[dashboard] Allocate 4 shards for torchbench (#103280)
Summary: Enabling inference nightly run unproportionally slows down
torchbench, so allocate more shards for it.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/103280
Approved by: https://github.com/huydhn