllm_config)
@@ -147,15 +161,30 @@ def test_engine_core_client(monkeypatch, multiprocessing_mode: bool):
client.abort_requests([request.request_id])
if multiprocessing_mode:
"""Utility method invocation"""
@fork_new_process_for_each_test
@pytest.mark.asyncio
core_client: SyncMPClient = client
result = core_client._call_utility("echo", "testarg")
assert result == "testarg"
with pytest.raises(Exception) as e_info:
core_client._call_utility("echo", None, "help!")
assert str(e_info.value) == "Call to echo method failed: help!"
@pytest.mark.asyncio(loop_scope="function")
async def test_engine_core_client_asyncio(monkeypatch):
with monkeypatch.context() as m:
with monkeypatch.context() as m, ExitStack() as after:
m.setenv("VLLM_USE_V1", "1")
engine_args = EngineArgs(model=MODEL_NAME)
# Monkey-patch core engine utility function to test.
m.setattr(EngineCore, "echo", echo, raising=False)
engine_args = EngineArgs(model=MODEL_NAME, enforce_eager=True)
vllm_config = engine_args.create_engine_config(
usage_context=UsageContext.UNKNOWN_CONTEXT)
executor_class = Executor.get_class(vllm_config)
@@ -166,6 +195,7 @@ async def test_engine_core_client_asyncio(monkeypatch):
executor_class=executor_class,
log_stats=True,
)
after.callback(client.shutdown)
MAX_TOKENS = 20
params = SamplingParams(max_tokens=MAX_TOKENS)
@@ -204,3 +234,14 @@ async def test_engine_core_client_asyncio(monkeypatch):
else:
assert len(outputs[req_id]) == MAX_TOKENS, (
f"{len(outputs[req_id])=}, {MAX_TOKENS=}")
"""Utility method invocation"""
core_client: AsyncMPClient = client
result = await core_client._call_utility_async("echo", "testarg")
assert result == "testarg"
with pytest.raises(Exception) as e_info:
await core_client._call_utility_async("echo", None, "help!")
assert str(e_info.value) == "Call to echo method failed: help!"