Open mls1999725 opened 3 years ago
all results are tested using fp16 if the gpu has tensor core, otherwise fp32.
all results are tested using fp16 if the gpu has tensor core, otherwise fp32.
Thanks.
By the way, use model.half() to achieve testing by fp16, is that right ?
yes.
Hi, I wonder whether Yolov4-CSP and Yolov4-Large models are tested using FP32 or FP16 ? Looking forward to your reply.