一、背景

0-Size Tensor是指，Tensor.shape中存在某维度值为0，进而该Tensor虽有shape、dtype、place等信息，但他的元素个数为0。在切割等特定业务中，0-Size出现频率较高。Paddle目前对0-Size的支持非常薄弱，需要系统性摸排、建设。参与本项活动，你将学习到Paddle算子库框架的设计，并了解Paddle动态图的整体执行流程。对深度学习框架0-Size业务有深刻了解。也有可能会涉猎到Paddle组合算子机制和编译器符号推导机制。遇到问题我们将有专人解答。

二、任务描述

本次开源任务主要用于完善Paddle的一些API对0-size Tensor的支持，这些不支持0-size Tensor的API遇到0-size Tensor时会出现例如精度异常、Coredump、Cuda Error、报错等问题。这些存在问题的API大概率是由于Kernel没有考虑到0-Size场景而产生的，当然也涉及到API shape的推导。其他的成因需要具体分析。需要修复的API如下：

[!IMPORTANT]
每个任务难度：0.025×🌟 题目讲解见录屏文件：https://meeting.tencent.com/crm/l59EWmRZc4 （00:16:00~00:36:00）详细介绍见：

https://github.com/PaddlePaddle/Paddle/issues/72637#issuecomment-2879353324

报名格式参考：【报名】：2、3，多个任务之间需要使用中文顿号分隔，报名多个连续任务可用横线表示，如 2-5
PR 提交格式：在 PR 的标题中以【0-size Tensor Job2 No.xxx】开头，注明任务编号。请注意，标题中一定要加Job2，以和第一批任务区分。

序号	API	报名人/状态	CPU/GPU/testcase
1	paddle.add		gpu
2	paddle.add_n		gpu,cpu
3	paddle.bincount		gpu,cpu
4	paddle.broadcast_tensors		gpu
5	paddle.cdist		gpu,cpu
6	paddle.crop	@co63oc #73333 #73299	gpu,cpu
7	paddle.cumprod	@co63oc #73333	gpu
8	paddle.divide		gpu
9	paddle.dot		gpu,cpu
10	paddle.equal_all		gpu
11	paddle.expand_as		gpu,cpu
12	paddle.full		gpu,cpu
13	paddle.gammainc		gpu
14	paddle.gammaincc	@co63oc #73207	gpu
15	paddle.gather		gpu,cpu
16	paddle.gather_nd		gpu,cpu
17	paddle.geometric.segment_max	@co63oc #73341	gpu,cpu
18	paddle.geometric.segment_mean	@co63oc #73341	gpu,cpu
19	paddle.geometric.segment_min	@co63oc #73341	gpu,cpu
20	paddle.geometric.segment_sum	@co63oc #73341	gpu,cpu
21	paddle.geometric.send_u_recv		gpu,cpu
22	paddle.geometric.send_ue_recv		gpu,cpu
23	paddle.geometric.send_uv		gpu,cpu
24	paddle.histogram		gpu,cpu
25	paddle.increment		gpu,cpu
26	paddle.incubate.nn.functional.fused_bias_dropout_residual_layer_norm		gpu
27	paddle.incubate.nn.functional.fused_linear		gpu
28	paddle.incubate.nn.functional.fused_linear_activation		gpu
29	paddle.incubate.nn.functional.fused_multi_head_attention		gpu
30	paddle.incubate.nn.functional.swiglu	@co63oc #73346	gpu,cpu
31	paddle.incubate.segment_max	@co63oc #73341	gpu,cpu
32	paddle.incubate.segment_mean	@co63oc #73341	gpu,cpu
33	paddle.incubate.segment_min	@co63oc #73341	gpu,cpu
34	paddle.incubate.segment_sum	@co63oc #73341	gpu,cpu
35	paddle.incubate.softmax_mask_fuse_upper_triangle	@co63oc #73353	gpu,cpu
36	paddle.index_add		gpu,cpu
37	paddle.index_put		gpu,cpu
38	paddle.index_sample	@co63oc #73276	gpu,cpu
39	paddle.index_select	@co63oc #73270	gpu,cpu
40	paddle.linalg.cholesky		gpu,cpu
41	paddle.linalg.eigh		gpu,cpu
42	paddle.linalg.eigvals		gpu,cpu
43	paddle.linalg.lu		gpu,cpu
44	paddle.linalg.lu_unpack		gpu,cpu
45	paddle.linalg.svdvals		gpu,cpu
46	paddle.masked_scatter		gpu,cpu
47	paddle.max		gpu,cpu
48	paddle.median		gpu,cpu
49	paddle.min		gpu,cpu
50	paddle.mode		gpu,cpu
51	paddle.multiplex		gpu,cpu
52	paddle.multiply		gpu
53	paddle.nn.functional.adaptive_log_softmax_with_loss		gpu,cpu
54	paddle.nn.functional.binary_cross_entropy		gpu,cpu
55	paddle.nn.functional.cross_entropy		gpu,cpu
56	paddle.nn.functional.dice_loss		gpu,cpu
57	paddle.nn.functional.embedding		gpu,cpu
58	paddle.nn.functional.gather_tree		gpu,cpu
59	paddle.nn.functional.gaussian_nll_loss		gpu,cpu
60	paddle.nn.functional.instance_norm		gpu,cpu
61	paddle.nn.functional.log_loss		gpu,cpu
62	paddle.nn.functional.margin_cross_entropy		gpu,cpu
63	paddle.nn.functional.nll_loss		gpu,cpu
64	paddle.nn.functional.one_hot		gpu
65	paddle.nn.functional.smooth_l1_loss	@co63oc #73354	gpu
66	paddle.nn.functional.softmax_with_cross_entropy		gpu,cpu
67	paddle.nn.functional.upsample		gpu,cpu
68	paddle.pow	@co63oc #73204	gpu
69	paddle.prod		gpu,cpu
70	paddle.put_along_axis		gpu,cpu
71	paddle.scale		gpu,cpu
72	paddle.scatter		gpu,cpu
73	paddle.scatter_nd		gpu,cpu
74	paddle.scatter_nd_add		gpu,cpu
75	paddle.shard_index		gpu,cpu
76	paddle.slice_scatter		cpu
77	paddle.squeeze		gpu,cpu
78	paddle.sum	@co63oc #73355	gpu,cpu
79	paddle.Tensor.add		gpu
80	paddle.Tensor.mul		gpu
81	paddle.Tensor.truediv		gpu
82	paddle.Tensor.add		gpu
83	paddle.Tensor.cumprod		gpu
84	paddle.Tensor.gather		gpu,cpu
85	paddle.Tensor.gather_nd		gpu,cpu
86	paddle.Tensor.index_select	@co63oc #73270	gpu,cpu
87	paddle.Tensor.masked_fill		gpu
88	paddle.Tensor.max		gpu,cpu
89	paddle.Tensor.mean		gpu,cpu
90	paddle.Tensor.min		gpu,cpu
91	paddle.Tensor.multiply		gpu
92	paddle.Tensor.pow	@co63oc #73204	gpu
93	paddle.Tensor.prod		gpu,cpu
94	paddle.Tensor.put_along_axis		gpu,cpu
95	paddle.Tensor.real		gpu,cpu
96	paddle.Tensor.reshape		gpu,cpu
97	paddle.Tensor.set_		gpu,cpu
98	paddle.Tensor.sum	@co63oc #73355	gpu,cpu
99	paddle.Tensor.take_along_axis		gpu,cpu
100	paddle.topk	@co63oc #73352	gpu,cpu
101	paddle.unfold		gpu
102	paddle.vander		gpu,cpu
103	paddle.vision.ops.deform_conv2d		gpu,cpu
104	paddle.vision.ops.distribute_fpn_proposals		gpu,cpu
105	paddle.vision.ops.roi_align		gpu,cpu
106	paddle.vision.ops.roi_pool		gpu,cpu