FastDeploy icon indicating copy to clipboard operation
FastDeploy copied to clipboard

docker容器部署后,每次调用gpu显存不释放,直到溢出,这个问题有很多issues,但没找到解决方案

Open zhuxiaobin opened this issue 11 months ago • 3 comments

容器镜像:registry.baidubce.com/paddlepaddle/fastdeploy:1.0.7-gpu-cuda11.4-trt8.5-21.10 调用一万次后,显存直接爆了 W0314 04:50:46.438977 62225 memory.cc:135] Failed to allocate CUDA memory with byte size 79027200 on GPU 1: CNMEM_STATUS_OUT_OF_MEMORY, falling back to pinned system memory 0314 05:01:17.338640 62420 pb_stub.cc:402] Failed to process the request(s) for model 'det_postprocess_0_0', message: TritonModelException: in ensemble 'rec_pp', softmax_2.tmp_0: failed to perform CUDA copy: an illegal memory access was encountered

zhuxiaobin avatar Mar 14 '24 05:03 zhuxiaobin

triton client 收到显存异常退出连接后,docker server端还占用着显存

zhuxiaobin avatar Mar 14 '24 05:03 zhuxiaobin

子进程模式用后销毁,不适合triton这种高并发server模式,还有什么解决方法么?模型是自己训练的高精度模型,模型较大,所以cpu推理速度太慢,gpu比较合适,但是一直占着显存,这就尴尬了,再多卡也扛不住

zhuxiaobin avatar Mar 14 '24 10:03 zhuxiaobin

子进程模式用后销毁,不适合triton这种高并发server模式,还有什么解决方法么?模型是自己训练的高精度模型,模型较大,所以cpu推理速度太慢,gpu比较合适,但是一直占着显存,这就尴尬了,再多卡也扛不住

大佬找到解决方案了吗?

KyleWang-Hunter avatar Mar 20 '24 01:03 KyleWang-Hunter