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

Convert formatting to use ruff instead of yapf + isort (#26247)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-10-05 15:06:22 +01:00
committed by GitHub
parent 17edd8a807
commit d6953beb91
1508 changed files with 115244 additions and 94146 deletions
Download Patch File Download Diff File Expand all files Collapse all files

170
tests/entrypoints/openai/test_lora_adapters.py
View File

@@ -20,26 +20,18 @@ MODEL_NAME = "HuggingFaceH4/zephyr-7b-beta"
BADREQUEST_CASES = [
(
"test_rank",
{
"r": 1024
},
{"r": 1024},
"is greater than max_lora_rank",
),
(
"test_bias",
{
"bias": "all"
},
{"bias": "all"},
"Adapter bias cannot be used without bias_enabled",
),
("test_dora", {
"use_dora": True
}, "does not yet support DoRA"),
("test_dora", {"use_dora": True}, "does not yet support DoRA"),
(
"test_modules_to_save",
{
"modules_to_save": ["lm_head"]
},
{"modules_to_save": ["lm_head"]},
"only supports modules_to_save being None",
),
]
@@ -48,24 +40,23 @@ BADREQUEST_CASES = [
@pytest.fixture(scope="module")
def monkeypatch_module():
from _pytest.monkeypatch import MonkeyPatch
mpatch = MonkeyPatch()
yield mpatch
mpatch.undo()
@pytest.fixture(scope="module", params=[True])
def server_with_lora_modules_json(request, monkeypatch_module,
zephyr_lora_files):
def server_with_lora_modules_json(request, monkeypatch_module, zephyr_lora_files):
use_v1 = request.param
assert use_v1
monkeypatch_module.setenv('VLLM_USE_V1', '1')
monkeypatch_module.setenv("VLLM_USE_V1", "1")
# Define the json format LoRA module configurations
lora_module_1 = {
"name": "zephyr-lora",
"path": zephyr_lora_files,
"base_model_name": MODEL_NAME
"base_model_name": MODEL_NAME,
}
args = [
@@ -96,14 +87,12 @@ def server_with_lora_modules_json(request, monkeypatch_module,
@pytest_asyncio.fixture
async def client(server_with_lora_modules_json):
async with server_with_lora_modules_json.get_async_client(
) as async_client:
async with server_with_lora_modules_json.get_async_client() as async_client:
yield async_client
@pytest.mark.asyncio
async def test_static_lora_lineage(client: openai.AsyncOpenAI,
zephyr_lora_files):
async def test_static_lora_lineage(client: openai.AsyncOpenAI, zephyr_lora_files):
models = await client.models.list()
models = models.data
served_model = models[0]
@@ -111,22 +100,18 @@ async def test_static_lora_lineage(client: openai.AsyncOpenAI,
assert served_model.id == MODEL_NAME
assert served_model.root == MODEL_NAME
assert served_model.parent is None
assert all(lora_model.root == zephyr_lora_files
for lora_model in lora_models)
assert all(lora_model.root == zephyr_lora_files for lora_model in lora_models)
assert all(lora_model.parent == MODEL_NAME for lora_model in lora_models)
assert lora_models[0].id == "zephyr-lora"
@pytest.mark.asyncio
async def test_dynamic_lora_lineage(client: openai.AsyncOpenAI,
zephyr_lora_files):
response = await client.post("load_lora_adapter",
cast_to=str,
body={
"lora_name": "zephyr-lora-3",
"lora_path": zephyr_lora_files
})
async def test_dynamic_lora_lineage(client: openai.AsyncOpenAI, zephyr_lora_files):
response = await client.post(
"load_lora_adapter",
cast_to=str,
body={"lora_name": "zephyr-lora-3", "lora_path": zephyr_lora_files},
)
# Ensure adapter loads before querying /models
assert "success" in response
@@ -141,37 +126,37 @@ async def test_dynamic_lora_lineage(client: openai.AsyncOpenAI,
@pytest.mark.asyncio
async def test_dynamic_lora_not_found(client: openai.AsyncOpenAI):
with pytest.raises(openai.NotFoundError):
await client.post("load_lora_adapter",
cast_to=str,
body={
"lora_name": "notfound",
"lora_path": "/not/an/adapter"
})
await client.post(
"load_lora_adapter",
cast_to=str,
body={"lora_name": "notfound", "lora_path": "/not/an/adapter"},
)
@pytest.mark.asyncio
async def test_dynamic_lora_invalid_files(client: openai.AsyncOpenAI,
tmp_path):
async def test_dynamic_lora_invalid_files(client: openai.AsyncOpenAI, tmp_path):
invalid_files = tmp_path / "invalid_files"
invalid_files.mkdir()
(invalid_files / "adapter_config.json").write_text("this is not json")
with pytest.raises(openai.BadRequestError):
await client.post("load_lora_adapter",
cast_to=str,
body={
"lora_name": "invalid-json",
"lora_path": str(invalid_files)
})
await client.post(
"load_lora_adapter",
cast_to=str,
body={"lora_name": "invalid-json", "lora_path": str(invalid_files)},
)
@pytest.mark.asyncio
@pytest.mark.parametrize("test_name,config_change,expected_error",
BADREQUEST_CASES)
async def test_dynamic_lora_badrequests(client: openai.AsyncOpenAI, tmp_path,
zephyr_lora_files, test_name: str,
config_change: dict,
expected_error: str):
@pytest.mark.parametrize("test_name,config_change,expected_error", BADREQUEST_CASES)
async def test_dynamic_lora_badrequests(
client: openai.AsyncOpenAI,
tmp_path,
zephyr_lora_files,
test_name: str,
config_change: dict,
expected_error: str,
):
# Create test directory
test_dir = tmp_path / test_name
@@ -191,29 +176,28 @@ async def test_dynamic_lora_badrequests(client: openai.AsyncOpenAI, tmp_path,
# Test loading the adapter
with pytest.raises(openai.BadRequestError, match=expected_error):
await client.post("load_lora_adapter",
cast_to=str,
body={
"lora_name": test_name,
"lora_path": str(test_dir)
})
await client.post(
"load_lora_adapter",
cast_to=str,
body={"lora_name": test_name, "lora_path": str(test_dir)},
)
@pytest.mark.asyncio
async def test_multiple_lora_adapters(client: openai.AsyncOpenAI, tmp_path,
zephyr_lora_files):
async def test_multiple_lora_adapters(
client: openai.AsyncOpenAI, tmp_path, zephyr_lora_files
):
"""Validate that many loras can be dynamically registered and inferenced
with concurrently"""
# This test file configures the server with --max-cpu-loras=2 and this test
# will concurrently load 10 adapters, so it should flex the LRU cache
async def load_and_run_adapter(adapter_name: str):
await client.post("load_lora_adapter",
cast_to=str,
body={
"lora_name": adapter_name,
"lora_path": str(zephyr_lora_files)
})
await client.post(
"load_lora_adapter",
cast_to=str,
body={"lora_name": adapter_name, "lora_path": str(zephyr_lora_files)},
)
for _ in range(3):
await client.completions.create(
model=adapter_name,
@@ -223,8 +207,7 @@ async def test_multiple_lora_adapters(client: openai.AsyncOpenAI, tmp_path,
lora_tasks = []
for i in range(10):
lora_tasks.append(
asyncio.create_task(load_and_run_adapter(f"adapter_{i}")))
lora_tasks.append(asyncio.create_task(load_and_run_adapter(f"adapter_{i}")))
results, _ = await asyncio.wait(lora_tasks)
@@ -234,8 +217,8 @@ async def test_multiple_lora_adapters(client: openai.AsyncOpenAI, tmp_path,
@pytest.mark.asyncio
async def test_loading_invalid_adapters_does_not_break_others(
client: openai.AsyncOpenAI, tmp_path, zephyr_lora_files):
client: openai.AsyncOpenAI, tmp_path, zephyr_lora_files
):
invalid_files = tmp_path / "invalid_files"
invalid_files.mkdir()
(invalid_files / "adapter_config.json").write_text("this is not json")
@@ -266,20 +249,18 @@ async def test_loading_invalid_adapters_does_not_break_others(
# Run a bunch of bad adapter loads
for _ in range(25):
with suppress(openai.NotFoundError):
await client.post("load_lora_adapter",
cast_to=str,
body={
"lora_name": "notfound",
"lora_path": "/not/an/adapter"
})
await client.post(
"load_lora_adapter",
cast_to=str,
body={"lora_name": "notfound", "lora_path": "/not/an/adapter"},
)
for _ in range(25):
with suppress(openai.BadRequestError):
await client.post("load_lora_adapter",
cast_to=str,
body={
"lora_name": "invalid",
"lora_path": str(invalid_files)
})
await client.post(
"load_lora_adapter",
cast_to=str,
body={"lora_name": "invalid", "lora_path": str(invalid_files)},
)
# Ensure all the running requests with lora adapters succeeded
stop_good_requests_event.set()
@@ -288,12 +269,11 @@ async def test_loading_invalid_adapters_does_not_break_others(
assert not isinstance(r, Exception), f"Got exception {r}"
# Ensure we can load another adapter and run it
await client.post("load_lora_adapter",
cast_to=str,
body={
"lora_name": "valid",
"lora_path": zephyr_lora_files
})
await client.post(
"load_lora_adapter",
cast_to=str,
body={"lora_name": "valid", "lora_path": zephyr_lora_files},
)
await client.completions.create(
model="valid",
prompt=["Hello there", "Foo bar bazz buzz"],
@@ -310,12 +290,11 @@ async def test_beam_search_with_lora_adapters(
"""Validate that async beam search can be used with lora."""
async def load_and_run_adapter(adapter_name: str):
await client.post("load_lora_adapter",
cast_to=str,
body={
"lora_name": adapter_name,
"lora_path": str(zephyr_lora_files)
})
await client.post(
"load_lora_adapter",
cast_to=str,
body={"lora_name": adapter_name, "lora_path": str(zephyr_lora_files)},
)
for _ in range(3):
await client.completions.create(
model=adapter_name,
@@ -326,8 +305,7 @@ async def test_beam_search_with_lora_adapters(
lora_tasks = []
for i in range(3):
lora_tasks.append(
asyncio.create_task(load_and_run_adapter(f"adapter_{i}")))
lora_tasks.append(asyncio.create_task(load_and_run_adapter(f"adapter_{i}")))
results, _ = await asyncio.wait(lora_tasks)