o Zh1@sddlZddlZddlmZmZddlmZGdddeZGdddeZ Gd d d e Z Gd d d e Z Gd ddeZ GdddeZ dS)N)DatasetIterableDataset) ModelOutputc@s$eZdZddZddZddZdS)PipelineDatasetcC||_||_||_dSNdatasetprocessparams)selfr r r rN/var/www/auris/lib/python3.10/site-packages/transformers/pipelines/pt_utils.py__init__  zPipelineDataset.__init__cC t|jSrlenr r rrr__len__ zPipelineDataset.__len__cCs"|j|}|j|fi|j}|Srr )r iitem processedrrr __getitem__s zPipelineDataset.__getitem__N)__name__ __module__ __qualname__rrrrrrrrs rc@s6eZdZd ddZddZddZdd Zd d ZdS) PipelineIteratorNcCs4||_||_||_|dkrd}||_d|_d|_dS)a Roughly equivalent to ``` for item in loader: yield infer(item, **params) ``` Arguments: loader (`torch.utils.data.DataLoader` or `Iterable`): The iterator that will be used to apply `infer` on. infer (any function): The function to apply of each element of `loader`. params (`dict`): The parameters passed to `infer` along with every item loader_batch_size (`int`, *optional*): If specified, the items of `loader` are supposed to come as batch, and are loader_batched here making it roughly behave as ``` for items in loader: for i in loader_batch_size: item = items[i] yield infer(item, **params) ```N)loaderinferr loader_batch_size_loader_batch_index_loader_batch_datar r!r"r r#rrrrs zPipelineIterator.__init__cCrr)rr!rrrrr?rzPipelineIterator.__len__cCt|j|_|Sriterr!iteratorrrrr__iter__B zPipelineIterator.__iter__cstjtjrjjd}ni}jD]\}}t|trQ|}t|dtjr;t fdd|D||<nt|dt j rPt fdd|D||<q|dvrt|t rt|dtjrpt fdd|D||<nt|dt j rt fdd|D||<q|durd||<qt|jtjr|jd||<qt|jt j rt |jd||<q|j||<qj |}jd 7_|S) ze Return item located at `loader_batch_index` within the current `loader_batch_data`. rc3 |] }|jdVqdSrNr$ unsqueeze.0elrrr Uz5PipelineIterator.loader_batch_item..c3"|] }t|jdVqdSr.np expand_dimsr$r1rrrr4W >Z hidden_statesZ attentionsZpast_key_valuesc3r-r.r/r1rrrr4\r5c3r6r.r7r1rrrr4^r:Nr ) isinstancer%torchTensorr$r0itemsrZto_tupletupler8Zndarrayr9 __class__)r resultZloader_batchedkelementrrrloader_batch_itemFs6   z"PipelineIterator.loader_batch_itemcCs|jdur|j|jkr|St|j}|j|fi|j}|jdurtt|tj r,|}nt|t r6|d}n t | d}||}t|t rLt |}n|jd}d|kr\|jkrann||_t|t rj|dn||_d|_|S|S)Nr)r$r#rDnextr*r"r r;r<r=r?listkeysrshaper%)r rr first_tensorkeyobserved_batch_sizerrr__next__us(        zPipelineIterator.__next__r)rrrrrr+rDrLrrrrrs  ' /rcs.eZdZdfdd ZddZddZZS) PipelineChunkIteratorNcst|||dS)aI Roughly equivalent to ``` for iterator in loader: for item in iterator: yield infer(item, **params) ``` Arguments: loader (`torch.utils.data.DataLoader` or `Iterable`): The iterator that will be used to apply `infer` on. infer (any function): The function to apply of each element of `loader`. params (`dict`): The parameters passed to `infer` along with every item N)superrr&r@rrrszPipelineChunkIterator.__init__cCst|j|_d|_|Sr)r)r!r* subiteratorrrrrr+s zPipelineChunkIterator.__iter__cCst|jdur |jt|jfi|j|_zt|j}W|Sty9|jt|jfi|j|_t|j}Y|Swr)rPr"rEr*r StopIteration)r rrrrrLs    zPipelineChunkIterator.__next__r)rrrrr+rL __classcell__rrrOrrMsrMc@s eZdZdZddZddZdS)PipelinePackIteratora Roughly equivalent to ``` packed = [] for item in loader: packed.append(item) if item["is_last"]: yield packed packed = [] ``` but it also handles cases where `item` are batched (meaning it's a dict of Tensor with first dimension > 1. In that case it does ``` packed = [] for batch in loader: # item is batched for item in batch: packed.append(item) if item["is_last"]: yield packed packed = [] ``` Arguments: loader (`torch.utils.data.DataLoader` or `Iterable`): The iterator that will be used to apply `infer` on. infer (any function): The function to apply of each element of `loader`. params (`dict`): The parameters passed to `infer` along with every item loader_batch_size (`int`, *optional*): If specified, the items of `loader` are supposed to come as batch, and are loader_batched here making it roughly behave as ``` for items in loader: for i in loader_batch_size: item = items[i] yield infer(item, **params) ```cCr'rr(rrrrr+r,zPipelinePackIterator.__iter__cCsTd}g}|jdur-|j|jkr-|j|jkr-|}|d}|||r'|S|j|jks|s|jt|jfi|j}|jdurt |t j rJ|}n t | d}||}t |t r`t|}n|jd}d|krp|jkrunn||_||_d|_|j|jkr|}|d}|||r|S|j|jksn |}|d}|||r/|S)NFis_lastr)r$r#rDpopappendr"rEr*r r;r<r=rFrGrrHr%)r rTZ accumulatorrrrIrJrKrrrrLsJ                zPipelinePackIterator.__next__N)rrr__doc__r+rLrrrrrSs- rSc@s.eZdZdedefddZddZddZd S) KeyDatasetr rJcCs||_||_dSrr rJ)r r rJrrrr,s zKeyDataset.__init__cCrrrrrrrr0rzKeyDataset.__len__cCs|j||jSrrYr rrrrr3szKeyDataset.__getitem__NrrrrstrrrrrrrrrX+s rXc@s2eZdZdededefddZddZdd Zd S) KeyPairDatasetr key1key2cCrrr r^r_)r r r^r_rrrr8rzKeyPairDataset.__init__cCrrrrrrrr=rzKeyPairDataset.__len__cCs"|j||j|j||jdS)N)textZ text_pairr`rZrrrr@s"zKeyPairDataset.__getitem__Nr[rrrrr]7s r])numpyr8r<Ztorch.utils.datarrZ utils.genericrrrrMrSrXr]rrrrs -d