xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-23 17:51:21 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fixing Chunked Prefill Test. (#19762)

Browse Source 
Signed-off-by: Alexei V. Ivanov <alexei.ivanov@amd.com>
This commit is contained in:
Alexei-V-Ivanov-AMD 2025-06-19 03:36:16 -05:00 committed by GitHub
parent 466166dcfd
commit 4719460644
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 17 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
.buildkite/test-pipeline.yaml
View File

@ -89,7 +89,7 @@ steps:
- VLLM_TEST_ENABLE_ARTIFICIAL_PREEMPT=1 pytest -v -s basic_correctness/test_preemption.py - VLLM_TEST_ENABLE_ARTIFICIAL_PREEMPT=1 pytest -v -s basic_correctness/test_preemption.py
- label: Chunked Prefill Test - label: Chunked Prefill Test
mirror_hardwares: [amdexperimental] mirror_hardwares: [amdexperimental, amdproduction]
source_file_dependencies: source_file_dependencies:
- vllm/ - vllm/
- tests/basic_correctness/test_chunked_prefill - tests/basic_correctness/test_chunked_prefill

18
tests/basic_correctness/test_chunked_prefill.py
View File

@ -49,7 +49,13 @@ def use_v0_only(monkeypatch: pytest.MonkeyPatch):
# NOTE: Increasing this in this suite will fail CI because we currently cannot # NOTE: Increasing this in this suite will fail CI because we currently cannot
# reset distributed env properly. Use a value > 1 just when you test. # reset distributed env properly. Use a value > 1 just when you test.
@pytest.mark.parametrize("tensor_parallel_size", [1]) @pytest.mark.parametrize("tensor_parallel_size", [1])
@pytest.mark.parametrize("attention_backend", ["FLASHINFER", "FLASH_ATTN"]) @pytest.mark.parametrize("attention_backend", [
pytest.param("FLASHINFER",
marks=pytest.mark.skipif(
current_platform.is_rocm(),
reason="FLASHINFER isn't supported on ROCm")),
"FLASH_ATTN"
])
def test_models( def test_models(
hf_runner: HfRunner, hf_runner: HfRunner,
vllm_runner: VllmRunner, vllm_runner: VllmRunner,
@ -99,7 +105,13 @@ def test_models(
@multi_gpu_test(num_gpus=2) @multi_gpu_test(num_gpus=2)
@pytest.mark.parametrize("distributed_executor_backend", ["ray", "mp"]) @pytest.mark.parametrize("distributed_executor_backend", ["ray", "mp"])
@pytest.mark.parametrize("model", MODELS) @pytest.mark.parametrize("model", MODELS)
@pytest.mark.parametrize("attention_backend", ["FLASHINFER", "FLASH_ATTN"]) @pytest.mark.parametrize("attention_backend", [
pytest.param("FLASHINFER",
marks=pytest.mark.skipif(
current_platform.is_rocm(),
reason="FLASHINFER isn't supported on ROCm")),
"FLASH_ATTN"
])
def test_models_distributed( def test_models_distributed(
hf_runner: HfRunner, hf_runner: HfRunner,
vllm_runner: VllmRunner, vllm_runner: VllmRunner,
@ -172,6 +184,8 @@ def test_models_distributed(
# Due to low-precision numerical divergence, this test is too sensitive to # Due to low-precision numerical divergence, this test is too sensitive to
# the async postprocessor # the async postprocessor
@pytest.mark.parametrize("disable_async_output_proc", [True]) @pytest.mark.parametrize("disable_async_output_proc", [True])
@pytest.mark.skipif(current_platform.is_rocm(),
reason="machete_prepack_B isn't supported on ROCm")
def test_models_with_fp8_kv_cache( def test_models_with_fp8_kv_cache(
vllm_runner: VllmRunner, vllm_runner: VllmRunner,
example_prompts, example_prompts,