xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-29 14:13:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

[ROCm][CI] Skip multi-GPU speculative decoding tests when insufficient GPUs available (#30527)

Browse Source 
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
This commit is contained in:
Andreas Karatzas 2025-12-11 21:54:56 -06:00 committed by GitHub
parent 197473c4e7
commit 783644e4ac
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 13 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
tests/v1/e2e/test_spec_decode.py
View File

@ -16,6 +16,16 @@ from vllm.platforms import current_platform
MTP_SIMILARITY_RATE = 0.8
def _skip_if_insufficient_gpus_for_tp(tp_size: int):
"""Skip test if available GPUs < tp_size on ROCm."""
if current_platform.is_rocm():
available_gpus = torch.cuda.device_count()
if available_gpus < tp_size:
pytest.skip(
f"Test requires {tp_size} GPUs, but only {available_gpus} available"
)
def get_test_prompts(mm_enabled: bool):
prompt_types = ["repeat", "sentence"]
if mm_enabled:
@ -455,6 +465,8 @@ def test_eagle_correctness(
m.setenv("VLLM_ROCM_USE_AITER", "1")
method, model_name, spec_model_name, tp_size = model_setup
_skip_if_insufficient_gpus_for_tp(tp_size)
max_model_len = 2048
max_num_batched_tokens = 128 if enable_chunked_prefill else max_model_len
@ -525,6 +537,7 @@ def test_mtp_correctness(
m.setenv("VLLM_MLA_DISABLE", "1")
method, model_name, tp_size = model_setup
_skip_if_insufficient_gpus_for_tp(tp_size)
ref_llm = LLM(
model=model_name,