[Core] Add an environment variable which needs to be set explicitly to allow BlockSpaceManagerV1 (#9149)
This commit is contained in:
@@ -12,7 +12,7 @@ from contextlib import nullcontext
|
||||
import pytest
|
||||
|
||||
from ..models.utils import check_logprobs_close, check_outputs_equal
|
||||
from ..utils import multi_gpu_test
|
||||
from ..utils import check_deprecated_block_manager_usage, multi_gpu_test
|
||||
|
||||
MODELS = [
|
||||
"facebook/opt-125m",
|
||||
@@ -20,6 +20,12 @@ MODELS = [
|
||||
]
|
||||
|
||||
|
||||
@pytest.fixture(scope="module", autouse=True)
|
||||
def check_deprecated_block_manager():
|
||||
check_deprecated_block_manager_usage(
|
||||
'tests/basic_correctness/test_chunked_prefill.py')
|
||||
|
||||
|
||||
@pytest.mark.parametrize("model", MODELS)
|
||||
@pytest.mark.parametrize("dtype", ["half"])
|
||||
@pytest.mark.parametrize("max_tokens", [32])
|
||||
|
||||
@@ -2,11 +2,18 @@ from itertools import cycle
|
||||
|
||||
import pytest
|
||||
|
||||
from tests.utils import check_deprecated_block_manager_usage
|
||||
from vllm import SamplingParams
|
||||
|
||||
from .conftest import get_token_ids_from_llm_generator
|
||||
|
||||
|
||||
@pytest.fixture(scope="module", autouse=True)
|
||||
def check_deprecated_block_manager():
|
||||
check_deprecated_block_manager_usage(
|
||||
'tests/core/block/e2e/test_correctness.py')
|
||||
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
"common_llm_kwargs",
|
||||
[{
|
||||
|
||||
@@ -3,6 +3,7 @@ from typing import List
|
||||
|
||||
import pytest
|
||||
|
||||
from tests.utils import check_deprecated_block_manager_usage
|
||||
from vllm import LLM, SamplingParams
|
||||
|
||||
from .conftest import get_text_from_llm_generator
|
||||
@@ -12,6 +13,12 @@ MODEL = "bigcode/starcoder2-3b"
|
||||
BLOCK_SIZE = 16
|
||||
|
||||
|
||||
@pytest.fixture(scope="module", autouse=True)
|
||||
def check_deprecated_block_manager():
|
||||
check_deprecated_block_manager_usage(
|
||||
'tests/core/block/e2e/test_correctness_sliding_window.py')
|
||||
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
"common_llm_kwargs",
|
||||
[{
|
||||
|
||||
@@ -8,6 +8,7 @@ from vllm.core.interfaces import AllocStatus
|
||||
from vllm.core.scheduler import Scheduler
|
||||
from vllm.sequence import Logprob, SequenceGroup
|
||||
|
||||
from ..utils import check_deprecated_block_manager_usage
|
||||
from .utils import create_dummy_prompt
|
||||
|
||||
|
||||
@@ -27,6 +28,12 @@ def schedule_and_update_computed_tokens(scheduler):
|
||||
return metas, out
|
||||
|
||||
|
||||
@pytest.fixture(scope="module", autouse=True)
|
||||
def check_deprecated_block_manager():
|
||||
check_deprecated_block_manager_usage(
|
||||
'tests/core/test_chunked_prefill_scheduler.py')
|
||||
|
||||
|
||||
@pytest.mark.parametrize('use_v2_block_manager', [True, False])
|
||||
def test_simple(use_v2_block_manager: bool):
|
||||
"""Verify basic scheduling works."""
|
||||
|
||||
@@ -12,11 +12,18 @@ from vllm.core.scheduler import Scheduler, SchedulingBudget
|
||||
from vllm.lora.request import LoRARequest
|
||||
from vllm.sequence import SequenceGroup, SequenceStatus
|
||||
|
||||
from ..utils import check_deprecated_block_manager_usage
|
||||
from .utils import (append_new_token, append_new_token_seq_group,
|
||||
create_dummy_prompt, get_sequence_groups,
|
||||
schedule_and_update_computed_tokens)
|
||||
|
||||
|
||||
@pytest.fixture(scope="module", autouse=True)
|
||||
def check_deprecated_block_manager():
|
||||
check_deprecated_block_manager_usage(
|
||||
"tests/core/test_chunked_prefill_scheduler.py")
|
||||
|
||||
|
||||
@pytest.mark.parametrize('use_v2_block_manager', [True, False])
|
||||
def test_scheduler_add_seq_group(use_v2_block_manager: bool):
|
||||
block_size = 4
|
||||
|
||||
@@ -7,6 +7,7 @@ from typing import List
|
||||
import pytest
|
||||
|
||||
from tests.kernels.utils import override_backend_env_variable
|
||||
from tests.utils import check_deprecated_block_manager_usage
|
||||
from vllm.block import PhysicalTokenBlock
|
||||
from vllm.core.block_manager_v1 import CachedBlockAllocator
|
||||
from vllm.utils import Device
|
||||
@@ -18,6 +19,12 @@ MODELS = [
|
||||
]
|
||||
|
||||
|
||||
@pytest.fixture(scope="module", autouse=True)
|
||||
def check_deprecated_block_manager():
|
||||
check_deprecated_block_manager_usage(
|
||||
'tests/prefix_caching/test_prefix_caching.py')
|
||||
|
||||
|
||||
@pytest.mark.parametrize("block_size", [16])
|
||||
@pytest.mark.parametrize("num_blocks", [16])
|
||||
def test_block_allocator(
|
||||
|
||||
@@ -1,10 +1,17 @@
|
||||
import pytest
|
||||
|
||||
from tests.utils import check_deprecated_block_manager_usage
|
||||
from vllm import SamplingParams
|
||||
|
||||
from .conftest import get_output_from_llm_generator
|
||||
|
||||
|
||||
@pytest.fixture(scope="module", autouse=True)
|
||||
def check_deprecated_block_manager():
|
||||
check_deprecated_block_manager_usage(
|
||||
'tests/spec_decode/e2e/test_compatibility.py')
|
||||
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
"common_llm_kwargs",
|
||||
[{
|
||||
|
||||
@@ -678,3 +678,12 @@ def get_client_text_logprob_generations(
|
||||
return [(text_generations, text,
|
||||
(None if x.logprobs is None else x.logprobs.top_logprobs))
|
||||
for completion in completions for x in completion.choices]
|
||||
|
||||
|
||||
def check_deprecated_block_manager_usage(test_name: str):
|
||||
assert envs.VLLM_ALLOW_DEPRECATED_BLOCK_MANAGER_V1 is True, (
|
||||
f"To allow the use of deprecated BlockSpaceManagerV1, set the "
|
||||
f"environment variable VLLM_ALLOW_DEPRECATED_BLOCK_MANAGER_V1=1. "
|
||||
f"You can run the tests with: "
|
||||
f"`VLLM_ALLOW_DEPRECATED_BLOCK_MANAGER_V1=1 pytest {test_name}`" #noqa
|
||||
)
|
||||
|
||||
Reference in New Issue
Block a user