102 Pull requests merged by 43 people

• [Dy2St] Add place hash to scope cache key to avoid conflict with executor cache

  #71505 merged Mar 9, 2025

• [Dy2St] Clear InplaceMap after program is completed

  #71503 merged Mar 8, 2025

• [XPU] fix xpu pp bug

  #71500 merged Mar 8, 2025

• [CINN] Fix broadcast loop axis mapping

  #71482 merged Mar 8, 2025

• [SOT]Support dynamic shape for infermeta of BroadcastTensors

  #71499 merged Mar 8, 2025

• [Dy2St] Update no_need_buffer names if original_name no need buffer

  #71476 merged Mar 8, 2025

• [SOT] Optimize env and subgraph information printing

  #71460 merged Mar 8, 2025

• [CINN] Fix strided_slice op infer symbolic interface bug

  #71486 merged Mar 7, 2025

• [XPU] fix error in conv3d_transpose test

  #71492 merged Mar 7, 2025

• [AutoParallel] local layer api adapts to pp

  #71373 merged Mar 7, 2025

• [XPU] feat: support xpu ipc and stream to enable zero cost checkpoint

  #71178 merged Mar 7, 2025

• fix swgilu matmul 0 size tensor bug

  #71442 merged Mar 7, 2025

• [XPU] add conv3d_transpose

  #71451 merged Mar 7, 2025

• [CINN] Optimize the compare func of Div expressions

  #71465 merged Mar 7, 2025

• [SOT] Rename self.hold to self.holds in iter.py

  #71473 merged Mar 7, 2025

• [SOT] Split InlineCallBreak

  #71458 merged Mar 7, 2025

• Modify Hook_intermidiate to Hook_intermediate

  #71447 merged Mar 7, 2025

• 【pir backward】Passing the gradient output of tuple_pop between if_grad_grad and if_grad

  #71445 merged Mar 7, 2025

• [XPU] Support print runtime error log for xdnn/xfa/xpudnn error

  #71431 merged Mar 7, 2025

• Fix logsumexp

  #71454 merged Mar 7, 2025

• [XPU] fix bugs of depthwise conv test and change default quant type

  #70859 merged Mar 7, 2025

• [XPU] fix batch_norm_grad when use global status

  #71423 merged Mar 7, 2025

• 【Paddle TensorRT】fix PaddleX pir-trt 3

  #71287 merged Mar 7, 2025

• [CINN] Fix block reduce CUDA template

  #71471 merged Mar 7, 2025

• [PIR saveload]Fix bug

  #71452 merged Mar 7, 2025

• support dynamic shape in pp scheduler (#71123)

  #71290 merged Mar 7, 2025

• [CINN] fix scalar tensor cast to local buffer type bug

  #71455 merged Mar 7, 2025

• [CINN] Fix common factor extraction of EntailLoopConditionPass

  #71466 merged Mar 7, 2025

• Revert "[CINN] Fix fp32 OOM in some models caused by too much unfused gemm epilogues."

  #71468 merged Mar 7, 2025

• 【save_load】fix Jit load func_name bug

  #71457 merged Mar 7, 2025

• [CINN] Insert if and inline compute for repeatedly accessed global var

  #71380 merged Mar 7, 2025

• Fix CUDA 12.8 compilation error

  #71404 merged Mar 7, 2025

• [Dy2St] Custom deepcopy behavior for WeakMethod to ensure hold correct instance

  #71463 merged Mar 7, 2025

• [SOT] make sure parameter holds in IterVariable.__init__() is a list

  #71437 merged Mar 6, 2025

• [XPU] Fix error for dynamic shape send recv

  #71425 merged Mar 6, 2025

• [CINN] add a skip for grad op and null value in cache check

  #71415 merged Mar 6, 2025

• Fix typos sptial spatial

  #71446 merged Mar 6, 2025

• [PHI] Update depthwise conv kernel

  #71403 merged Mar 6, 2025

• modify eigen patch linux & win

  #71419 merged Mar 6, 2025

• [CINN] Fix bug of Split on dynamic dim with more than 2 factors

  #71420 merged Mar 6, 2025

• [CINN] Increase SM resource utilization for grid reduce

  #71301 merged Mar 6, 2025

• 【CINN】Remove cas file

  #71374 merged Mar 6, 2025

• Refine license of adapted DeepEP files

  #71449 merged Mar 6, 2025

• add DeepEP intranode all-to-all

  #71358 merged Mar 6, 2025

• [CINN] Add fold_full_op pass

  #71443 merged Mar 6, 2025

• [CustomOp] Pop name option from kwargs to avoid duplicate name option

  #71438 merged Mar 6, 2025

• [CINN] Fix auto recompute bug

  #71436 merged Mar 6, 2025

• 【快乐开源】Paddle Tensor 规范化二期：array_api_tests/test_creation_functions.py::test_arange鲁棒性增强

  #71394 merged Mar 6, 2025

• [XPU] add quantize_linear and dequantize_linear op

  #71375 merged Mar 6, 2025

• [Inference] MoE Use macro to compress code impl

  #71396 merged Mar 6, 2025

• Fix typos divisble divisible

  #71428 merged Mar 6, 2025

• [SOT] Add PEP508LikeEnvironmentVariable

  #71430 merged Mar 6, 2025

• [CINN][new hardware] SYCL third PR: complete the SYCL logic

  #71204 merged Mar 6, 2025

• [PIR] Fix issue of no op_info when enabling oneDNN

  #71426 merged Mar 6, 2025

• [Paddle TensorRT] Naming of layers that is complementary to the rest of the converter

  #71354 merged Mar 5, 2025

• fix issue of expand pir build

  #71388 merged Mar 5, 2025

• [XPU] add isfinite/isinf support

  #71364 merged Mar 5, 2025

• [CINN] Fix shape mismatch in axis transform simulation

  #71405 merged Mar 5, 2025

• [AutoParallel] Fix pipeline visualization tool

  #71386 merged Mar 5, 2025

• Fix typos simplied simplified

  #71383 merged Mar 5, 2025

• [AutoParallel] Handle dynamic shape in InferGlobalShape and InferLocalShape

  #71320 merged Mar 5, 2025

• [CINN] Add InputOutputMaximumConstrain for Trivial Recompute

  #71408 merged Mar 5, 2025

• [Prim] Add index_select_double_grad

  #71352 merged Mar 5, 2025

• 【custom】add Custom pass list in LoadCustomRuntimeLib and analysis_predictor will use it in customplace

  #71362 merged Mar 5, 2025

• 【CINN】Remove cas head file --Part2

  #71369 merged Mar 5, 2025

• [XPU] reduce_xxx and broadcast_xxx use int64_t shape

  #71361 merged Mar 5, 2025

• [SOT] Standardize the from_iterator method of MapVariable and ZipVariable

  #71407 merged Mar 5, 2025

• [SOT] GraphLogger->SubGraphInfo to collect graph info

  #71412 merged Mar 5, 2025

• add win eigen patch

  #71414 merged Mar 5, 2025

• fix cuda12.6 linux

  #71363 merged Mar 5, 2025

• [CINN] fix vectorize info bug

  #71324 merged Mar 5, 2025

• 【CINN】Remove cas head file --Part1

  #71368 merged Mar 5, 2025

• 【CINN】Remove cas head file --Part0

  #71367 merged Mar 5, 2025

• 【PIR】fused_bn_add_act_pass set channel to align 4

  #71209 merged Mar 5, 2025

• [Serde] Ensure run save hook under dygraph mode

  #71400 merged Mar 5, 2025

• [xpu] support: each parameter has different lr in merged_momentum

  #71212 merged Mar 5, 2025

• [CINN] Remove redundant group output before divide to fusion ops

  #71401 merged Mar 5, 2025

• Revert "[AutoPrallel] Fix some bug"

  #71384 merged Mar 4, 2025

• 【CINN】Add simplify util

  #71376 merged Mar 4, 2025

• [CINN] Correct mistakes in using shape_or_data

  #71382 merged Mar 4, 2025

• 【CINN】Use ArithSimplify instead of Autosimplify --part4

  #71314 merged Mar 4, 2025

• [SOT] Fix lost error info

  #71385 merged Mar 4, 2025

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

  #71230 merged Mar 4, 2025

• support dict pp split point

  #71342 merged Mar 4, 2025

• [CINN] Optimize indivisble loops by condition entailment

  #71340 merged Mar 4, 2025

• [SOT] Support use iterable for call with varargs

  #71377 merged Mar 4, 2025

• [Prim][CINN] Use reshape to decompose squeeze grad

  #71387 merged Mar 4, 2025

• 【CINN】Move casInterval to integer_set

  #71381 merged Mar 4, 2025

• [CINN] Update group substitute_dimexpr_map for broadcast tree

  #71370 merged Mar 4, 2025

• [CINN] Add group substitute_dimexpr_map clone

  #71365 merged Mar 4, 2025

• [CINN] Fix bug of AnchorFusion

  #71349 merged Mar 4, 2025

• [CINN] preload scalar tensor for vectorize situation

  #71249 merged Mar 3, 2025

• 【CINN】Optimize use of simplify

  #71321 merged Mar 3, 2025

• [CINN] Init OriginalAttributesFilter before ShapeOptimizationPass

  #71348 merged Mar 3, 2025

• [CI] add job name

  #71353 merged Mar 3, 2025

• [SOT] Refactor MapVariable to align with Python

  #71346 merged Mar 3, 2025

• Fix typos shoulde should

  #71350 merged Mar 3, 2025

• 【CINN】Remove Ginac

  #71323 merged Mar 3, 2025

• [CINN] Add InferSymbolicShapeInterface for pd_op.strided_slice

  #70541 merged Mar 3, 2025

• Use ArithSimplify instead of Autosimplify --part3

  #71309 merged Mar 3, 2025

• 【Paddle Tensor 第二期 API鲁棒性增强】 paddle.where support bool

  #71238 merged Mar 3, 2025

• Fix cos/sin double_grad functor error when meets null ptr

  #71332 merged Mar 3, 2025

51 Pull requests opened by 39 people

• Modify the setup file to add cuda12.6

  #71351 opened Mar 3, 2025

• 【custom】add Custom xcclcommcontext init in new_executor

  #71357 opened Mar 3, 2025

• 【custom】support Custom device commcontext init and set in phi_api and support static process_group.py for createxcclcommcontext

  #71366 opened Mar 3, 2025

• test fp16 old-ir trt

  #71378 opened Mar 3, 2025

• [CINN] Remove reduce scale constraint for Trivial-Reduce AnchorFusion

  #71379 opened Mar 3, 2025

• add save pad_tensor feature"

  #71395 opened Mar 4, 2025

• [AutoParallel] remove loss shape assert

  #71397 opened Mar 4, 2025

• Paddle xpu chunk

  #71402 opened Mar 4, 2025

• [XPU] feat: implement xpu pinned memory and sync load with pinned memory

  #71409 opened Mar 4, 2025

• [CINN] Optimize nvrtc compile time

  #71410 opened Mar 4, 2025

• [XPU] support all_to_all_unequal_split_size

  #71411 opened Mar 4, 2025

• change openmp to threads for save bug

  #71418 opened Mar 5, 2025

• [AutoParallel] fix problems in ClipGradByNorm

  #71421 opened Mar 5, 2025

• add npu dockerfile

  #71422 opened Mar 5, 2025

• [XPU] support set numa affinity

  #71424 opened Mar 5, 2025

• [Distribution] Support DualPipeV

  #71427 opened Mar 5, 2025

• [XPU] add grid sampler support.

  #71432 opened Mar 5, 2025

• Support send/recv when mesh is pp_mesh

  #71433 opened Mar 5, 2025

• Add deepep internode implementations.

  #71435 opened Mar 5, 2025

• Local depthwise shm

  #71439 opened Mar 5, 2025

• [CINN] Reset CINN pass related singleton state after apply cinn pass

  #71441 opened Mar 5, 2025

• add new func of int64 to int32

  #71444 opened Mar 5, 2025

• paddle.distributed.all_to_all supports unequal split(#71429)

  #71448 opened Mar 6, 2025

• [Inference] MoE Code Clean Job

  #71456 opened Mar 6, 2025

• Add dtensor_idx in ShardDataloader

  #71459 opened Mar 6, 2025

• Optimize the op of c_softmax_with_cross_entropy

  #71461 opened Mar 6, 2025

• [CINN] TileBroadcastTactic NHWC layout broadcast

  #71464 opened Mar 6, 2025

• 【DCU】support dtk25 and put_along_axis

  #71467 opened Mar 6, 2025

• 4090

  #71469 opened Mar 6, 2025

• 【Paddle TensorRT】fix document

  #71470 opened Mar 6, 2025

• [CINN] add easy simplify for min

  #71472 opened Mar 6, 2025

• fix

  #71474 opened Mar 7, 2025

• [WIP][Dy2St] Restore patch forward after layer call to avoid side effect

  #71475 opened Mar 7, 2025

• Modify hook_test_intermidiate to hook_test_intermediate

  #71478 opened Mar 7, 2025

• Optimize the performance of NMSFast in the multiclass_nms3_kernel.

  #71479 opened Mar 7, 2025

• CI测试不review

  #71480 opened Mar 7, 2025

• fix cuda arch support for DeepEP

  #71481 opened Mar 7, 2025

• [CINN] Fix the out-of-bounds bug in the vectorize.

  #71483 opened Mar 7, 2025

• update

  #71485 opened Mar 7, 2025

• [Inference] remove useless code

  #71488 opened Mar 7, 2025

• [CINN]Add The TileDiscreteReductionTactic

  #71489 opened Mar 7, 2025

• Support XPU deepseek

  #71490 opened Mar 7, 2025

• [XPU]add full_with_tensor to xpu3 list

  #71493 opened Mar 7, 2025

• Fix DCU Build

  #71494 opened Mar 7, 2025

• support_adamw_moment1_bfloat16

  #71495 opened Mar 7, 2025

• [AutoParallel] Add flash_mask spmd

  #71496 opened Mar 7, 2025

• [AutoParallel] Add shape from_dtensor in input spec

  #71497 opened Mar 7, 2025

• [AutoParallel] Reshard master weights in dynamic mode

  #71498 opened Mar 7, 2025

• 【Paddle TensorRT】Pir-trt support TensorRT Refittable

  #71501 opened Mar 8, 2025

• 【快乐开源】Paddle Tensor 规范化二期 API 支持 0-size Tensor No.13-17：paddle.mean、paddle.sum、paddle.prod、paddle.var、paddle.std

  #71504 opened Mar 8, 2025

• [WIP][Dy2St] Move scope cache to cpp side

  #71506 opened Mar 8, 2025

15 Issues closed by 8 people

• 修改单测名称，报设备找不到错误

  #71487 closed Mar 8, 2025

• paddle3.0的安装方式

  #71398 closed Mar 7, 2025

• paddle动态图转静态图报错

  #71450 closed Mar 6, 2025

• paddlenlp/bin/../lib/libstdc++.so.6: version `GLIBCXX_3.4.30'

  #71372 closed Mar 5, 2025

• 怎么使用inference进行多卡推理呢

  #71355 closed Mar 5, 2025

• 【Hackathon 8th】开源贡献个人挑战赛（尝鲜版）

  #70746 closed Mar 5, 2025

• 使用AMP O2 模式训练模型，jit.save报错

  #71356 closed Mar 5, 2025

• 昇腾编译出现错误

  #71360 closed Mar 5, 2025

• 流水并行无法打印结果

  #71359 closed Mar 4, 2025

• Paddle2.6动转静

  #62185 closed Mar 4, 2025

• 使用Docker编译失败

  #61829 closed Mar 4, 2025

• risc-v芯片上编译paddle报错

  #61770 closed Mar 4, 2025

• RuntimeError: (NotFound) The kernel `assign_value` is not registered 数据加载报错

  #52287 closed Mar 4, 2025

• ft1500上编译paddle报错，Could NOT find PY_numpy (missing: PY_NUMPY)

  #40988 closed Mar 4, 2025

• /bin/../lib/libstdc++.so.6: version `GLIBCXX_3.4.30'

  #71371 closed Mar 4, 2025

7 Issues opened by 7 people

• CI中pre-commit错误

  #71502 opened Mar 8, 2025

• 【HACKATHON 预备营】飞桨启航计划集训营（第五期）

  #71491 opened Mar 7, 2025

• Reduce 操作中，重复维度可能导致错误规约

  #71477 opened Mar 7, 2025

• paddle.zeros的参数里面没有place

  #71462 opened Mar 6, 2025

• paddle.distributed.all_to_all不支持unequal_split_size的语义

  #71429 opened Mar 5, 2025

• 环境配置不兼容,执行出现段错误

  #71417 opened Mar 5, 2025

• 2张卡流水并行，显存都累积到了1号卡

  #71413 opened Mar 4, 2025

33 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.

• 【Hackathon 8th No.1】add `lu_solve` api for paddle

  #71030 commented on Mar 7, 2025 • 32 new comments

• [SOT][Faster Guard][3.13] add `ENV_SOT_ENABLE_STRICT_GUARD_CHECK`

  #71239 commented on Mar 7, 2025 • 24 new comments

• [PIR-Auto-Parallel] Add sync shared param pass

  #71167 commented on Mar 8, 2025 • 14 new comments

• [XPU] feat: add xpu async memory copy to enable zero cost checkpoint

  #71168 commented on Mar 7, 2025 • 9 new comments

• 【Paddle Tensor】Fix bugs related to converting unit tests about collect shape

  #71305 commented on Mar 6, 2025 • 2 new comments

• 【Paddle Tensor】fix converter old ir issues -1

  #70849 commented on Mar 7, 2025 • 1 new comment

• 【Paddle Tensor】Fix bugs related to converting unit tests of the old ir-trt into pir-trt

  #71083 commented on Mar 6, 2025 • 1 new comment

• Paddle cannot load a saved model file in Cambricon MLU370 Card

  #71331 commented on Mar 3, 2025 • 0 new comments

• Check tensorrt engin op

  #70652 commented on Mar 5, 2025 • 0 new comments

• 【CINN】longlong2int for dynamic shape

  #71186 commented on Mar 5, 2025 • 0 new comments

• Add fused_mt dybatch

  #71188 commented on Mar 7, 2025 • 0 new comments

• [WIP]【Paddle Tensor 规范化第二期】paddle.svd support complex and 0-size

  #71250 commented on Mar 5, 2025 • 0 new comments

• [feat]: add fused adaLN scale residual xpu kernel

  #71282 commented on Mar 7, 2025 • 0 new comments

• Test lu solve static

  #71285 commented on Mar 7, 2025 • 0 new comments

• 【CINN】Apply constraint in CINN backend

  #71306 commented on Mar 3, 2025 • 0 new comments

• [Inference]Fix PIR-TRT bugs in PaddleX Part-3

  #71319 commented on Mar 7, 2025 • 0 new comments

• [XPU] enable FLAGS_log_memory_stats in xpu

  #71334 commented on Mar 5, 2025 • 0 new comments

• ExternalError: CUBLAS error(15)

  #49519 commented on Mar 4, 2025 • 0 new comments

• Internal server error. Retry later.

  #62346 commented on Mar 4, 2025 • 0 new comments

• Padlle源码编译错误，基于docker，with CINN

  #70728 commented on Mar 5, 2025 • 0 new comments

• 【Hackathon 8th】Fundable Projects

  #71311 commented on Mar 6, 2025 • 0 new comments

• 【HACKATHON 8th Code Camp】飞桨正式实习招聘（可在校）

  #71313 commented on Mar 6, 2025 • 0 new comments

• Illegal instruction (core dumped)

  #69926 commented on Mar 6, 2025 • 0 new comments

• 【HACKATHON 8th Code Camp】黑客松护航计划集训营（正式批）

  #71312 commented on Mar 6, 2025 • 0 new comments

• 【快乐开源】Paddle Tensor 规范化二期

  #69908 commented on Mar 6, 2025 • 0 new comments

• paddle.logsumexp存在Bug，输入大shape时报cuda error 700错误

  #71225 commented on Mar 7, 2025 • 0 new comments

• 【Hackathon 8th】开源贡献个人挑战赛

  #71310 commented on Mar 7, 2025 • 0 new comments

• Paddle主框架文档修复任务，速来！

  #71203 commented on Mar 7, 2025 • 0 new comments

• 同时import paddle和torch报错找不到cudnn

  #66947 commented on Mar 7, 2025 • 0 new comments

• 发布 paddlepaddle-gpu 的 pre-release 版本

  #71007 commented on Mar 9, 2025 • 0 new comments

• [oneDNN] Upgrade oneDNN to v3.6

  #69386 commented on Mar 5, 2025 • 0 new comments

• [Paddle TensorRT] support TensorRT Refittable

  #70286 commented on Mar 7, 2025 • 0 new comments

• CI测试不review[fluid_ops] c_allreduce_sum 2

  #70348 commented on Mar 7, 2025 • 0 new comments