[CI] fix compile test | refactor VLLM_DISABLE_COMPILE_CACHE for tests (#49770)

Signed-off-by: Divakar Verma <divakar.verma@amd.com>
This commit is contained in:
Divakar Verma
2026-07-24 23:15:19 -05:00
committed by GitHub
parent d9cd774198
commit aaaeda98dc
4 changed files with 23 additions and 14 deletions
+1 -4
View File
@@ -290,12 +290,9 @@ def test_attention_quant_pattern(
model_class: type[AttentionQuantPatternModel],
backend: AttentionBackendEnum,
dist_init,
monkeypatch,
use_fresh_inductor_cache,
disable_vllm_compile_cache,
):
"""Test AttentionStaticQuantPattern fusion pass"""
monkeypatch.setenv("VLLM_DISABLE_COMPILE_CACHE", "1")
if backend == AttentionBackendEnum.FLASHINFER and (
not current_platform.is_device_capability((10, 0)) or not has_flashinfer()
):
@@ -419,8 +419,7 @@ def test_mla_attention_quant_pattern(
model_class: type[MLAAttentionQuantPatternModel],
backend: AttentionBackendEnum,
dist_init,
monkeypatch,
use_fresh_inductor_cache,
disable_vllm_compile_cache,
):
"""Test MLA AttentionQuantPattern fusion pass"""
if (
@@ -429,8 +428,6 @@ def test_mla_attention_quant_pattern(
):
pytest.skip("NVFP4 is not supported on this GPU (requires SM 100+).")
monkeypatch.setenv("VLLM_DISABLE_COMPILE_CACHE", "1")
custom_ops_list = custom_ops.split(",") if custom_ops else []
device = torch.device(f"{DEVICE_TYPE}:0")