g^(h>> # xdoctest: +SKIP >>> sum_reducer = _CustomReducer( >>> torch.tensor(0.0), >>> lambda a, b: a + b >>> ) c ||_||_yN) init_value reduce_fn)selfrrs e/var/www/skyplay_api_hub/venv/lib/python3.12/site-packages/torch/distributed/pipelining/microbatch.py__init__z_CustomReducer.__init__(s$"N)__name__ __module__ __qualname____doc__rrrr r s  #rr c eZdZy) _LossReducerNrrrrrrrr-rrgc ||zSrr)abs rr"1s 1q5rcneZdZUdZdZeed<dZdZe de edffdZ e de e effd Zy ) r z2 Class used to specify chunking of inputs c||_yr split_dim)rr&s rrzTensorChunkSpec.__init__=s "rr&c||jjd|jjd|jdS)N.()) __class__rrr&rs r__repr__zTensorChunkSpec.__repr__Bs9~~(()4>>+B+B*C1T^^DTTU V rc"d|jdS)NzTensorChunkSpec(r*r%r,s r__str__zTensorChunkSpec.__str__Gs!$..!133r chunk_dims.c t|d}|S)a A helper for creating a tuple of `TensorChunkSpec` from a tuple of chunk dimensions (int's). Example: >>> # xdoctest: +SKIP >>> # There are three positional arguments to the model, and >>> # we are chunking them along dimension 0, 0 and 1, respectively >>> args_chunk_spec = TensorChunkSpec.from_tuple((0, 0, 1)) ct|Srr dims rr"z,TensorChunkSpec.from_tuple..Y ,rr)r0args_chunk_specs r from_tuplezTensorChunkSpec.from_tupleJs(  , rc t|d}|S)a\ A helper for creating a dictionary of `TensorChunkSpec` from a dictionary of chunk dimensions (int's). Example: >>> # xdoctest: +SKIP >>> # Chunk dimension 0 for the "id" argument, 1 for the "mask" argument >>> kwargs_chunk_spec = TensorChunkSpec.from_dict({"id": 0, "mask": 1}) ct|Srr3r4s rr"z+TensorChunkSpec.from_dict..kr6rr)r0kwargs_chunk_specs r from_dictzTensorChunkSpec.from_dict]s*  , ! rN)rrrrrint__annotations__r-r/ staticmethodtupler8dictstrr<rrrr r 8se#N 4#s(O$!cN!!rr c eZdZy) _ReplicateNrrrrrDrDqrrrDc i}g}|}d}t|t|k(s;Jdt|jdt|j|jD]G\}}t |\} } |j | ||} | Jt | \} } t| t| k7rt d|d| g}t| | D]\}}|tust|tjs|j |g|z?t|trqt|tjs J|d|j|j}||kr9|r"tj!d|d |d |d |}nt#d |d |d|dtj$|||j}t&rg}d}|D]}tj(|}||j|jz}t+dddg|j,z}t+||||j<|||<|j |||j|jz }|j |n|j |d}t/d||||<Jg}t1|D]D}i}|jD]\}}|Dcgc]}|| }}|||<|j |Fg}|D]b} i}!t|t| k(sJt| j|D]\\}}}"t3||"|!|<|j |!d|Scc}w)aW Given a dictionary of args, and a dictionary of chunking specs, shard the args according to the chunking specs. Args: args_dict: Dictionary of args args_chunk_spec: Dictionary of chunking specs num_chunks: Number of chunks to shard the args into Returns: args_split: List of sharded args Tzargs_dict.keys() = z args_chunk_spec.keys() = NzArgument value z9 did not have the same number of values as as chunk spec z is not a tensorz%Tensor size on chunking dimension is z', downsizing the number of chunks from z to r(zArg z% on chunking dimension has a size of z$, smaller than the number of chunks z. PiPPy cannot reduce the number of chunks because other arguments have bigger chunk-dimension sizes. Please adjust your num_chunks setting.rFzUnrecognized chunk spec: )lenlistkeysitemsrappend ValueErrorziprD isinstancetorchTensorr sizer&loggerwarning RuntimeError tensor_split_debug_mask_minibatches zeros_likeslicendim TypeErrorranger)# args_dictr7 num_chunksargs_sharded_replicated arg_specsreal_num_chunks first_tensorarg_keyargflatspec chunk_specchunk_spec_flat_sharded_arg_flatvchunk_vv_split_dim_size chunk_tensorsexpanded_chunks split_dim_idx chunk_tensornew_val upper_idx slice_indices chunks_flat chunk_idx chunk_argskeyv_flatarg_single_chunk args_splitchunkper_chunk_argsarg_specs# r_shard_dict_of_argsr}us(!I OL y>S1 1 d9>>#3455OPTUdUiUiUkPlOmn 1")I< !#& d$W- %%%)*5 t9O, ,!#'++5,8  dO48 GJAw*$Jq%,,,G ''o(=>G_5"!U\\2Jqc9I4JJ2#$66'*;*;#< #o5#CDTCUVDDN .'..w7%):):7;L;L)MM  N%++O<$++M:$ ";G9 EFFq8 Gt,<(SIz0split_args_kwargs_into_chunks..Vs <jm v> 0  !# "#>*c+.>>>>14^[1Q4-M:$3joocmm6T$TM%*4t%<$= @R@R$RM38-3XM#--0*=9F!((0&3O4!/  ! !%))Ms}}"M N ^ ,..K"3'7#89  !mm!1)!rs8 '<    8 $  ##$ > <5<<,.@A 5!5!r  |F>B>B f$ S/f$ T#s(^ $f$ f$eOS$89: f$  S/%9 :; f$  4;T "# f$Rw8 Iw8r