o ߥiv @szddlZddlZddlZddlZddlmZmZmZddlZddl Zddl Z ddl m Z ddl mZmZmZmZmZddlmZddlmZddlmZz dd l mZejZWn eyce jZYnweje jejd krved d d gZ ddZ!ddZ"e j#$rdndddfde%dee%e j&fde'de%fdd Z(  d dee%ee%fde)de'dee j*e j+ffdd Z,dS)!N)AnyListUnion)Image) CenterCropCompose NormalizeResizeToTensor)tqdm) build_model)SimpleTokenizer)InterpolationModez1.7.1z.PyTorch version 1.7.1 or higher is recommendedloadtokenizecCs |dS)NRGB)convert)imagerf/home/ubuntu/.local/lib/python3.10/site-packages/modelscope/models/cv/text_driven_segmentation/clip.py_convert_image_to_rgb"s rcCs&tt|tdt|tttddgS)N) interpolation)g3<4'?gwg M?gy{ ?)gB91?gwt.?g U?)rr BICUBICrrr r)n_pxrrr _transform&s rcudacpuFnamedevicejitrootc s|st}tdkr||t|jjfStjj fddgd}dd|j dDdfd d }| |||j ||jtdkrtjj d dgd}t|j d d }|fdd} | | | |j | |j||t|jfS)NrcstgtSN)torchonestorr)rrr?szload..)example_inputscSsg|] }dt|vr|qS)Device)repr).0nrrr @s  zload..prim::Constantcsz t|dr |jgng}Wn tyg}Ynwt|dr%||jj|D]}|dD]}d|vrDt|ddrD| q.q'dS)Ngraphforward1r-valuer) hasattrr/ RuntimeErrorappendr0 findAllNodesattributeNamesstr startswithcopyAttributes)modulegraphsr/node) device_noderr patch_deviceEs$   zload..patch_devicecSstgSr")r#r$floatrrrrr&[saten::tor csz t|dr |jgng}Wn tyg}Ynwt|dr%||jj|D](}|dD] }t|}dD]}||ddkrM|| q8q.q'dS)Nr/r0r@)r r1) r2r/r3r4r0r5listinputsr<r9)r:r;r/r<rDi) float_noderr patch_float_s"   zload..patch_float)r r%r7r?rvisualinput_resolutionr#r tracer/r5apply encode_image encode_textrCfindNoderDr<item) rrr r!model device_holderr> float_holder float_inputrGr)rr=rFrr1s:             Mtextscontext_lengthtruncatereturncst|tr|g}jdjdfdd|D}tjtjtjdkr6tjt ||tj d}n tjt ||tj d}t |D]0\}}t ||krh|r\|d|}|d<n t d ||d |t|||dt |f<qE|S) a Returns the tokenized representation of given input string(s) Parameters ---------- texts : Union[str, List[str]] An input string or a list of input strings to tokenize context_length : int The context length to use; all CLIP models use 77 as the context length truncate: bool Whether to truncate the text in case its encoding is longer than the context length Returns ------- A two-dimensional tensor containing the resulting tokens, shape = [number of input strings, context_length]. We return LongTensor when torch version is <1.8.0, since older index_select requires indices to be long. z<|startoftext|>z <|endoftext|>cs"g|] }g|gqSr)encode)r*text _tokenizer eot_token sot_tokenrrr,sztokenize..z1.8.0)dtypeNr.zInput z is too long for context length ) isinstancer7encoder packagingversionparser# __version__zeroslenlongint enumerater3tensor)r\rUrVrW all_tokensresultrEtokensrr[rrzs0       )rTF)-hashlibosurllibwarningstypingrrrrbpackaging.versionr#PILrtorchvision.transformsrrrr r r rPr simple_tokenizerr _Tokenizerrr ImportErrorrcrdrewarn__all__rrr is_availabler7rboolrri IntTensor LongTensorrrrrrsf             L