Paddle

Paddle copied to clipboard

[PHI] fix paddle.allclose for big tensor

4

### PR Category Execute Infrastructure ### PR Types Bug fixes ### Description ### 存在的问题： 输出tensor的numel超过int的最大值时出现int溢出进而出现访存越界 ### 修复方法： numel超过int 最大值时使用int64 否则使用用int32 ### 回测结果：  cuda error 700 已经修复。有一些case报invalid arguement，这不是big Tensor造成的，是allclose不支持shape不同的输入Tensors pcard-67164

DanielSun11

[PHI] fix paddle.assign for big tensor

1

### PR Category Execute Infrastructure ### PR Types Bug fixes ### Description **报错：**  报错是因为assign里面limit设置的太小了 **PaddleAPITest回测结果：**  pcard-67164

DanielSun11

[0-size Tensor No.175] Add 0-size Tensor support for group_norm

3

### PR Category Execute Infrastructure ### PR Types Improvements ### Description 待修改 [0-size Tensor No.175] Add 0-size Tensor support for group_norm 修改前向和反向 infermeta没有修改 kernel修改cpu/gpu/xpu group和channel维度需要一致并大于0 PaddleAPITest 测试通过 

co63oc

Add offload activation

1

### PR Category Performance Optimization ### PR Types New features ### Description Add offload activation.

sneaxiy

Migrate device_event to fluid

1

### PR Category User Experience ### PR Types Others ### Description 迁移 device_event 到 fluid， fluid 头文件不在phi中包含

co63oc

### PR Category Execute Infrastructure ### PR Types Improvements ### Description pcard-67164

xiaoguoguo626807

### PR Category Execute Infrastructure ### PR Types Improvements ### Description paddle.nn.functional.smooth_l1_loss PaddleAPITest 测试通过，增加单测 

co63oc

### PR Category Execute Infrastructure ### PR Types Improvements ### Description paddle.incubate.softmax_mask_fuse_upper_triangle 修改GPU 前向和反向 infermeta 没有修改 PaddleAPITest测试通过 

co63oc

[0-size Tensor Job2 No.100] Add 0-size Tensor support for topk

1

### PR Category Execute Infrastructure ### PR Types Improvements ### Description 修改前向和后向 infermeta没有修改 修改kernel CPU/GPU/XPU PaddleAPITest 为numpy error GPU numpy error  CPU numpy error 

co63oc

[0-size Tensor No.180、182-183] Add 0-size Tensor support for hinge_embedding_loss

7

### PR Category Execute Infrastructure ### PR Types Improvements ### Description 180 paddle.nn.functional.hinge_embedding_loss GPU 测试通过 CPU 测试通过 182 paddle.nn.functional.kl_div GPU 测试通过  CPU 测试通过  183 paddle.nn.functional.l1_loss GPU 测试通过 CPU...

co63oc