66 Pull requests merged by 34 people

• [AP] Implement pcc compile engine frontend of trivial fusion.

  #72640 merged May 13, 2025

• [XPU] fix index's datatype, using int64 instead of int, part 2 (g-z)

  #72519 merged May 13, 2025

• [ThirdParty] Bump zlib version to v1.3.1 (re-apply #72524)

  #72675 merged May 12, 2025

• [Auto Parallel] Add spmd rule No.1 for topk and topk_grad ops

  #72499 merged May 12, 2025

• [CINN] Fixed arange float16 and bfloat16 support

  #72669 merged May 12, 2025

• [Science] Fix dpa bug

  #72635 merged May 12, 2025

• [CI] xpu download

  #72642 merged May 12, 2025

• 【Paddle Tensor 规范化第二期】elementwise_pow support complex

  #72518 merged May 12, 2025

• Fix example output in amp/debugging.py

  #72656 merged May 12, 2025

• 【XPU】update_xft_to_0507

  #72610 merged May 12, 2025

• Update base/framework.py

  #72661 merged May 12, 2025

• Update python/paddle/base/backward.py

  #72660 merged May 12, 2025

• Fix parameter name in dygraph/base.py

  #72658 merged May 12, 2025

• Update tensor_patch_methods.py

  #72659 merged May 12, 2025

• Fix typo in grad_scaler.py

  #72657 merged May 12, 2025

• [CI] Add WITH_XPU_FFT

  #72653 merged May 12, 2025

• [PIR slice] Optimize bool index logic for setitem and getitem

  #72644 merged May 12, 2025

• Use gather to replace gather nd when the bool index is one-dimensional

  #72625 merged May 12, 2025

• [PHI] Optimize GatherNdCUDAKernel

  #72623 merged May 12, 2025

• [CI] coverage download

  #72632 merged May 12, 2025

• [PHI] Fixed paddle.var & paddle.std float16 overflow problem

  #72650 merged May 12, 2025

• [cherry-pick][Distributed] destroy && recreate ncclComm (#72626)

  #72648 merged May 10, 2025

• [ThirdParty] Remove unused absl submodule

  #72633 merged May 9, 2025

• [Distributed] destroy && recreate ncclComm

  #72626 merged May 9, 2025

• [CINN] Fix WhileOp::InferSymbolicShape

  #72630 merged May 9, 2025

• [[Distributed] fix sp layer dist info

  #72470 merged May 9, 2025

• [CINN] fix broadcast vectorize tiling bug

  #72621 merged May 9, 2025

• [PHI][CINN] Align linalg.solve kernel with torch

  #72608 merged May 9, 2025

• [xpu] add weight_only_linear

  #72602 merged May 9, 2025

• 为paddle.less_than、paddle.less_equal、paddle.greater_than、paddle.greater_equal添加复数类型支持

  #72619 merged May 9, 2025

• [CI] fix develop build docker

  #72609 merged May 9, 2025

• Fix typo FlashAttention

  #72627 merged May 9, 2025

• [Auto Parallel] Add spmd rule No.6 for mean_all and mean_all_grad ops.

  #72479 merged May 9, 2025

• 【infer】cuda check for block_attn dynamic smem allocation

  #72605 merged May 9, 2025

• [CINN] CINN supports arange for fusion

  #72598 merged May 9, 2025

• [Docs] Fix docs for randn and rand_like

  #72549 merged May 9, 2025

• [AutoParallel] send/recv_object_list function and serialize method for placement object

  #72098 merged May 8, 2025

• [AP] Add paddle.cc.ap.FacadeOp: a custom op machanism for ap pass only

  #72525 merged May 8, 2025

• [XPU] UpdatePadding has already modified the value of pad

  #72618 merged May 8, 2025

• To Support CUDA kernel CustomDevice

  #72604 merged May 8, 2025

• [Typing] update pybind11 stub

  #72553 merged May 8, 2025

• [CINN] Limit merge non related group with large program

  #72550 merged May 8, 2025

• Fix typos 4-d

  #72591 merged May 8, 2025

• [CINN] Update cinn jit data

  #72510 merged May 8, 2025

• fix result shape of __ne__ and __eq__ when comparing with complex scalar

  #72611 merged May 8, 2025

• [CINN] Fix the bug in the preloaded tensor which has a write dependency.

  #72365 merged May 7, 2025

• [CINN]Delete the dependency of absl

  #72596 merged May 7, 2025

• [PHI] Preliminary fix for elementwise broadcast int32 shape overflow

  #72584 merged May 7, 2025

• Fix maximum and minimum for Nan input

  #71933 merged May 7, 2025

• 【Inference】fix moe tune

  #72597 merged May 7, 2025

• Fix typos <=

  #72590 merged May 7, 2025

• [CI] Add build and static-check

  #72537 merged May 7, 2025

• [CI] single coverage workflow

  #72586 merged May 7, 2025

• ResizeAndAllocate API fix place release bug

  #72587 merged May 7, 2025

• [XPU]add support for XPUFFT

  #72477 merged May 7, 2025

• 【CINN】Update indexexpr verify condition

  #72569 merged May 7, 2025

• [Inference]Support control trt_glog_info in PIR

  #72429 merged May 7, 2025

• 【Inference】update moe ffn config

  #72583 merged May 7, 2025

• [XPU] fix index's datatype, using int64 instead of int, part 1 (a-f)

  #72431 merged May 7, 2025

• [NPU] release GIL in constant folding pass

  #72561 merged May 7, 2025

• [SOT][NumPy] Fix TypeError: unhashable type that occurs when attempting to construct a NumPyAPIVariable

  #72581 merged May 6, 2025

• [SOT][CodeStyle] Replace Numpy with NumPy

  #72580 merged May 6, 2025

• [Auto Parallel] Add there ops spmd_rule(argmin, abs, cosh)

  #72264 merged May 6, 2025

• [PHI][CINN] Fix cum kernel for big tensor

  #72562 merged May 6, 2025

• Change the dependency of absl

  #72546 merged May 6, 2025

• update release disable test

  #72563 merged May 6, 2025

35 Pull requests opened by 24 people

• [Distributed] Support overlapped offload for refine recompute

  #72594 opened May 7, 2025

• Add _moe_unzip op

  #72600 opened May 7, 2025

• [WIP] support fa3 varlen

  #72606 opened May 7, 2025

• Update cub to v2.1.0

  #72607 opened May 7, 2025

• pr test by zhaopeisen

  #72613 opened May 8, 2025

• Support co shard

  #72614 opened May 8, 2025

• Update cub to version 2.1.0

  #72620 opened May 8, 2025

• [Distributed] Support custom state setter and getter for recompute to save custom states

  #72622 opened May 8, 2025

• [PHI][CINN] Fix grid sample kernel for big tensor

  #72628 opened May 8, 2025

• [Science] Fix DPA bug

  #72629 opened May 8, 2025

• [SOT] Mark `item` to breakgraph && add unittest of tensor array related APIs

  #72631 opened May 8, 2025

• [PHI] Fix shape int overflow for flatten/split/expand_as

  #72634 opened May 9, 2025

• [Auto Parallel] Fix enable fused linear param grad add in dynamic auto mode

  #72636 opened May 9, 2025

• [API] fix paddle.diag with big tensor

  #72638 opened May 9, 2025

• [PHI] Fix adaptivate pool2d kernel for big tensor

  #72639 opened May 9, 2025

• [CI] clone third party checkout 'Needed a single revision'

  #72641 opened May 9, 2025

• State dict

  #72645 opened May 9, 2025

• [SOT][DynamicShape] Force break for symbolic variable when dispatch not found

  #72646 opened May 9, 2025

• [CI] Add Api-Benchmark

  #72647 opened May 9, 2025

• [XPU] add `complex`,`real`,`imag` op for xpu

  #72649 opened May 9, 2025

• [WIP][API] fix paddle.dist with big tensor

  #72651 opened May 9, 2025

• [API] fix paddle.cross with big tensor

  #72652 opened May 9, 2025

• Update strided copy kernel

  #72662 opened May 11, 2025

• tmp

  #72663 opened May 11, 2025

• [PHI] Optimize GPUScatterNdAdd

  #72666 opened May 12, 2025

• CUDA Kernel on Custom Device - FIX 1

  #72668 opened May 12, 2025

• [Distributed] Support custom state setter and getter for recompute to save custom states

  #72670 opened May 12, 2025

• [CINN]Add remove_redundant_full_ops_pass

  #72671 opened May 12, 2025

• [CI] docker first

  #72673 opened May 12, 2025

• [PIR slice]Add 0-d index slice tests and remove single bool getitem branch

  #72676 opened May 12, 2025

• [Auto Parallel] Add spmd_rule for backward ops with one arg

  #72677 opened May 12, 2025

• [WIP][API] fix paddle.bmm api with big tensor

  #72679 opened May 12, 2025

• [CI] Add Api-benchmark GPU

  #72680 opened May 12, 2025

• [PHI] Fixed logsumexp precision problem

  #72681 opened May 13, 2025

• [CINN] Denied arange op for x86 end codegen

  #72682 opened May 13, 2025

4 Issues closed by 4 people

• 【严重数值错误，危险】paddlepaddle 3.0.0 py310_cpu_windows 版本，Tensor.reshape().cast_()以后，数值居然会发生变化。厉害了。这个神奇现象Bug我花了2天定位到它。

  #72654 closed May 11, 2025

• 大佬们，如果子模块更新出错，如何让它和远程保持一致呢？找了好多命令，都没有用呢

  #72573 closed May 7, 2025

• 安装paddle后，import paddle 打印“信息: 用提供的模式无法找到文件”

  #72526 closed May 6, 2025

• OSError: (External) CUBLAS error(7)

  #54701 closed May 6, 2025

18 Issues opened by 18 people

• 增加 3.0版本的 C 推理库

  #72678 opened May 12, 2025

• paddlepaddle-gpu 3.0.0 与 CUDA12.6 无法通过 run_check()测试

  #72674 opened May 12, 2025

• Support for rtx 50 series gpu

  #72672 opened May 12, 2025

• 【开源任务】Paddle CPU/GPU Kernel 精度问题推全

  #72667 opened May 12, 2025

• cpu后端是否支持AMD或者Hygon BLAS库

  #72665 opened May 12, 2025

• paddlepaddle 3.0.0 py310_cpu_windows 版本，cpu版本下Tensor.reshape().cast_()数值异常

  #72664 opened May 11, 2025

• raise Exception( Exception: Process 0 terminated with exit code 1.

  #72655 opened May 10, 2025

• 【快乐开源】Paddle2Torch API 转换功能开发

  #72643 opened May 9, 2025

• 【开源任务】 Paddle API 0-size 机制建设

  #72637 opened May 9, 2025

• 飞桨自动并行实战-入门/进阶任务

  #72624 opened May 8, 2025

• 无法安装paddlenlp

  #72616 opened May 8, 2025

• #关于PaddleMIX的一些能力完善优化建议

  #72612 opened May 8, 2025

• PyLayer.backward方法中需不需要对不需要梯度的参数对应返回None占位

  #72601 opened May 7, 2025

• MSVC 编译错误 C2593: 在 elementwise\grad\_kernel\_impl.h 中 bfloat16 和 double 的 operator* 不明确

  #72593 opened May 6, 2025

• CastOpTransformFunctor 和 CastDataTypeFunctor 是不是在vs2022下会报错？

  #72592 opened May 6, 2025

• ModuleNotFoundError: No module named 'paddle.base.dataloader'

  #72589 opened May 6, 2025

• 哪里可以下载NO_AVX512 paddlepaddle==3.0.0 cuda118 python3.10的包啊？

  #72588 opened May 6, 2025

• ├─ paddlepaddle-cuda 11.8** is not installable because it requires

  #72585 opened May 6, 2025

58 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• [API] `paddle.slogdet` 返回值规范化

  #72505 commented on May 8, 2025 • 4 new comments

• paddle.cumsum support 0-size tensor

  #72532 commented on May 13, 2025 • 3 new comments

• [SOT][FasterGuard] `ShapeMatchGuard` support dynamic shape

  #72564 commented on May 12, 2025 • 3 new comments

• 【快乐开源】Paddle Tensor 规范化二期 test_inv&test_linalg_matmul

  #72574 commented on May 12, 2025 • 2 new comments

• 【Paddle Tensor 第二期 API支持 0-size Tensor】Fix frobenius_norm to support 0-size tensor

  #72570 commented on May 9, 2025 • 2 new comments

• [Auto Parallel] Add spmd_rule for 11 ops.

  #72312 commented on May 12, 2025 • 2 new comments

• Lazy init decorator

  #72398 commented on May 12, 2025 • 1 new comment

• [SOT][FasterGuard] add `ExprNodeBase` cleanup func

  #72552 commented on May 10, 2025 • 1 new comment

• [SOT][Exception][3.10-][3.13] Add exception handler for Py3.10-

  #72559 commented on May 7, 2025 • 1 new comment

• try to fix fused gemm api

  #72473 commented on May 10, 2025 • 0 new comments

• Schedules

  #72480 commented on May 10, 2025 • 0 new comments

• win11 test

  #72481 commented on May 10, 2025 • 0 new comments

• Fix Cmake 4.0

  #72486 commented on May 6, 2025 • 0 new comments

• [AP] Remove pcc apis from `__all__` temporarily.

  #72469 commented on May 10, 2025 • 0 new comments

• [SOT] Add new internal API `paddle.jit.skip_transform` to mark an API as unified in dynamic and static mode and support custom op

  #72466 commented on May 10, 2025 • 0 new comments

• try to fix fused gemm api

  #72488 commented on May 10, 2025 • 0 new comments

• [Auto Parallel] add main_grad for sharding in auto dy

  #72493 commented on May 13, 2025 • 0 new comments

• [SOT][DynamicShape] remove TypeError catch when fallback

  #72464 commented on May 10, 2025 • 0 new comments

• [AP] Add documentation of pcc apis.

  #72506 commented on May 10, 2025 • 0 new comments

• [Auto Parallel] add FLAGS_enable_tensor_fusion

  #72508 commented on May 13, 2025 • 0 new comments

• [XPU] update xhpc to impove performance of strided_copy

  #72513 commented on May 8, 2025 • 0 new comments

• [Cherry-Pick] Add hybrid expert parallel communication group

  #72520 commented on May 10, 2025 • 0 new comments

• supports fa3_varlen

  #72539 commented on May 12, 2025 • 0 new comments

• [AutoParallel] support auto dp comm

  #72540 commented on May 13, 2025 • 0 new comments

• 【Hackathon 8th No.3】del oldIR in engine -part

  #72542 commented on May 10, 2025 • 0 new comments

• 【Hackathon 8th No.3】del oldIR in api -part

  #72548 commented on May 9, 2025 • 0 new comments

• [Auto Parallel] Add tensor_fusion and overlap in auto dy sharding

  #72551 commented on May 13, 2025 • 0 new comments

• Implement paddle.min/paddle.max for 0-size tensors along non-zero dimensions

  #72565 commented on May 13, 2025 • 0 new comments

• remove redundant code

  #72567 commented on May 8, 2025 • 0 new comments

• 在maxpool函数中加入dilation参数

  #72579 commented on May 13, 2025 • 0 new comments

• 安装paddlepaddle3.0-gpu报错

  #71987 commented on May 6, 2025 • 0 new comments

• 加载远程模型文件导致RCE

  #72575 commented on May 7, 2025 • 0 new comments

• PaddleInference 源码编译报错: fatal error LNK1248: 映像大小(1002F8BB5)超过允许的最大大小(FFFFFFFF)

  #55550 commented on May 7, 2025 • 0 new comments

• Add `paddlepaddle-gpu` aarch64 binaries to support NVIDIA Jetson devices

  #72578 commented on May 7, 2025 • 0 new comments

• PaddleTS的DeepAR模型 在用paddle inference推理的时候，PreconditionNotMetError: Tensor holds no memory. Call Tensor::mutable_data firstly.

  #66187 commented on May 8, 2025 • 0 new comments

• 实测活动：飞桨框架 3.0自动并行实战！官方case开源代码直给、官方驻群答疑为你护航、不同难度任务体系任你自选，更有现金奖学金等你来拿！

  #72432 commented on May 8, 2025 • 0 new comments

• Reason: cannot convert from 'const paddle::bfloat16' to 'const phi::dtype::complex<double>'

  #72576 commented on May 9, 2025 • 0 new comments

• Paddle主框架文档修复任务，速来！

  #71203 commented on May 9, 2025 • 0 new comments

• 【开源任务】算子切分推导规则开发，支持更多模型使用自动并行，简化更多用户的分布式开发成本

  #72415 commented on May 9, 2025 • 0 new comments

• 有计划提供cuda 12.4、12.8的镜像或安装包吗？

  #72523 commented on May 9, 2025 • 0 new comments

• `paddlepaddle` 包污染 `logging`，导致功能不可用

  #57165 commented on May 11, 2025 • 0 new comments

• paddle_inference.dll never works on windows 10

  #72377 commented on May 12, 2025 • 0 new comments

• 【HACKATHON 预备营】飞桨启航计划集训营（第五期）

  #71491 commented on May 12, 2025 • 0 new comments

• 【Hackathon 8th】开源贡献个人挑战赛

  #71310 commented on May 12, 2025 • 0 new comments

• [Inference] Support group-wize quantize for weight_quantize op in GPU

  #71549 commented on May 13, 2025 • 0 new comments

• 【Hackathon 8th No.3】 clean oldIR for engine

  #71929 commented on May 10, 2025 • 0 new comments

• [AutoParallel] Add take_along_axis spmd rules

  #72063 commented on May 7, 2025 • 0 new comments

• Add mac ci actions

  #72115 commented on May 13, 2025 • 0 new comments

• PipelineStage

  #72155 commented on May 10, 2025 • 0 new comments

• [AutoParallel] Batch data processing of pp

  #72179 commented on May 12, 2025 • 0 new comments

• 【CINN】simplify min max

  #72311 commented on May 10, 2025 • 0 new comments

• Test K100-AI

  #72331 commented on May 12, 2025 • 0 new comments

• Pr searlize placement

  #72374 commented on May 10, 2025 • 0 new comments

• [spmd_rules] Add SPMD rules for ArgSort operator

  #72388 commented on May 9, 2025 • 0 new comments

• api_tracer optimize

  #72421 commented on May 10, 2025 • 0 new comments

• [CINN] add _moe_unzip op

  #72425 commented on May 10, 2025 • 0 new comments

• Pr lazy init

  #72439 commented on May 10, 2025 • 0 new comments

• Add moe_zip op

  #72442 commented on May 10, 2025 • 0 new comments