ó Α®iÁãó*•SSKJr SSKJr SSKJrJr SSKJr SSK J r J r SSK J r JrJr SSKJr \(aSSKJr SS KJr SS K Jr S S KJr SS jrSSjr\ R4SSS4SSjjrg)é)Ú annotations)Ú TYPE_CHECKING)Ú_C_opsÚ framework)Ú data_feeder)Ú_get_global_groupÚ_warn_cur_rank_not_in_group)ÚReduceOpÚ_get_reduce_opÚ_to_inplace_op)Ú in_pir_mode)ÚTensor)Útask)ÚGroupé)Ú _ReduceOpcóÈ•[U5nU(aURRX5$URRXU5nU(aUR"5 U$)N)r Ú process_groupÚall_reduce_on_calc_streamÚ all_reduceÚwait)ÚtensorÚopÚgroupÚsync_opÚuse_calc_streamÚop_typers Úr/var/www/html/banglarbhumi/venv/lib/python3.13/site-packages/paddle/distributed/communication/stream/all_reduce.pyÚ_all_reduce_in_dygraphr'sQ€ô˜RÓ €GæØ×"Ñ"×<Ñ<¸VÓMÐMà × Ñ × )Ñ )¨&¸7Ó C€DÞØ Š Œ à €Kócóˆ•[R"US/SQS5 UcSO URn[U[5(d [ S5e[ 5(a#[U5n[R"XU5 g[U5n[R"U40[5D6nURUSU/0SU/0XSS.S 9 g) Nr) Úfloat16Úfloat32Úfloat64Úint32Úint64Úint8Úuint8ÚboolÚuint16rrz3The type of 'ring_id' for all_reduce should be int.ÚXÚOut)Úring_idr)ÚtypeÚinputsÚoutputsÚattrs)rÚcheck_variable_and_dtypeÚidÚ isinstanceÚintÚ ValueErrorr r rÚ all_reduce_r rÚ LayerHelperÚlocalsÚ append_op)rrrrrr-rÚhelpers rÚ_all_reduce_in_static_moder<:s€ô×(Ò(ØØò ð ôð"‘=‰a e§h¡h€Gä gœs× #Ñ #ÜÐNÓOÐOä‡}}Ü% bÓ)ˆÜ×Ò˜6¨BÔ/Øô˜RÓ €GÜ × "Ò " 7Ñ 7¬f«hÑ 7€FØ ×ÑØ ØfXˆØ˜˜Ð!Ø!Ñ>ð òr NTFcóð•[U5(agU(dU(a [S5e[R"5(aUc [ 5OUn[ XX#U5$UbS5e[ XX#U5$)aL Perform specific reduction (for example, sum, max) on inputs across devices. Args: tensor (Tensor): The input tensor on each rank. The result will overwrite this tenor after communication. Support float16, float32, float64, int32, int64, int8, uint8 or bool as the input data type. op (ReduceOp.SUM|ReduceOp.MAX|ReduceOp.MIN|ReduceOp.PROD, optional): The reduction used. If none is given, use ReduceOp.SUM as default. group (Group|None, optional): Communicate in which group. If none is given, use the global group as default. sync_op (bool, optional): Indicate whether the communication is sync or not. If none is given, use true as default. use_calc_stream (bool, optional): Indicate whether the communication is done on calculation stream. If none is given, use false as default. This option is designed for high performance demand, be careful to turn it on except you are clearly know its meaning. Returns: Return a task object. Examples: .. code-block:: python >>> # doctest: +REQUIRES(env: DISTRIBUTED) >>> import paddle >>> import paddle.distributed as dist >>> dist.init_parallel_env() >>> local_rank = dist.get_rank() >>> data = None >>> if local_rank == 0: ... data = paddle.to_tensor([[4, 5, 6], [4, 5, 6]]) >>> else: ... data = paddle.to_tensor([[1, 2, 3], [1, 2, 3]]) >>> task = dist.stream.all_reduce(data, sync_op=False) >>> task.wait() # type: ignore[union-attr] >>> out = data >>> print(out) [[5, 7, 9], [5, 7, 9]] Nz5use_calc_stream can only be true in sync op behavior.z3Group can not be used in static graph mode for now.)r Ú RuntimeErrorrÚin_dynamic_moderrr<)rrrrrs rrrhsˆ€ôV# 5×)Ñ)Øæ –ÜØ Có ð ô× Ò ×"Ñ"Ø',¡}Ô!Ô#¸%ˆÜ%Ø ˜¨ó ð ð‰}ð Ø Aó ˆ}ô*Ø ˜¨ó ð r ) rrrrrrrr)rr)Úreturnr) rrrrrrrr)rr)r@ÚNone) rrrrrz Group | Nonerr)rr)r@z task | None)Ú __future__rÚtypingrÚpaddlerrÚ paddle.baserÚ&paddle.distributed.communication.grouprr Ú'paddle.distributed.communication.reducer r r Úpaddle.frameworkr rÚpaddle.base.corerrrrrr<ÚSUM©r rÚrLsðõ#å ç$Ý#÷÷ñõ )æÝÝ%Ý<å&ðØ ðàðð ððð ð ð ð  ô ð&+Ø ð+àð+ð ð+ðð +ð ð +ð  ô +ð`—L‘LØØØ!ð > Ø ð> àð> ð ð> ðð > ð ð > ð ö > r