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

[Mypy] Better fixes for the mypy issues in vllm/config (#37902)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2026-03-25 13:14:43 +00:00
committed by GitHub
parent 34d317dcec
commit d215d1efca
35 changed files with 153 additions and 182 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
tests/compile/test_config.py
View File

@@ -414,9 +414,12 @@ def test_cudagraph_sizes_post_init(
ctx,
patch("vllm.config.parallel.cuda_device_count_stateless", return_value=tp_size),
):
kwargs = {}
if cudagraph_capture_sizes is not None:
kwargs["cudagraph_capture_sizes"] = cudagraph_capture_sizes
if max_cudagraph_capture_size is not None:
kwargs["max_cudagraph_capture_size"] = max_cudagraph_capture_size
compilation_config = CompilationConfig(
cudagraph_capture_sizes=cudagraph_capture_sizes,
max_cudagraph_capture_size=max_cudagraph_capture_size,
pass_config=PassConfig(
enable_sp=enable_sp,
fuse_norm_quant=True,
@@ -425,6 +428,7 @@ def test_cudagraph_sizes_post_init(
sp_min_token_num=512 if enable_sp else None,
),
cudagraph_mode=cudagraph_mode,
**kwargs,
)
engine_args = EngineArgs(
model="facebook/opt-125m",

5
tests/entrypoints/offline_mode/test_offline_mode.py
View File

@@ -2,7 +2,6 @@
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
"""Tests for HF_HUB_OFFLINE mode"""
import dataclasses
import importlib
import sys
@@ -12,7 +11,6 @@ import urllib3
from vllm import LLM
from vllm.distributed import cleanup_dist_env_and_memory
from vllm.engine.arg_utils import EngineArgs
MODEL_CONFIGS = [
{
@@ -160,8 +158,7 @@ def test_model_from_huggingface_offline(monkeypatch: pytest.MonkeyPatch):
# Need to re-import huggingface_hub
# and friends to set up offline mode
_re_import_modules()
engine_args = EngineArgs(model="facebook/opt-125m")
LLM(**dataclasses.asdict(engine_args))
LLM(model="facebook/opt-125m")
finally:
# Reset the environment after the test
# NB: Assuming tests are run in online mode

19
tests/models/multimodal/generation/test_keye.py
View File

@@ -1,6 +1,5 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from dataclasses import asdict
from typing import NamedTuple
import pytest
@@ -29,14 +28,6 @@ def test_keye_vl(image_assets, question: str):
images = [asset.pil_image for asset in image_assets]
image_urls = [encode_image_url(image) for image in images]
engine_args = EngineArgs(
model=MODEL_NAME,
trust_remote_code=True,
max_model_len=8192,
max_num_seqs=5,
limit_mm_per_prompt={"image": len(image_urls)},
)
placeholders = [{"type": "image", "image": url} for url in image_urls]
messages = [
{
@@ -54,8 +45,14 @@ def test_keye_vl(image_assets, question: str):
messages, tokenize=False, add_generation_prompt=True
)
engine_args = asdict(engine_args) | {"seed": 42}
llm = LLM(**engine_args)
llm = LLM(
model=MODEL_NAME,
trust_remote_code=True,
max_model_len=8192,
max_num_seqs=5,
limit_mm_per_prompt={"image": len(image_urls)},
seed=42,
)
sampling_params = SamplingParams(
temperature=0.0, max_tokens=256, stop_token_ids=None

15
tests/models/multimodal/generation/test_vit_backend_functionality.py
View File

@@ -7,13 +7,12 @@ This test validates that each multimodal model can successfully generate outputs
using different ViT attention backends. Tests are parametrized by model and backend.
"""
from dataclasses import asdict
from typing import Any
import pytest
from transformers import AutoProcessor
from vllm import LLM, EngineArgs, SamplingParams
from vllm import LLM, SamplingParams
from vllm.multimodal.utils import encode_image_url
from vllm.multimodal.video import sample_frames_from_video
from vllm.platforms import current_platform
@@ -274,7 +273,7 @@ def run_llm_generate_test(config, mm_encoder_attn_backend, image_assets):
limit_mm_per_prompt = config.get("limit_mm_per_prompt", {"image": len(images)})
# Create engine
engine_args = EngineArgs(
llm = LLM(
model=config["model_name"],
trust_remote_code=True,
max_model_len=config["max_model_len"],
@@ -283,11 +282,9 @@ def run_llm_generate_test(config, mm_encoder_attn_backend, image_assets):
mm_encoder_attn_backend=mm_encoder_attn_backend,
hf_overrides=dummy_hf_overrides,
load_format="dummy",
seed=42,
)
engine_dict = asdict(engine_args) | {"seed": 42}
llm = LLM(**engine_dict)
# Generate
sampling_params = SamplingParams(**config["sampling_params"])
outputs = llm.generate(
@@ -318,7 +315,7 @@ def run_llm_chat_test(config, mm_encoder_attn_backend, image_assets):
messages = build_dots_ocr_prompt([stop_sign_image], config)
# Create engine
engine_args = EngineArgs(
llm = LLM(
model=config["model_name"],
trust_remote_code=True,
max_model_len=config["max_model_len"],
@@ -327,11 +324,9 @@ def run_llm_chat_test(config, mm_encoder_attn_backend, image_assets):
mm_encoder_attn_backend=mm_encoder_attn_backend,
hf_overrides=dummy_hf_overrides,
load_format="dummy",
seed=42,
)
engine_dict = asdict(engine_args) | {"seed": 42}
llm = LLM(**engine_dict)
# Generate using chat
sampling_params = SamplingParams(**config["sampling_params"])
outputs = llm.chat(messages=messages, sampling_params=sampling_params)

3
tests/models/multimodal/generation/test_voxtral_realtime.py
View File

@@ -1,7 +1,6 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import contextlib
from dataclasses import asdict
import pytest
import pytest_asyncio
@@ -75,7 +74,7 @@ def tokenizer() -> MistralTokenizer:
@pytest.fixture
def engine():
engine_args = EngineArgs(**ENGINE_CONFIG)
llm = LLM(**asdict(engine_args))
llm = LLM.from_engine_args(engine_args)
try:
yield llm
finally:

37
tests/v1/kv_connector/unit/test_example_connector.py
View File

@@ -1,12 +1,11 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from dataclasses import asdict
from typing import NamedTuple
import pytest
from PIL import Image
from vllm import LLM, EngineArgs, SamplingParams
from vllm import LLM, SamplingParams
from vllm.assets.image import ImageAsset
from vllm.config import AttentionConfig, KVTransferConfig
from vllm.multimodal.utils import encode_image_url
@@ -129,24 +128,6 @@ def test_shared_storage_connector_hashes(tmp_path, attn_backend):
# Using tmp_path as the storage path to store KV
print(f"KV storage path at: {str(tmp_path)}")
# Configure the ExampleConnector
kv_transfer_config = KVTransferConfig(
kv_connector="ExampleConnector",
kv_role="kv_both",
kv_connector_extra_config={"shared_storage_path": str(tmp_path)},
)
engine_args = EngineArgs(
model=MODEL_NAME,
max_model_len=8192,
max_num_seqs=1,
gpu_memory_utilization=0.4,
attention_config=AttentionConfig(backend=attn_backend),
enforce_eager=True,
kv_transfer_config=kv_transfer_config,
limit_mm_per_prompt={"image": 2},
)
# don't put this import at the top level
# it will call torch.accelerator.device_count()
from transformers import AutoProcessor
@@ -163,8 +144,20 @@ def test_shared_storage_connector_hashes(tmp_path, attn_backend):
assert image_1 != image_2, "The images should not be identical"
# Create the LLM instance
engine_args = asdict(engine_args)
llm = LLM(**engine_args)
llm = LLM(
model=MODEL_NAME,
max_model_len=8192,
max_num_seqs=1,
gpu_memory_utilization=0.4,
attention_config=AttentionConfig(backend=attn_backend),
enforce_eager=True,
kv_transfer_config=KVTransferConfig(
kv_connector="ExampleConnector",
kv_role="kv_both",
kv_connector_extra_config={"shared_storage_path": str(tmp_path)},
),
limit_mm_per_prompt={"image": 2},
)
# Prepare the input cases
input_cases = [