benchmark

2ee0fd06 - Enable fp16 for torchvision models inference CUDA tests by default (#758)

Enable fp16 for torchvision models inference CUDA tests by default (#758) Summary: Pull Request resolved: https://github.com/pytorch/benchmark/pull/758 Reviewed By: frank-wei Differential Revision: D34294547 Pulled By: xuzhao9 fbshipit-source-id: 5c2834ffc11ed64f82489c56a290c3b9c569a24a