[Tests] Clarify pytest skip reasons with actionable context (#32981)

Signed-off-by: 7. Sun <jhao.sun@gmail.com>
This commit is contained in:
7. Sun
2026-01-24 06:38:50 +00:00
committed by GitHub
parent 14d03b8ddb
commit 0b9a735e11
2 changed files with 7 additions and 3 deletions

View File

@@ -20,7 +20,7 @@ MM_BEAM_WIDTHS = [2]
MODELS = ["TinyLlama/TinyLlama-1.1B-Chat-v1.0"]
@pytest.mark.skip_v1 # FIXME: This fails on V1 right now.
@pytest.mark.skip_v1 # V1 engine does not yet support beam search
@pytest.mark.parametrize("model", MODELS)
@pytest.mark.parametrize("dtype", ["half"])
@pytest.mark.parametrize("max_tokens", MAX_TOKENS)
@@ -62,7 +62,7 @@ def test_beam_search_single_input(
)
@pytest.mark.skip_v1 # FIXME: This fails on V1 right now.
@pytest.mark.skip_v1 # V1 engine does not yet support beam search
@pytest.mark.parametrize("model", MODELS)
@pytest.mark.parametrize("dtype", ["half"])
@pytest.mark.parametrize("max_tokens", MAX_TOKENS)

View File

@@ -48,7 +48,11 @@ def test_topk_impl_equivalence():
assert torch.allclose(result1, result2)
@pytest.mark.skip(reason="FIXME: This test is failing right now.")
@pytest.mark.skip(
reason="FlashInfer top-k/top-p renorm comparison fails; "
"needs investigation of tolerance threshold or "
"interface differences between Python and FlashInfer implementations"
)
def test_flashinfer_sampler():
"""
This test verifies that the FlashInfer top-k and top-p sampling