add rose tests

Signed-off-by: Bill Nell <bnell@redhat.com>

Author: bnellnm

tests/kernels/moe/modular_kernel_tools/common.py

@@ -23,7 +23,7 @@
  FusedMoEQuantConfig,
 )
 from vllm.model_executor.layers.fused_moe.fused_moe import fused_topk
-from vllm.utils.import_utils import has_deep_ep, has_deep_gemm, has_pplx
+from vllm.utils.import_utils import has_deep_ep, has_deep_gemm, has_pplx, has_rose
 
 from .mk_objects import (
  TestMoEQuantConfig,
@@ -199,6 +199,10 @@ def needs_pplx(self):
  info = prepare_finalize_info(self.prepare_finalize_type)
  return info.backend == "pplx"
 
+ def needs_rose(self):
+ info = prepare_finalize_info(self.prepare_finalize_type)
+ return info.backend == "rose"
+
  def needs_deep_ep(self):
  info = prepare_finalize_info(self.prepare_finalize_type)
  return (
@@ -273,6 +277,8 @@ def is_valid(self) -> tuple[bool, str | None]:
  return False, "Needs DeepGEMM, but DeepGEMM not available."
  if self.needs_pplx() and not has_pplx(): # noqa: SIM103
  return False, "Needs PPLX, but PPLX not available."
+ if self.needs_rose() and not has_rose(): # noqa: SIM103
+ return False, "Needs Rose, but Rose not available."
 
  return True, None
 

tests/kernels/moe/modular_kernel_tools/mk_objects.py

@@ -40,7 +40,7 @@
 from vllm.platforms import current_platform
 from vllm.utils.deep_gemm import is_deep_gemm_supported
 from vllm.utils.flashinfer import has_flashinfer_cutlass_fused_moe
-from vllm.utils.import_utils import has_deep_ep, has_deep_gemm, has_pplx
+from vllm.utils.import_utils import has_deep_ep, has_deep_gemm, has_pplx, has_rose
 
 
 @dataclass
@@ -234,6 +234,19 @@ def expert_info(kind) -> ExpertInfo:
  backend="pplx",
  )
 
+if has_rose():
+ from vllm.model_executor.layers.fused_moe.rose_prepare_finalize import (
+ RosePrepareAndFinalize,
+ )
+
+ register_prepare_and_finalize(
+ RosePrepareAndFinalize,
+ batched_format,
+ common_float_and_int_types,
+ blocked_quantization_support=True,
+ backend="rose",
+ )
+
 if has_flashinfer_cutlass_fused_moe() and current_platform.has_device_capability(100):
  from vllm.model_executor.layers.fused_moe.flashinfer_cutlass_moe import (
  FlashInferExperts,