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, 22:30 UTC
[CI/Build][AMD] Skip marlin, machete, and hadacore tests since these require _C functions not defined for ROCm (#30109)
Signed-off-by: Randall Smith <ransmith@amd.com> Co-authored-by: Randall Smith <ransmith@amd.com>
rasmith committed
Dec 5, 2025 at 22:54 UTC
62079d86004448a42c4205d00e4a977fe85b69a6
3 files changed
+21
tests/kernels/quantization/test_hadacore.py
+7
@@ -8,6 +8,13 @@ import torch
8
from compressed_tensors.transform import deterministic_hadamard_matrix
9
10
from vllm import _custom_ops as ops
11
+from vllm.platforms import current_platform
12
+
13
+if current_platform.is_rocm():
14
+ pytest.skip(
15
+ "These tests require hadacore_transform, not supported on ROCm.",
16
+ allow_module_level=True,
17
+ )
18
19
20
@pytest.mark.parametrize("batch_size", [1, 32])
tests/kernels/quantization/test_machete_mm.py
+6
@@ -23,6 +23,12 @@ from vllm.model_executor.layers.quantization.utils.quant_utils import (
23
from vllm.platforms import current_platform
24
from vllm.scalar_type import ScalarType, scalar_types
25
26
+if current_platform.is_rocm():
27
+ pytest.skip(
28
+ "These tests require machete_prepack_B, not supported on ROCm.",
29
+ allow_module_level=True,
30
+ )
31
+
32
CUDA_DEVICES = [f"cuda:{i}" for i in range(1 if torch.cuda.device_count() == 1 else 2)]
33
34
# TODO: in future PR refactor this and `is_quant_method_supported` in the kernel
tests/kernels/quantization/test_marlin_gemm.py
+8
@@ -56,6 +56,14 @@ from vllm.model_executor.layers.quantization.utils.quant_utils import (
56
from vllm.platforms import current_platform
57
from vllm.scalar_type import scalar_types
58
59
+if current_platform.is_rocm():
60
+ pytest.skip(
61
+ "These tests require gptq_marlin_repack,"
62
+ "marlin_int4_fp8_preprocess, gptq_marlin_24_gemm,"
63
+ "or gptq_marlin_gemm which are not supported on ROCm.",
64
+ allow_module_level=True,
65
+ )
66
+
67
ACT_ORDER_OPTS = [False, True]
68
K_FULL_OPTS = [False, True]
69
USE_ATOMIC_ADD_OPTS = [False, True]