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 11, 01:15 UTC
[AMD][CI] fix v1/engine test_preprocess_error_handling (#31192)
Signed-off-by: Divakar Verma <divakar.verma@amd.com>
Divakar Verma committed
Dec 22, 2025 at 19:28 UTC
78e5e62bbf9d59c1ba1fac3ec57d7899d80e3e39
1 file changed
+6
tests/v1/engine/test_preprocess_error_handling.py
+6
@@ -5,6 +5,7 @@ import pytest
5
import torch.cuda
6
7
from vllm import LLM, SamplingParams
8
+from vllm.platforms import current_platform
9
from vllm.v1.engine import EngineCoreRequest
10
from vllm.v1.engine.core import EngineCore
11
@@ -14,6 +15,11 @@ MODEL_NAME = "hmellor/tiny-random-LlamaForCausalLM"
15
def test_preprocess_error_handling(monkeypatch: pytest.MonkeyPatch):
16
"""Test that preprocessing errors are handled gracefully."""
17
18
+ if current_platform.is_rocm():
19
+ pytest.skip(
20
+ "Skipped on ROCm: this test only works with 'fork', but ROCm uses 'spawn'."
21
+ )
22
+
23
assert not torch.cuda.is_initialized(), (
24
"fork needs to be used for the engine "
25
"core process and this isn't possible if cuda is already initialized"