[ROCm][CI] Use a platform-independent GEMM in the merged-column fuser test (#56130)
Signed-off-by: Djordje Ramic <djoramic@amd.com>
djramic committed
Sep 9, 2026 at 22:08 UTC
cce50657b742598a5b8874fc9fe3a94dc4344a14
1 file changed
+6
-2
tests/models/transformers/fusers/test_linear.py
+6
-2
@@ -1145,11 +1145,16 @@ def test_merged_column_fuser_supports_any_number_of_linears(
1145
from vllm.model_executor import parameter
1146
from vllm.model_executor.layers import linear
1147
from vllm.model_executor.layers.linear import MergedColumnParallelLinear
1148
- from vllm.model_executor.layers.utils import dispatch_cpu_unquantized_gemm
1148
+ from vllm.model_executor.layers.utils import default_unquantized_gemm
1149
from vllm.model_executor.models.utils import AutoWeightsLoader, WeightsMapper
1150
1151
monkeypatch.setattr(linear, "get_tensor_model_parallel_rank", lambda: 0)
1152
monkeypatch.setattr(linear, "get_tensor_model_parallel_world_size", lambda: 1)
1153
+ monkeypatch.setattr(
1154
+ linear,
1155
+ "dispatch_unquantized_gemm",
1156
+ lambda linear_backend="auto": default_unquantized_gemm,
1157
+ )
1158
monkeypatch.setattr(parameter, "get_tensor_model_parallel_rank", lambda: 0)
1159
monkeypatch.setattr(parameter, "get_tensor_model_parallel_world_size", lambda: 1)
1160
with torch.device("meta"):
@@ -1176,7 +1181,6 @@ def test_merged_column_fuser_supports_any_number_of_linears(
1181
1182
merged = getattr(fused, fuser.merged_name)
1183
assert isinstance(merged, MergedColumnParallelLinear)
1179
- dispatch_cpu_unquantized_gemm(merged, remove_weight=False)
1184
for actual, reference in zip(fused(x), expected):
1185
torch.testing.assert_close(actual, reference)
1186