[Bugfix] Add init_workspace_manager to moe kernel benchmarks (#31042)

Signed-off-by: mgoin <mgoin64@gmail.com>

benchmarks/kernels/benchmark_cutlass_moe_fp8.py

@@ -15,6 +15,7 @@ from vllm.model_executor.layers.fused_moe.cutlass_moe import cutlass_moe_fp8
 from vllm.model_executor.layers.fused_moe.fused_moe import fused_experts, fused_topk
 from vllm.platforms import current_platform
 from vllm.utils.argparse_utils import FlexibleArgumentParser
+from vllm.v1.worker.workspace import init_workspace_manager
 
 # Weight shapes for different models: [num_experts, topk, hidden_size,
 # intermediate_size]
@@ -297,6 +298,10 @@ def bench_run(
 
 
 def main(args):
+    # Initialize workspace manager (required for CUTLASS MoE kernels)
+    device = torch.device("cuda:0")
+    init_workspace_manager(device)
+
     print("Benchmarking models:")
     for i, model in enumerate(args.models):
         print(f"[{i}]  {model}")

benchmarks/kernels/benchmark_cutlass_moe_nvfp4.py

@@ -21,6 +21,7 @@ from vllm.model_executor.layers.fused_moe.cutlass_moe import cutlass_moe_fp4
 from vllm.model_executor.layers.fused_moe.fused_moe import fused_experts, fused_topk
 from vllm.scalar_type import scalar_types
 from vllm.utils.argparse_utils import FlexibleArgumentParser
+from vllm.v1.worker.workspace import init_workspace_manager
 
 WEIGHT_SHAPES_MOE = {
     "nvidia/DeepSeek-R1-FP4": [
@@ -441,6 +442,10 @@ def bench_run(
 
 
 def main(args):
+    # Initialize workspace manager (required for CUTLASS MoE kernels)
+    device = torch.device("cuda:0")
+    init_workspace_manager(device)
+
     print("Benchmarking models:")
     for i, model in enumerate(args.models):
         print(f"[{i}]  {model}")

benchmarks/kernels/benchmark_grouped_gemm_cutlass.py

@@ -14,6 +14,7 @@ from vllm.model_executor.layers.fused_moe.fused_moe import (
     fused_topk,
 )
 from vllm.utils.argparse_utils import FlexibleArgumentParser
+from vllm.v1.worker.workspace import init_workspace_manager
 
 DEFAULT_MODELS = [
     "mistralai/Mixtral-8x7B-Instruct-v0.1",
@@ -364,6 +365,10 @@ def bench_run(
 
 
 def main(args):
+    # Initialize workspace manager (required for CUTLASS MoE kernels)
+    device = torch.device("cuda:0")
+    init_workspace_manager(device)
+
     print("Benchmarking models:")
     for i, model in enumerate(args.models):
         print(f"[{i}]  {model}")