Awesome video understanding toolkits based on PaddlePaddle. It supports video data annotation tools, lightweight RGB and skeleton based action recognition model, practical applications for video tagging and sport action detection.
Apache License 2.0
1.46k
stars
374
forks
source link
【benchmark】add max_mem_reserved for benchmark #657
训练benchmark 使用 API paddle.device.cuda.max_memory_reserved() 收集模型训练时的显存占用，因此本PR 修改 benchmark 打印日志部分，新增 max_mem_reserved max_mem_allocated 指标打印；如下： [11/20 19:22:41] epoch:[ 1/1 ] train step:199 loss: 1.40318 lr: 0.005365 top1: 0.81250 top5: 1.00000 batch_cost: 0.43920 sec, reader_cost: 0.00056 sec, ips: 36.42998 instance/sec, eta: 0:00:24, max_mem_reserved: 19813035008 B, max_mem_allocated: 18339896576 B