[sgl-kernel][3/N]Support Expert Specialization Grouped GEMM (#11674)

sgl-kernel/benchmark/bench_es_fp8_blockwise_grouped_gemm.py

@@ -133,6 +133,7 @@ def bench_es(
     d_strides = torch.full(
         (num_groups,), c_out.stride(0), device=device, dtype=torch.int64
     )
+    workspace = torch.empty((1024 * 1024 * 1024), device=device, dtype=torch.uint8)
 
     def run_cutlass():
         es_fp8_blockwise_scaled_grouped_mm(
@@ -146,6 +147,7 @@ def bench_es(
             d_strides,
             problem_sizes,
             expert_offsets[:-1],
+            workspace,
         )
 
     run_cutlass()