o i @sddlmZmZmZmZmZddlZddlZddl m Z m Z ddl m Z GdddZ   dd eeeeeejffd eeefd ed eedeeeeeejfff ddZdS)) CollectionDictListTupleUnionN)check_argument_typescheck_return_type)pad_listc@seZdZdZ   ddeeefdedeefdd Z d d Z d ee ee ee jffd e eee eejfffddZdS)CommonCollateFnz$Functor class of common_collate_fn()float_pad_value int_pad_value not_sequencecCs$tsJ||_||_t||_dS)N)rrrsetr)selfrrrr r L/home/ubuntu/.local/lib/python3.10/site-packages/espnet2/train/collate_fn.py__init__ s zCommonCollateFn.__init__cCs|jd|jd|jdS)Nz(float_pad_value=z, int_pad_value=)) __class__r)rr r r__repr__szCommonCollateFn.__repr__datareturncCst||j|j|jdS)N)rrr)common_collate_fnrrr)rrr r r__call__s zCommonCollateFn.__call__Nr r r )__name__ __module__ __qualname____doc__rfloatintrstrrrrrnpndarrayrtorchTensorrr r r rr s$   r r r r rrrrrc stsJddD}ddDtfddDs"JdtdddDs8Jd tdi}dD]Cdjjd krM|}n|}fd dD}d d|D}t||} | |<|vrtjfd dDtjd} | |d<q>||f}t |sJ|S)asConcatenate ndarray-list to an array and convert to torch.Tensor. Examples: >>> from espnet2.samplers.constant_batch_sampler import ConstantBatchSampler, >>> import espnet2.tasks.abs_task >>> from espnet2.train.dataset import ESPnetDataset >>> sampler = ConstantBatchSampler(...) >>> dataset = ESPnetDataset(...) >>> keys = next(iter(sampler) >>> batch = [dataset[key] for key in keys] >>> batch = common_collate_fn(batch) >>> model(**batch) Note that the dict-keys of batch are propagated from that of the dataset as they are. cSsg|]\}}|qSr r ).0u_r r r Az%common_collate_fn..cSsg|]\}}|qSr r )r(r*dr r rr+Br,c3s$|] }tdt|kVqdS)rN)rr(r-)rr r Ds"z$common_collate_fn..zdict-keys mismatchingcss|] }|d VqdS)_lengthsN)endswith)r(kr r rr/Es  rz*_lengths is reserved: icsg|]}|qSr r r.keyr rr+Sr,cSsg|]}t|qSr )r& from_numpy)r(ar r rr+Wscsg|] }|jdqS)r)shaper.r4r rr+^s)dtyper0) ralllistr9kindr r&tensorlongr) rrrruttidsoutput pad_value array_list tensor_listr=lensr )rr5rr)s2      rr)typingrrrrrnumpyr$r& typeguardrr&espnet.nets.pytorch_backend.nets_utilsr r r#r%r!r"r'rr r r rs( !