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 10, 01:30 UTC
[ROCm][CI] Skip Qwen3-30B-A3B-MXFP4A16 Eval Test On Non-CUDA Platforms (#32460)
Signed-off-by: Micah Williamson <micah.williamson@amd.com>
Micah Williamson committed
Jan 16, 2026 at 02:17 UTC
b84c426a8c48c25db6a4a1a14860e845347db1c1
1 file changed
+11
tests/evals/gsm8k/test_gsm8k_correctness.py
+11
@@ -11,9 +11,11 @@ pytest -s -v tests/evals/gsm8k/test_gsm8k_correctness.py \
11
12
import shlex
13
14
+import pytest
15
import yaml
16
17
from tests.utils import RemoteOpenAIServer
18
+from vllm.platforms import current_platform
19
20
from .gsm8k_eval import evaluate_gsm8k
21
@@ -53,6 +55,15 @@ def test_gsm8k_correctness(config_filename):
55
"""Test GSM8K correctness for a given model configuration."""
56
eval_config = yaml.safe_load(config_filename.read_text(encoding="utf-8"))
57
58
+ if (
59
+ not current_platform.is_cuda()
60
+ and "Qwen3-30B-A3B-MXFP4A16" in eval_config["model_name"]
61
+ ):
62
+ pytest.skip(
63
+ "Skipping Qwen3-30B-A3B-MXFP4A16 on non-CUDA platforms. "
64
+ "Marlin kernels are not supported."
65
+ )
66
+
67
# Parse server arguments from config (use shlex to handle quoted strings)
68
server_args_str = eval_config.get("server_args", "")
69
server_args = shlex.split(server_args_str) if server_args_str else []