ϑi:<SSKrSSKJrJr SSKrSSKrSrSrSr g)N)MappingSequence_paddle_field_c^U4Sjm[U[5(d/nT"U/U/S5up#XS4$/nT"X/S5up#X4$)a  For lod_blocking_queue only receive tensor array, flatten batch data, extract numpy.array data out as a list of numpy.array to send to lod_blocking_queue, and save the batch data structure such as fields in other types (str, int, etc) or key-value map of dictionaries cr>[U[5(GaSUGHIn[U[R[R [R RRR 45(a1UR[U35 URU5 US- nM[U[[[R45(aURU5 M[U[5(a T"XA/U5upSURU5 GM[U[5(a T"XA0U5upSURU5 GM8URU5 GML X#4$[U[5(Ga!UR!5GH upd[U[R[R [R RRR 45(a$[U3X&'URU5 US- nM[U[[[R45(aXBU'M[U[5(aT"XA/U5upSXRU'M[U[5(aT"XA0U5upSXRU'GMXBU'GM X#4$[#S[%U535e)Nwrong flat data type: ) isinstancernpndarraypaddleTensorbasecoreeagerappend FIELD_PREFIXstrbytesnumbersNumberritems TypeErrortype)batch flat_batch structure field_idxfield field_structk_flattens Y/var/www/html/banglarbhumi/venv/lib/python3.13/site-packages/paddle/io/dataloader/flat.pyr" _flatten_batch.._flatten"sD eX & &ZZ 0@0@0F0F0M0MN$$ ~i[%AB%%e,NIUGNN'CDD$$U+x00.62y/+L$$\2w//.62y/+L$$\2$$U++`##5w ' '!KKMZZ 0@0@0F0F0M0MN'3^I;#?IL%%e,NIUGNN'CDD#(aLx00.62y/+L$0aLw//.62y/+L$0aL#(aL+*2##4T%[MBC Cr)r r)rrr_r"s @r#_flatten_batchr's\2$j eX & & R; Q<''JEr15LI   r%c&^^UU4Sjm[T[5(dS5e[T5S:XaU$[U[[45(aU[ S3:Xd SU35eTS$T"US5nUS-[T5:XdS5eU$)z After reading list of Tensor data from lod_blocking_queue outputs, use this function to restore the batch data structure, replace :attr:`_paddle_field_x` with data from flat_batch c>[U[5(a[U5Hup#[U[5(a_UR [ 5(aE[ UR[ S55n[X5nTUcS5eTUX'STU'My[U[[[R45(aM[U[[45(dMT"XU5nM U$[U[5(aUR5HupS[U[5(a_UR [ 5(aE[ UR[ S55n[X5nTUcS5eTUX'STU'My[U[[[R45(aM[U[[45(dMT"XU5nM U$[S[U535e)Nz flat_batch[{}] parsed repeatedlyr )r r enumerater startswithrintreplacemaxrrrrrrr)rrir cur_field_idxr!_restorers r#r2 _restore_batch.._restoregs i * *%i0eS))e.>.>|.L.L$' lB(G$HM #I =I%m4@:@$.m#.>|.L.L$' lB(G$HM #I =I%m4@:@$.m#r9s$-  D!N8r%