benchmark
f7a62d6f - Force fp16 for vision_maskrcnn inference (#113110)

Commit
2 years ago
Force fp16 for vision_maskrcnn inference (#113110) Summary: For fp16 for maskrcnn inference (doesnt support bf16). Also skip phi_1_5 in training - it OOMs even with batch size 1 X-link: https://github.com/pytorch/pytorch/pull/113110 Approved by: https://github.com/xmfan Reviewed By: PaliC Differential Revision: D51190243 Pulled By: eellison fbshipit-source-id: 94b0b674b73cf92dabee5121747071daf9ad710d
Author
Parents
Loading