[CI] Speed up quantization test group (#53291)

Signed-off-by: Felix Marty <Felix.Marty@amd.com>
Co-authored-by: Andreas Karatzas <akaratza@amd.com>
This commit is contained in:
fxmarty-amd
2026-09-01 13:15:24 -05:00
committed by GitHub
co-authored by Andreas Karatzas
parent e7cf4730d6
commit d1c15e589d
20 changed files with 824 additions and 725 deletions
+14
View File
@@ -11,6 +11,7 @@ from vllm.utils.torch_utils import (
get_kv_cache_torch_dtype,
is_lossless_cast,
is_quantized_kv_cache,
set_default_torch_dtype,
set_torch_threads_for_runtime,
startup_omp_num_threads,
)
@@ -28,6 +29,19 @@ def test_nvfp4_4over6_cache_dtype() -> None:
assert get_kv_quant_mode(cache_config.cache_dtype) == KVQuantMode.NVFP4
def test_set_default_torch_dtype_restores_dtype_after_exception() -> None:
original_dtype = torch.get_default_dtype()
with (
pytest.raises(RuntimeError, match="expected failure"),
set_default_torch_dtype(torch.bfloat16),
):
assert torch.get_default_dtype() == torch.bfloat16
raise RuntimeError("expected failure")
assert torch.get_default_dtype() == original_dtype
@pytest.mark.parametrize(
("src_dtype", "tgt_dtype", "expected_result"),
[