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

86
tests/models/test_initialization.py
View File

@@ -8,13 +8,19 @@ import pytest
from vllm import LLM
from vllm.utils import GiB_bytes
from vllm.v1.core.kv_cache_utils import (generate_scheduler_kv_cache_config,
get_kv_cache_configs)
from vllm.v1.core.kv_cache_utils import (
generate_scheduler_kv_cache_config,
get_kv_cache_configs,
)
from vllm.v1.engine.core import EngineCore as V1EngineCore
from ..utils import create_new_process_for_each_test
from .registry import (_TRANSFORMERS_BACKEND_MODELS, AUTO_EXAMPLE_MODELS,
HF_EXAMPLE_MODELS, HfExampleModels)
from .registry import (
_TRANSFORMERS_BACKEND_MODELS,
AUTO_EXAMPLE_MODELS,
HF_EXAMPLE_MODELS,
HfExampleModels,
)
from .utils import dummy_hf_overrides
# This minimal list of model architectures is smaller than the total list of
@@ -24,23 +30,32 @@ from .utils import dummy_hf_overrides
# generation, sequence classification, causal LM, ranking, chat, reward model,
# multimodal, geospatial, voice, embedding, MTP)
MINIMAL_MODEL_ARCH_LIST = [
"LlavaForConditionalGeneration", "Llama4ForConditionalGeneration",
"BertForSequenceClassification", "Gemma3nForCausalLM", "JinaVLForRanking",
"InternVLChatModel", "InternLM2ForRewardModel",
"TransformersForMultimodalLM", "PrithviGeoSpatialMAE", "UltravoxModel",
"DeepSeekMTPModel", "XLMRobertaModel"
"LlavaForConditionalGeneration",
"Llama4ForConditionalGeneration",
"BertForSequenceClassification",
"Gemma3nForCausalLM",
"JinaVLForRanking",
"InternVLChatModel",
"InternLM2ForRewardModel",
"TransformersForMultimodalLM",
"PrithviGeoSpatialMAE",
"UltravoxModel",
"DeepSeekMTPModel",
"XLMRobertaModel",
]
# This list is the complement of the minimal list above. The intention is that
# this list of models is only tested in a "special case" i.e. most PRs should
# not test these models
OTHER_MODEL_ARCH_LIST = (set(HF_EXAMPLE_MODELS.get_supported_archs()) -
set(MINIMAL_MODEL_ARCH_LIST))
OTHER_MODEL_ARCH_LIST = set(HF_EXAMPLE_MODELS.get_supported_archs()) - set(
MINIMAL_MODEL_ARCH_LIST
)
@create_new_process_for_each_test()
def can_initialize(model_arch: str, monkeypatch: pytest.MonkeyPatch,
EXAMPLE_MODELS: HfExampleModels):
def can_initialize(
model_arch: str, monkeypatch: pytest.MonkeyPatch, EXAMPLE_MODELS: HfExampleModels
):
"""The reason for using create_new_process_for_each_test is to avoid
the WARNING:
"We must use the 'spawn' multiprocessing start method. Overriding
@@ -53,12 +68,12 @@ def can_initialize(model_arch: str, monkeypatch: pytest.MonkeyPatch,
model_info.check_available_online(on_fail="skip")
model_info.check_transformers_version(on_fail="skip")
hf_overrides_fn = partial(dummy_hf_overrides,
model_arch=model_arch,
exist_overrides=model_info.hf_overrides,
use_original_num_layers=getattr(
model_info, 'use_original_num_layers',
False))
hf_overrides_fn = partial(
dummy_hf_overrides,
model_arch=model_arch,
exist_overrides=model_info.hf_overrides,
use_original_num_layers=getattr(model_info, "use_original_num_layers", False),
)
# Avoid calling model.forward()
def _initialize_kv_caches_v1(self, vllm_config):
@@ -68,14 +83,15 @@ def can_initialize(model_arch: str, monkeypatch: pytest.MonkeyPatch,
kv_cache_specs,
[10 * GiB_bytes],
)
scheduler_kv_cache_config = generate_scheduler_kv_cache_config(
kv_cache_configs)
scheduler_kv_cache_config = generate_scheduler_kv_cache_config(kv_cache_configs)
# gpu_blocks (> 0), cpu_blocks, scheduler_kv_cache_config
return 1, 0, scheduler_kv_cache_config
with (patch.object(V1EngineCore, "_initialize_kv_caches",
_initialize_kv_caches_v1), monkeypatch.context() as m):
with (
patch.object(V1EngineCore, "_initialize_kv_caches", _initialize_kv_caches_v1),
monkeypatch.context() as m,
):
if model_info.v0_only:
# NOTE(woosuk): skip the test for V0-only models
return
@@ -97,21 +113,24 @@ def can_initialize(model_arch: str, monkeypatch: pytest.MonkeyPatch,
speculative_config={
"model": model_info.speculative_model,
"num_speculative_tokens": 1,
} if model_info.speculative_model else None,
}
if model_info.speculative_model
else None,
trust_remote_code=model_info.trust_remote_code,
max_model_len=model_info.max_model_len,
# these tests seem to produce leftover memory
gpu_memory_utilization=0.80,
load_format="dummy",
model_impl="transformers"
if model_arch in _TRANSFORMERS_BACKEND_MODELS else "vllm",
if model_arch in _TRANSFORMERS_BACKEND_MODELS
else "vllm",
hf_overrides=hf_overrides_fn,
max_num_seqs=model_info.max_num_seqs)
max_num_seqs=model_info.max_num_seqs,
)
@pytest.mark.parametrize("model_arch", MINIMAL_MODEL_ARCH_LIST)
def test_can_initialize_small_subset(model_arch: str,
monkeypatch: pytest.MonkeyPatch):
def test_can_initialize_small_subset(model_arch: str, monkeypatch: pytest.MonkeyPatch):
"""Test initializing small subset of supported models"""
if model_arch == "Lfm2ForCausalLM":
pytest.skip("Skipping until test supports V1-only models")
@@ -119,10 +138,9 @@ def test_can_initialize_small_subset(model_arch: str,
@pytest.mark.parametrize("model_arch", OTHER_MODEL_ARCH_LIST)
def test_can_initialize_large_subset(model_arch: str,
monkeypatch: pytest.MonkeyPatch):
def test_can_initialize_large_subset(model_arch: str, monkeypatch: pytest.MonkeyPatch):
"""Test initializing large subset of supported models
This test covers the complement of the tests covered in the "small subset"
test.
"""
@@ -131,8 +149,6 @@ def test_can_initialize_large_subset(model_arch: str,
can_initialize(model_arch, monkeypatch, HF_EXAMPLE_MODELS)
@pytest.mark.parametrize("model_arch",
AUTO_EXAMPLE_MODELS.get_supported_archs())
def test_implicit_converted_models(model_arch: str,
monkeypatch: pytest.MonkeyPatch):
@pytest.mark.parametrize("model_arch", AUTO_EXAMPLE_MODELS.get_supported_archs())
def test_implicit_converted_models(model_arch: str, monkeypatch: pytest.MonkeyPatch):
can_initialize(model_arch, monkeypatch, AUTO_EXAMPLE_MODELS)