ó Α®iàãó•%SSKJr SSKJrJrJr SSKrSSKJr SSKJ r \(a"SSK J r SSKJ r SSK Jr SS KJr \S rS \S '"S S5rSrSr\R,SS4SSjjrSSjrg)é)Ú annotations)Ú TYPE_CHECKINGÚClassVarÚLiteralN)Ú framework)Ústream)Ú TypeAlias)ÚTensor)Útask)ÚGroup)réééér Ú _ReduceOpcó`•\rSrSr%SrSrS\S'SrS\S'S rS \S 'S r S \S'Sr S\S'Sr g)ÚReduceOpé!a² Specify the type of operation used for element-wise reductions. It should be one of the following values: ReduceOp.SUM ReduceOp.MAX ReduceOp.MIN ReduceOp.PROD Examples: .. code-block:: python >>> # doctest: +REQUIRES(env: DISTRIBUTED) >>> import paddle >>> import paddle.distributed as dist >>> dist.init_parallel_env() >>> if dist.get_rank() == 0: ... data = paddle.to_tensor([[4, 5, 6], [4, 5, 6]]) >>> else: ... data = paddle.to_tensor([[1, 2, 3], [1, 2, 3]]) >>> dist.all_reduce(data, op=dist.ReduceOp.SUM) >>> print(data) >>> # [[5, 7, 9], [5, 7, 9]] (2 GPUs) rzClassVar[Literal[0]]ÚSUMr zClassVar[Literal[1]]ÚMAXrzClassVar[Literal[2]]ÚMINrzClassVar[Literal[3]]ÚPRODrzClassVar[Literal[4]]ÚAVG©N) Ú__name__Ú __module__Ú __qualname__Ú__firstlineno__Ú__doc__rÚ__annotations__rrrrÚ__static_attributes__róÚg/var/www/html/banglarbhumi/venv/lib/python3.13/site-packages/paddle/distributed/communication/reduce.pyrr!s@‡ñð<!"€CÐ Ó!Ø !€CÐ Ó!Ø !€CÐ Ó!Ø!"€DÐ Ó"Ø !€CÐ Ö!r"rcóP•U[R:Xa$[RRR$U[R:Xa$[RRR$U[R :Xa$[RRR $U[R :Xa$[RRR$U[R:Xa$[RRR$[SUS35e)NzUnknown reduce_op type for Ú.) rrrÚcorerrrÚPRODUCTrÚ ValueError)Ú reduce_ops r#Ú_get_reduce_opr*GsÁ€Ø”H—L‘LÓ Ü~‰~×&Ñ&×*Ñ*Ð*Ø ”h—l‘lÓ "Ü~‰~×&Ñ&×*Ñ*Ð*Ø ”h—l‘lÓ "Ü~‰~×&Ñ&×*Ñ*Ð*Ø ”h—m‘mÓ #Ü~‰~×&Ñ&×.Ñ.Ð.Ø ”h—l‘lÓ "Ü~‰~×&Ñ&×*Ñ*Ð*ä Ð2°9°+¸QÐ?Ó @Ð@r"có •US3$)NÚ_r)Úop_names r#Ú_to_inplace_opr.Vs€ØˆYaˆ=Ðr"Tc ó^•U[R:Xa[5(drUc([RR R 5OUnURSUR- 5 [R"UU[RUUSS9$[R"UUUUUSS9$)a| Reduce a tensor to the destination from all others. As shown below, one process is started with a GPU and the data of this process is represented by its group rank. The destination of the reduce operator is GPU0 and the process is sum. Through reduce operator, the GPU0 will owns the sum of all data from all GPUs. .. image:: https://githubraw.cdn.bcebos.com/PaddlePaddle/docs/develop/docs/api/paddle/distributed/img/reduce.png :width: 800 :alt: reduce :align: center Args: tensor (Tensor): The output Tensor for the destination and the input Tensor otherwise. Its data type should be float16, float32, float64, int32, int64, int8, uint8, bool or bfloat16. dst (int): The destination rank id. op (ReduceOp.SUM|ReduceOp.MAX|ReduceOp.MIN|ReduceOp.PROD|ReduceOp.AVG, optional): The operation used. Default value is ReduceOp.SUM. group (Group|None, optional): The group instance return by new_group or None for global default group. sync_op (bool, optional): Whether this op is a sync op. The default value is True. Returns: Return a task object. Examples: .. code-block:: python >>> # doctest: +REQUIRES(env: DISTRIBUTED) >>> import paddle >>> import paddle.distributed as dist >>> dist.init_parallel_env() >>> if dist.get_rank() == 0: ... data = paddle.to_tensor([[4, 5, 6], [4, 5, 6]]) >>> else: ... data = paddle.to_tensor([[1, 2, 3], [1, 2, 3]]) >>> dist.reduce(data, dst=0) >>> print(data) >>> # [[5, 7, 9], [5, 7, 9]] (2 GPUs, out for rank 0) >>> # [[1, 2, 3], [1, 2, 3]] (2 GPUs, out for rank 1) gð?F)ÚdstÚopÚgroupÚsync_opÚuse_calc_stream)rrÚis_avg_reduce_op_supportedÚpaddleÚ distributedÚ collectiveÚ_get_global_groupÚscale_ÚnranksrÚreducerÚ is_memberÚidÚget_group_rankrrrÚ_C_opsr()Útensorr0r1r2r3r4Úring_idÚgdsts r#r<r<Zs¦€ð^ ŒX\‰\ÓÔ#=×#?Ñ#?ð‰}ô × Ñ × )Ñ )× ;Ñ ;Ô =àð ð  ‰ c˜EŸL™LÑ(Ô)Ü}Š}Ø ØÜ|‰|ØØØ!ñ  ð ô =Š=ØØ Ø ØØØñ  ðr"cóŽ•[R"5(a+[RRR 5S:¬$g)NiRF)r6Úis_compiled_with_cudaÚbaser&Ú nccl_versionrr"r#r5r5¾s2€Ü ×#Ò#×%Ñ%Ü{‰{×Ñ×,Ñ,Ó.°%Ñ7Ð7àr") rAr r0Úintr1rr2z Group | Noner3ÚboolÚreturnr )rJrI)Ú __future__rÚtypingrrrr6rÚ paddle.distributed.communicationrÚtyping_extensionsr r Úpaddle.base.corer Ú&paddle.distributed.communication.groupr rr rr*r.rr<r5rr"r#ÚrQs¦ðö#ç3Ñ3ã ÝÝ3æÝ+åÝ%Ý<à" =Ñ1€IˆyÓ1÷#"ñ#"òL Aòð—L‘LØØð a6Ø ða6à ða6ð ða6ð ð a6ð ð a6ð  õ a6õHr"