[ROCm][CI] Extended Fused MoE and FP8 MoE test support (#41100)

Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Signed-off-by: Andreas Karatzas <Andreas.Karatzas@amd.com>
This commit is contained in:
Andreas Karatzas
2026-08-19 10:19:42 -05:00
committed by GitHub
parent be06873198
commit 160f7f0840
9 changed files with 331 additions and 31 deletions
+13
View File
@@ -31,6 +31,9 @@ from vllm.model_executor.layers.quantization.kv_cache import BaseKVCacheMethod
from vllm.model_executor.layers.quantization.online.fp8 import (
Fp8PerTensorOnlineLinearMethod,
)
from vllm.model_executor.layers.quantization.utils.fp8_utils import (
process_fp8_input_tensor_strategy_moe,
)
from vllm.model_executor.model_loader.weight_utils import default_weight_loader
from vllm.platforms import current_platform
@@ -47,6 +50,16 @@ MODELS = [
]
def test_static_fp8_moe_input_scales_remain_scalar() -> None:
a1_scale, a2_scale = process_fp8_input_tensor_strategy_moe(
torch.tensor([0.25, 0.5]),
torch.tensor([0.75, 0.6]),
enable_eplb=False,
)
assert a1_scale.ndim == a2_scale.ndim == 0
@pytest.mark.skipif(
not is_quant_method_supported("fp8"),
reason="FP8 is not supported on this GPU type.",