Fix vbe benchmark for MTIA (#4171)

Summary:
Pull Request resolved: #4171

X-link: facebookresearch/FBGEMM#1251

This Diff fixes the VBE benchmark for MTIA.

Reviewed By: q10, evanleed

Differential Revision: D75195213

fbshipit-source-id: 1d4d8f25ce8f164effaca43e08ce7e27ab44e684

Author: shintaro-iwasaki

fbgemm_gpu/bench/tbe/split_table_batched_embeddings_benchmark.py

@@ -1350,6 +1350,7 @@ def vbe(
         output_dtype=embconfig.output_dtype,
         pooling_mode=embconfig.pooling_mode,
         bounds_check_mode=embconfig.bounds_check_mode,
+        device=get_device(),
     ).to(get_device())
 
     all_requests = {