[Th? %SSKrSSKJrJrJr SSKrSSKJr SSK J r SSK J r SSK Jr SSKJr SSKJr SS KJrJrJrJrJrJrJr SS KJrJrJrJrJ r J!r! SS K"J#r#J$r$ S /r%\&\'\(S 'S\S\S\)4Sjr*S\&\S\&\S\&\4Sjr+S\RXS\4Sjr-S\ S\4Sjr.S\S\ S\4Sjr/S\'S\S\ 4Sjr0S\'S\S\ S\ 4Sjr1S\'S\RXS\ 4Sjr2S\'S\4S jr3S!r4S"r5S\'S#\S$\&\S\&\4S%jr6S&\S\4S'jr7S\'S(\S\&\ 4S)jr8S\S\4S*jr9S\RXS\&\4S+jr:S\'S,\S-\S\&\4S.jr;S&\<\'\4S\4S/jr=S0\S1\S2\S3\4S4jr>g)5N)AnyCallablecast)_get_device_module) ShardMetadata) ShardedTensor)DTensor)%compute_local_shape_and_global_offset)BytesStorageMetadataChunkStorageMetadata MetadataIndexSTATE_DICT_TYPE STORAGE_TYPESTensorPropertiesTensorStorageMetadata) LoadItemTypeReadItemSavePlanTensorWriteData WriteItem WriteItemType)"_check_shard_metadata_pair_overlap+_shards_get_overlap_region_wrt_saved_tensor create_read_items_for_chunk_list__all__plan other_planreturncURUR:wag[UR5[UR5:wag[URUR5GHnup#URUR:wa gUR nUR nUR UR :wd4URUR:wdUR UR :wa gURnURnU(aU(aU(d U(a gU(dMU(dMURUR:wa gURnURn U(aU (aU(d U (a gU(dGM.U (dGM8URU R:wdURU R:wdGMo g g)z Compare the two Save plans and return True if they are equal. Args: plan (SavePlan): First SavePlan to compare. other_plan (SavePlan): Second SavePlan to compare. Returns: True if the two plans are equal, False otherwise. FT) usablelenitemsziptypeindexfqnoffset tensor_datasizechunkoffsetssizes) rr plan_itemother_plan_itemplan_metadata_indexother_plan_metadata_indexr)other_tensor_datar+ other_chunks d/var/www/auris/envauris/lib/python3.13/site-packages/torch/distributed/checkpoint/planner_helpers.py_compare_save_plansr5'sq {{j''' 4::#j..//'*$**j6F6F&G" >>_11 1'oo$3$9$9! # #'@'D'D D"))-F-M-MM"((,E,K,KK ++ +77  1 1 ;,,#4#9#99 %%E+11Kk5[uMM[%8%88{{k&7&77 S'HV  cached_plans delta_plansc/n[X5HAup4U(a$UR(dURU5 M0URU5 MC U$)aG Merge a list of delta plans into a single plan. Args: cached_plans (List[SavePlan]): A list of cached plans. delta_plans (List[SavePlan]): A list of delta plans to merge. It can contain empty plans Returns: A single merged plan. If a delta plan is not usable, use the cached plan. Otherwise, use the delta plan. )r$r!append)r7r8 merged_plans cached_plan delta_plans r4_merge_delta_local_plansr>hsKL#&|#A j//    ,    + $B r6tensorc [[R"S/[UR 55-5UR 5S9$)Nrr,r-)r torchSizer"r*)r?s r4_create_chunk_from_tensorrDs5  A3V[[]!334FKKM r6shard_mdc[[R"UR5[R"UR5S9$NrA)r rBrC shard_offsets shard_sizes)rEs r4_chunk_for_shardrJs3  8112jj--. r6sharded_tensorc UR5Rn[URURUR UR URS9n[[U5UUR5RS9$)N)dtypelayout requires_grad memory_format pin_memoryr+ propertiesr*) metadatatensor_propertiesrrMrNrOrPrQrrJr*)rKrEshard_propertiesrSs r4_sharded_tensor_metadatarWs|&..0BB!$$&&&44&44#.. J x(  $ $ & + + r6r'c x[URURUR5up#[R "U5[R "U5p2[ [X5[R[[UUS9[R"UR55UR5S9S9$)NrArRr&r%r))r shape device_mesh placementsrBrCrrrSHARDrr rcreate_from_tensorto_localr*)r'r?r-r,s r4_create_write_items_for_dtensorr`s: f((&*;*;NEZZ& 7(;7 C)  #&(::6??;LM    r6c[R"UR5n[[ X5[ R [X5S9$)NrY)rBrCrHrrrr]rW)r'rKrEr,s r4_create_write_item_for_shardrbs>jj//0G C)  ,^F r6c ([R"S/[UR55-5n[ [ X5[ R[[X!R5S9[R"U5UR5S9S9$)NrrArRrY) rBrCr"r*rrrTENSORrr rr^)r'r?r,s r4_create_write_item_for_tensorresmjj!s6;;=112G C)  ! !#&wkkmL'::6B  r6bytescD[[U5[RS9$)N)r&r%)rrrBYTE_IO)r'rfs r4_create_write_item_for_bytesioris C  " " r6c [[RU[R"U45U[R"U45[R"U45S9$N)r% dest_index dest_offsets storage_indexstorage_offsetslengths)rrrhrBrCrl dest_offsetrnstorage_offsetlengths r4_create_read_item_for_byteiorusK   ! !ZZ/# N#45 F9%  r6c [[RU[R"U5U[R"U5[R"U5S9$rk)rrrdrBrCrlrmrnrorps r4_create_read_item_for_tensorrxsD   ZZ -# ?3 7#  r6 checkpoint_md local_chunksc /n[U5HupE[UR5Hupg[XW5(dM/n/n /n [XuS9H;un n n nUR U 5 U R U 5 U R U5 M= UR [ [ XRU5U [ XRU5UU S95 M M U$)a' Create a list of ``ReadItem`` based on the checkpoint and local chunks. This applies the resharding algorithm and computes the reads needed to satisfy ``local_chunks`` with a checkpoint described by ``checkpoint_md``. Args: fqn (str) : The state_dict FQN to pass to ``ReadItem``. checkpoint_md (TensorStorageMetadata): metadata for a given tensor from a checkpoint. local_chunks (List[ChunkStorageMetadata]): Local chunks that needs to be loaded. Returns: A list of ``ReadItem`` that will satisfy all input chunks. ) saved_shard current_shardrw) enumeratechunksrrr:rxrr,)r'ryrz read_itemsidxshard storage_idx storage_mdrormrp_dimoffset_for_saved_tensoroffset_for_current_tensorrts r4rrs*J - '01E1E'F #K5eHH OLG =& ') &&'>?##$=>v&    ,,S--E!-"/5G5G"U$3#  '(G.: r6 state_dictc^^/nUR5Humm[T[5(aUR[ TT55 M8[T[ 5(a6UR UU4SjTR5R55 M[T[R5(aUR[TT55 MUR[TT55 M [U5$)Nc3>># UHn[TTU5v M g7f)N)rb).0rEr'objs r4 5_create_default_metadata_only_plan..,s$ >H-S#x@@ >s)r# isinstancer r:r`rextendrTshards_metadatarBTensorrerir)rrequestsr'rs @@r4"_create_default_metadata_only_planr&sH$$&S c7 # # OO;CE F ] + + OO #  > > U\\ * * OO9#sC D OO:3D E' H r6objectcR[US5(aURX5$[U[5(a5UR 5Vs/sHn[ XUR 5PM sn$[U[R5(a [X5/$[X5/$s snf)N__create_write_items__) hasattrrrr local_shardsrbrTrBrreri)r'rrs r4_create_write_itemsr7sv/00,,S99 FM * * ,,. . )enn E.   FELL ) )-c:;;.s;<< s B$c[URURUR5up[R "U5[R "U5p![ UUS9$rG)r rZr[r\rBrCr )r?r-r,s r4_create_chunk_from_dtensorrFsS: f((&*;*;NEZZ& 7(;7  r6cn[US5(aUR5nU$[U[5(a5UR 5Vs/sHn[ UR 5PM nnU$[U[R5(a[U5/nU$[S[U535es snf)N__create_chunk_list__zMUnsupported Type, expecting one of [Tensor, DTensor, ShardedTensor] ,but got ) rrrrrrJrTrBrrD ValueErrorr%)r?rzrs r4_create_chunk_listrQsv.//335   FM * *:@:M:M:O :O U^^ ,:O    FELL ) )1&9:    V ~ '   s B2mdrc[U[5(d[U5n[ XU5$[ [U5S[U5SSS9/$![a$n[SUS3S[ U53-5UeSnAff=f)Nz Invalid checkpoint metadata for z, z(expected BytesStorageMetadata but found rrq)rr rrr%rrur)r'rrrzexs r4_create_read_itemsrds b. / / -c2L0FF )(-+C0      23%r:.dtensor_funcs$/ U\\&) )//FFHMMK 0=LLNF %//0@P (( !-- ++jjl||~ GNLr6c[USS5nU[R"S5:Xa[S[ U5S35eU$)NrrzFound unsupported type z for meta device loading.)rrBr RuntimeErrorr%)rrs r4sharded_tensor_func-_init_state_dict..sharded_tensor_funcsF$/ U\\&) ))$u+6OP Lr6c2[USS5nU[R"S5:Xao[RR 5R n[[R[U5R55n[R"XS9nU$U$)Nrrr) rrBrrrrr%rrrr)rrrr?s r4 tensor_func%_init_state_dict..tensor_funcsy$/ U\\&) )//FFHMMK 0=LLNF%%e;FMLr6N)r rrBr_iterate_state_dict)rrrrs r4_init_state_dictr{s@ G*3 5<<  r6 iter_objectrrrc p[U[5(aU"U5$[U[5(aU"U5$[U[R5(aU"U5$[U[ [ [[[R45(dUcU$[U[5(a)UR5HupE[XQX#5X'M U$[U[[45(a>UVs/sHn[XaX#5PM nn[U[5(a [U5nU$gs snf)a Iterate through the state dict, applying the given functions to each tensor type and update the state dict in place. Args: iter_object (Any): the target state_dict. sharded_tensor_func (Callable): the function to apply to ShardedTensor dtensor_func (Callable): the function to apply to DTensor tensor_func (Callable): the function to apply to Tensor # TODO: let state_dict_util._iterate_state_dict() to support in place option so we don't need to have two versions of _iterate_state_dict. N)rr rrBrintfloatstrrfioBytesIOdictr#rlisttuple)rrrrkeyrvrets r4rrs(+w''K(( K / /";// K . .;'';eS% DEE   K & &%++-JC2%8 K . K$ / /!   1D R    k5 ) )*C  0 s8D3)?rtypingrrrrBtorch.distributed distributedr torch._utilsr!torch.distributed._shard.metadatar'torch.distributed._shard.sharded_tensorrtorch.distributed.tensorr torch.distributed.tensor._utilsr rTr r rrrrrplannerrrrrrr reshardingrrrrr__annotations__boolr5r>rrDrJrWr`rbrerirurxrrrrrrrrrr6r4rs] && +;A,Q 99c9>h>H>>Bx.h (^2ell7K }1E!-:(g)( +7D s ELL Y C  4 4(4+,4 (^ 4n?x" =S =# =$y/ =w3Gu||5I0J& C ]  h .4c3h4C4n,,,", ,r6