Winters Montagne
Winters Montagne
### PR Category CINN ### PR Types Others ### Description Symbolic shape inference support for pd_op.bce_loss, pd_op.sigmoid_cross_entropy_with_logits
- Description 本文档是第二部分的设计文档,主要关于底层实现。 - Related Links - 第五期黑客松:https://github.com/PaddlePaddle/Paddle/issues/57262 - 第一部分的设计文档:https://github.com/PaddlePaddle/community/pull/709
 可能是因为paddle develop分支中fluid已经废弃了,但是benchmark中没有做出相应的调整
在/benchmark/api/tests目录下运行指令 bash run.sh,报错如下:  环境: 显卡:gtx 960 paddle:根据源码编译安装 cuda:10.2 tensorflow:2.1.0 tensorrt:6.0.1.8 cudnn:7.6.5 python:3.7.12 torch:1.12.0
### PR Category CINN ### PR Types Bug fixes ### Description Bug fixes for GaussianOpInferSymbolicShape
### PR Category CINN ### PR Types Bug fixes ### Description PaddleTest 中 layercase/sublayer1000/Seg_cases/danet_danet_resnet50_os8_voc12aug_512x512_40k/SIR_34.py 报错,详细信息如下: ``` NotImplementedError: (Unimplemented) pd_op.add_grad DOES NOT have InferSymbolicShapeInterface! (at /home/aistudio/Paddle/paddle/pir/src/dialect/shape/transforms/shape_optimization_pass.cc:333) ``` 将 "pd_op.add" 添加至 ALLOW_DYNAMIC_SHAPE_VJP_OPS,解决了这个问题。...