A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[CI] Skip tests failing on main (#25326)
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Woosuk Kwon committed
Sep 20, 2025 at 19:57 UTC
72dd1595b466383c8c9c9e39cb519814df1a105f
4 files changed
+10
-2
tests/entrypoints/openai/test_completion_with_prompt_embeds.py
+1
@@ -60,6 +60,7 @@ def create_dummy_embeds(num_tokens: int = 5) -> str:
60
return base64.b64encode(buffer.getvalue()).decode('utf-8')
61
62
63
+@pytest.mark.skip("This test is skipped because it is flaky.")
64
@pytest.mark.asyncio
65
@pytest.mark.parametrize("model_name", [MODEL_NAME])
66
async def test_completions_with_prompt_embeds(
tests/models/quantization/test_fp8.py
+4
-1
@@ -32,7 +32,7 @@ from ..utils import check_logprobs_close
32
# Due to low-precision numerical divergence, we only test logprob of 4 tokens
33
@pytest.mark.parametrize("max_tokens", [4])
34
@pytest.mark.parametrize("enforce_eager", [True])
35
-@pytest.mark.parametrize("backend", ["FLASH_ATTN", "XFORMERS"])
35
+@pytest.mark.parametrize("backend", ["FLASH_ATTN"])
36
# NOTE: Increasing this in this suite will fail CI because we currently cannot
37
# reset distributed env properly. Use a value > 1 just when you test.
38
@pytest.mark.parametrize("tensor_parallel_size", [1])
@@ -57,6 +57,9 @@ def test_models(
57
pytest.skip(
58
f"{kv_cache_dtype} is currently not supported on ROCm/HIP.")
59
60
+ if not current_platform.is_kv_cache_dtype_supported(kv_cache_dtype, None):
61
+ pytest.skip(f"{kv_cache_dtype} is not supported on this platform.")
62
+
63
with monkeypatch.context() as m:
64
m.setenv("TOKENIZERS_PARALLELISM", 'true')
65
m.setenv(STR_BACKEND_ENV_VAR, backend)
tests/models/test_oot_registration.py
+1
@@ -63,6 +63,7 @@ def test_oot_registration_embedding(
63
image = convert_image_mode(ImageAsset("cherry_blossom").pil_image, "RGB")
64
65
66
+@pytest.mark.skip(reason="This test is skipped because it failed on V1.")
67
@create_new_process_for_each_test()
68
def test_oot_registration_multimodal(
69
monkeypatch: pytest.MonkeyPatch,
tests/quantization/test_compressed_tensors.py
+4
-1
@@ -357,6 +357,9 @@ def test_compressed_tensors_fp8(vllm_runner):
357
assert output
358
359
360
+@pytest.mark.skipif(
361
+ not current_platform.is_kv_cache_dtype_supported("fp8", None),
362
+ reason="FP8 KV cache is not supported on this device.")
363
@pytest.mark.skipif(not current_platform.is_cuda(),
364
reason="This test is skipped on non-CUDA platform.")
365
def test_compressed_tensors_kv_cache(vllm_runner):
@@ -738,4 +741,4 @@ def test_compressed_tensors_transforms_perplexity(vllm_runner, model, prompt,
741
with vllm_runner(model, enforce_eager=True) as llm:
742
perplexity = llm.generate_prompt_perplexity([prompt])[0]
743
print(perplexity)
741
- assert perplexity <= exp_perplexity
\ No newline at end of file
744
+ assert perplexity <= exp_perplexity