2 Commits

Author	SHA1	Message	Date
DefTruth	b2426aefa9	[Backend] add paddle custom ops compatible policy (#2070) ... * Add centerpoint * fix postprocess op file name * [Backend] add paddle custom ops compatible policy * [Backend] add paddle custom ops compatible policy * [Backend] add paddle custom ops compatible policy * upgrade linx paddle gpu -> 2.5 * add custom op compatible policy for paddle 2.5 * add custom op compatible policy for paddle 2.5 * add custom op compatible policy for paddle 2.5 * add collect_trt_shape_by_device option for paddle backend * add collect_trt_shape_by_device option for paddle backend * add custom op option for python build * fix python build bugs * update paddle linux x86 cpu only lib * update paddle linux gpu lib * update patchelf cmake * fix paddle backend option pybind * update paddle_inference.cmake * add cuda sm_80 support (A100) --------- Co-authored-by: zengshao0622 <peter_z96@163.com> Co-authored-by: qiuyanjun <qiuyanjun@baidu.com>	2023-06-29 22:32:14 +08:00
Wang Xinyu	e5c955dd3e	[Model] yolo use external stream, avoid reallocating output tensors (#447) ... * yolov5 use external stream * yolov5lite/v6/v7/v7e2etrt: optimize output tensor and cuda stream * avoid reallocating output tensors * add input output tensors to FastDeployModel * add cuda.cmake * rename to reused_input/output_tensors * eliminate cmake cuda arch error * use swap to release input and output tensors Co-authored-by: Jason <jiangjiajun@baidu.com>	2022-11-02 09:52:27 +08:00