[Bugfix] Register VLLM_BATCH_INVARIANT in envs.py to fix spurious unknown env var warning (#35007)
Signed-off-by: Ranran <1012869439@qq.com> Signed-off-by: Ranran <hzz5361@psu.edu> Signed-off-by: ran <hzz5361@psu.edu> Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
This commit is contained in:
@@ -55,37 +55,37 @@ def _clear_supports_cache():
|
||||
# supports_trtllm_attention
|
||||
|
||||
|
||||
@patch("vllm.utils.flashinfer.vllm_is_batch_invariant", return_value=True)
|
||||
def test_supports_batch_invariant_disables(_mock):
|
||||
@patch("vllm.envs.VLLM_BATCH_INVARIANT", True)
|
||||
def test_supports_batch_invariant_disables():
|
||||
assert supports_trtllm_attention() is False
|
||||
|
||||
|
||||
@patch("vllm.utils.flashinfer.vllm_is_batch_invariant", return_value=False)
|
||||
@patch("vllm.envs.VLLM_BATCH_INVARIANT", False)
|
||||
@patch(
|
||||
"vllm.utils.flashinfer.current_platform.is_device_capability_family",
|
||||
return_value=True,
|
||||
)
|
||||
@patch("vllm.utils.flashinfer.has_nvidia_artifactory", return_value=True)
|
||||
def test_supports_sm100_with_artifactory(_art, _cap, _bi):
|
||||
def test_supports_sm100_with_artifactory(_art, _cap):
|
||||
assert supports_trtllm_attention() is True
|
||||
|
||||
|
||||
@patch("vllm.utils.flashinfer.vllm_is_batch_invariant", return_value=False)
|
||||
@patch("vllm.envs.VLLM_BATCH_INVARIANT", False)
|
||||
@patch(
|
||||
"vllm.utils.flashinfer.current_platform.is_device_capability_family",
|
||||
return_value=False,
|
||||
)
|
||||
def test_supports_non_sm100_platform(_cap, _bi):
|
||||
def test_supports_non_sm100_platform(_cap):
|
||||
assert supports_trtllm_attention() is False
|
||||
|
||||
|
||||
@patch("vllm.utils.flashinfer.vllm_is_batch_invariant", return_value=False)
|
||||
@patch("vllm.envs.VLLM_BATCH_INVARIANT", False)
|
||||
@patch(
|
||||
"vllm.utils.flashinfer.current_platform.is_device_capability_family",
|
||||
return_value=True,
|
||||
)
|
||||
@patch("vllm.utils.flashinfer.has_nvidia_artifactory", return_value=False)
|
||||
def test_supports_sm100_without_artifactory(_art, _cap, _bi):
|
||||
def test_supports_sm100_without_artifactory(_art, _cap):
|
||||
assert supports_trtllm_attention() is False
|
||||
|
||||
|
||||
|
||||
@@ -8,7 +8,7 @@ Run `pytest tests/kernels/moe/test_grouped_topk.py`.
|
||||
import pytest
|
||||
import torch
|
||||
|
||||
import vllm.model_executor.layers.batch_invariant as batch_invariant
|
||||
import vllm.envs as envs
|
||||
from vllm.config import (
|
||||
CompilationConfig,
|
||||
VllmConfig,
|
||||
@@ -69,7 +69,7 @@ def test_grouped_topk(
|
||||
|
||||
with set_current_vllm_config(vllm_config), monkeypatch.context() as m:
|
||||
m.setenv("VLLM_USE_FUSED_MOE_GROUPED_TOPK", "0")
|
||||
m.setattr(batch_invariant, "VLLM_BATCH_INVARIANT", True)
|
||||
m.setattr(envs, "VLLM_BATCH_INVARIANT", True)
|
||||
grouped_topk = GroupedTopk(
|
||||
topk=topk,
|
||||
renormalize=renormalize,
|
||||
|
||||
@@ -2,11 +2,11 @@
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||
import pytest
|
||||
|
||||
import vllm.model_executor.layers.batch_invariant as batch_invariant
|
||||
import vllm.envs as envs
|
||||
|
||||
|
||||
@pytest.fixture(autouse=True)
|
||||
def enable_batch_invariant_mode(monkeypatch: pytest.MonkeyPatch):
|
||||
"""Automatically enable batch invariant kernel overrides for all tests."""
|
||||
monkeypatch.setattr(batch_invariant, "VLLM_BATCH_INVARIANT", True)
|
||||
monkeypatch.setattr(envs, "VLLM_BATCH_INVARIANT", True)
|
||||
monkeypatch.setenv("VLLM_BATCH_INVARIANT", "1")
|
||||
|
||||
@@ -15,7 +15,7 @@ from utils import (
|
||||
skip_unsupported,
|
||||
)
|
||||
|
||||
import vllm.model_executor.layers.batch_invariant as batch_invariant
|
||||
import vllm.envs as envs
|
||||
from vllm import LLM, SamplingParams
|
||||
|
||||
IS_DEVICE_CAPABILITY_BELOW_90 = is_device_capability_below_90()
|
||||
@@ -173,11 +173,9 @@ def test_logprobs_bitwise_batch_invariance_bs1_vs_bsN(
|
||||
|
||||
# For batch invariance, disable custom all-reduce to ensure deterministic
|
||||
# all-reduce operations (custom all-reduce may not be deterministic)
|
||||
from vllm.model_executor.layers.batch_invariant import (
|
||||
vllm_is_batch_invariant,
|
||||
)
|
||||
import vllm.envs as envs
|
||||
|
||||
disable_custom_ar = vllm_is_batch_invariant()
|
||||
disable_custom_ar = envs.VLLM_BATCH_INVARIANT
|
||||
|
||||
if disable_custom_ar:
|
||||
print(f"\n{'=' * 80}")
|
||||
@@ -454,7 +452,7 @@ def test_logprobs_without_batch_invariance_should_fail(
|
||||
"""
|
||||
# CRITICAL: Disable batch invariance for this test
|
||||
monkeypatch.setenv("VLLM_BATCH_INVARIANT", "0")
|
||||
monkeypatch.setattr(batch_invariant, "VLLM_BATCH_INVARIANT", False)
|
||||
monkeypatch.setattr(envs, "VLLM_BATCH_INVARIANT", False)
|
||||
seed = int(os.getenv("VLLM_TEST_SEED", "12345"))
|
||||
random.seed(seed)
|
||||
tp_size = int(os.getenv("VLLM_TEST_TP_SIZE", "1"))
|
||||
@@ -674,11 +672,9 @@ def test_decode_logprobs_match_prefill_logprobs(
|
||||
random.seed(seed)
|
||||
tp_size = int(os.getenv("VLLM_TEST_TP_SIZE", "1"))
|
||||
|
||||
from vllm.model_executor.layers.batch_invariant import (
|
||||
vllm_is_batch_invariant,
|
||||
)
|
||||
import vllm.envs as envs
|
||||
|
||||
disable_custom_ar = vllm_is_batch_invariant()
|
||||
disable_custom_ar = envs.VLLM_BATCH_INVARIANT
|
||||
|
||||
if disable_custom_ar:
|
||||
print(f"\n{'=' * 80}")
|
||||
|
||||
Reference in New Issue
Block a user