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, 12:45 UTC
[ROCm]Skip test_torchao.py::test_pre_quantized_model on CDNA3 arch (#31905)
Signed-off-by: ZhiweiYan-96 <zhiwei.yan@amd.com>
Zhiwei committed
Jan 8, 2026 at 15:47 UTC
573a1d1119af85613ff0cb90ac063ab669cbbd7f
1 file changed
+6
tests/quantization/test_torchao.py
+6
@@ -6,11 +6,17 @@ import importlib.util
6
import pytest
7
import torch
8
9
+from vllm.platforms import current_platform
10
+
11
DTYPE = ["bfloat16"]
12
13
TORCHAO_AVAILABLE = importlib.util.find_spec("torchao") is not None
14
15
16
+@pytest.mark.skipif(
17
+ current_platform.is_rocm() and current_platform.is_fp8_fnuz(),
18
+ reason="Only fp8_fnuz supported on CDNA3 architecture",
19
+)
20
@pytest.mark.skipif(not TORCHAO_AVAILABLE, reason="torchao is not available")
21
def test_pre_quantized_model(vllm_runner):
22
with vllm_runner(