VhddlZddlmZmZmZddlZddlmZmZe edfZ eee e fge edffZ gdZdedefdZd ej"j$deeeffd Zd eedeeeffd Zd edefdZy)N)AnyCallableUnion)tree_flatten_with_pathtree_map.)normalize_source_namemodule_to_nested_dicttrack_dynamism_across_examplesclone_and_convert_to_metanamereturnc0tjdd|S)Nz\.([a-zA-Z_][a-zA-Z0-9_]*)z['\1'])resub)r s O/home/dcms/DCMS/lib/python3.12/site-packages/torch/fx/experimental/_dynamism.pyrrs 66/D AAmodulecDi}i|d<i|d<t|D]}|jdrtt||r+t||}t |t j jr\t |ttt jfst|tus|||<|jdD] \}}||d|<|jdD] \}}||d|<|jD]\}}t!||d|<|S)ziRecursively converts an nn.Module into a nested dictionary with explicit 'parameters' and 'modules' keys. _parameters_modules_F)recurse)dir startswithcallablegetattr isinstancetorchnnModuleintfloatTensortypeboolnamed_parameters named_buffersnamed_childrenr )r self_dict attr_name attr_valuer parambuffer submodules rr r s2 "I!ImIj[2 ##C('&):T1U 3Jz588??;zC +EF$D0'1 )$2..u.=/ e). - &/,,U,;0 f)/ - &0"002Gi&;I&F *d#G rexample_inputsci}|D]m}d|vr8t|dtjjrt |d|d<t |\}}|D]\}}t|t ttjfs-t|tjrt|j}d}n|f}d}||vr0tt|Dcgc] }tc}|f||<nV||\} } | |k7r t| t|kr1| jtt| t|kr1t|D]\} } ||d| j!| !pi} |j#D]U\}\} }td| D}ddj%d|Dz}|dj&}|| vri| |<|| ||<W| Scc}w) a This function analyzes a list of example inputs to determine the dynamism of their shapes. It tracks whether the dimensions of tensors or non-tensor values change across different examples. The function returns a dictionary where each key represents a path to a value in the input examples, and the corresponding value is a tuple indicating which dimensions are dynamic (i.e., change across examples). This helps in understanding how the structure of data varies across different instances. selfTFrc38K|]}t|dkDyw)N)len).0ss r z1track_dynamism_across_examples..[s7#a&1*7sLc34K|]}t|yw)N)str)r5ks rr7z1track_dynamism_across_examples..\s>3q6( >s)rrrr r rr!r"r#tupleshaperanger4setappend enumerateadditemsjoinkey)r/trackingexleaves_with_pathsrkey_pathvaluer> is_tensordim_setsflagidimoutput _is_tensor final_dynkey_strrFs rr r 4s<>H2 RX>>>qkoo f F3K(s G ) M%'IsG" example_inputc:dtdtfd}t||S)z This function takes a list of example inputs and for each tensor, clones it and converts it to device=meta. For non-tensor values, it keeps the reference. It uses pytree to handle nested structures recursively. rKr czt|tjr |jj dS|S)Nmeta)device)rrr#cloneto)rKs r transform_fnz/clone_and_convert_to_meta..transform_fnjs/ eU\\ *;;=##6#2 2 r)rr)rUr\s rr r ds$ CC L- 00r)rtypingrrrrtorch.utils._pytreerrr=KeyPathr!r"NonTensorShapeFn__all__r;rrr dictr listr r rrres '' @ S/U3:./sCx@A BBB %((//d38n8-I- #s(^-` 1S 1S 1r