o "jm@sddlZddlmZddlZddlmZddlmZddlm Z m Z m Z ddl m Z mZdd Zd d Zd d ZddZ  d dddZdS)N)List)is_initialized)logger)LocalTensorIndexLocalTensorMetadataMetadata)%compute_local_shape_and_global_offsetflatten_state_dictcCsZt|}g}tj||||ddD]}||dks*Jd|d|dqdS)Nrrzkeys:z != first_keys: )listkeyspaddle distributedall_gather_object) state_dict process_groupZ local_keysZ gloabl_keysr rn/var/www/html/Deteccion_Ine/venv/lib/python3.10/site-packages/paddle/distributed/checkpoint/save_state_dict.pycheck_state_dicts rcCsfg}t|dddd}tj||||ddD]}||dks0Jd|d|qdS)N.r_rzid:z != all_unique_id[0]:)intsplitr rr) file_namerZ all_unique_id unique_ididrrrcheck_file_name'srcCsft|ts Jdi}|D]#}|D]\}}||vr*|||vr"q|||q|g||<qq |S)Nz'The global_state_dict should be a list.) isinstanceritemsappend)global_state_dict_metadataoutrkeyvalrrrmerge_state_dict_metadata3s  r$cCs6i}|D]}|D] \}}||vrq |||<q q|S)N)r)global_storage_metadatar!storage_metadatar"r#rrrdedup_storage_metadataCs r'returnc CstjjXt|tsJdt|}t|dkr,|D] }t|tj s+Jdqt j |s9t j |ddtjdkrBdnd}|rR|durRtsRtjd}d } tjd |d }t j t j ||snn|d7}qWtd ||rt||t||t}i} i} i} |D]e\} }t|tj r|sq|rt|jdkrt|j|jj|jj nd \} }| dus|durq|!}nt"|j} t|jdkrt"dgt|jnd}|}|| | <t#|| | | <|| t$| t"|<qg}g}|rtj%|| |tj%|| |n |&| |&| t'||_(t)||_*|tjkrDtd|t+|t j ||dtd| t+| t j ||WddS1scwYdS)a Save the state_dict of model to path. Args: state_dict(Dict[str, paddle.Tensor]): The state_dict to save. path(str): The directory to save state_dict. process_group(paddle.distributed.collective.Group): ProcessGroup to be used for cross-rank synchronization. Use the default process group which contains all cards. coordinator_rank(int): The rank used to save non distributed values. Rank0 is used by default. Examples: .. code-block:: python >>> # doctest: +SKIP('run in distributed mode') >>> import paddle >>> import paddle.distributed as dist >>> w1 = paddle.arange(32).reshape([4, 8]) >>> mesh = dist.ProcessMesh([0, 1]) >>> sharded_w1 = dist.shard_tensor(w1, mesh, [dist.Shard(0), dist.Replicate()]) >>> state_dict = {"w1": sharded_w1} >>> dist.save_state_dict(state_dict, "./checkpoint") >>> # doctest: -SKIP z&The state_dict should be a dictionary.rzGOnly support dygraph Tensor now, support static DistributedTensor laterT)exist_okrFNrz.distcpz file_name:)rrrz metadata:z .metadatazlocal_state_dict:),r baseZdygraphguardrdictr lenvaluesZTensorospathexistsmakedirsrZget_world_sizerZinit_parallel_envZget_rankjoinrdebugrrrrZ_is_initializedZis_distshaper Z dist_attrZ process_meshZ dims_mappingZ _local_valuetuplerrrrr$Zstate_dict_metadatar'r&save)rr1rZcoordinator_rankr#Zuse_distrrmetadataZlocal_state_dictZlocal_state_dict_metadataZlocal_storage_metadatar"Z local_shapeZ global_offsetZ local_tensorr r%rrrsave_state_dictMs              $r:)Nr)r(N)r0typingrr Z&paddle.distributed.communication.grouprZ'paddle.distributed.fleet.utils.log_utilrr9rrrutilsr r rrr$r'r:rrrrs