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 6, 13:15 UTC
[ROCm][CI] Skip trtllm kvfp8 dequant tests on ROCm (#37330)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Mar 17, 2026 at 22:12 UTC
58cde5c026efee42987fbc87681ecbf262f9db2b
1 file changed
+6
tests/kernels/attention/test_trtllm_kvfp8_dequant.py
+6
@@ -12,6 +12,12 @@ import torch
12
13
from vllm.platforms import current_platform
14
15
+if current_platform.is_rocm():
16
+ pytest.skip(
17
+ "trtllm kvfp8 dequant is not supported on ROCm.",
18
+ allow_module_level=True,
19
+ )
20
+
21
FP8_DTYPE = current_platform.fp8_dtype()
22
23
NUM_BLOCKS = 128