a d$ @sddlZddlZddlZddlZddlmZmZmZddlm Z ddl Z ddl m Z ddl mZmZmZmZmZddlmZddlmZdd lmZzdd l mZejZWneye jZYn0e je je jd kred gd Z eZ!dddddddddd Z"e#e#dddZ$ddZ%ddZ&ee#dd d!Z'e j()rPd"nd#d$dfe#ee#e j*fe+e#d%d&d'Z,d,ee#ee#fe-e+ee j.e j/fd)d*d+Z0dS)-N)AnyUnionList) packaging)Image)ComposeResize CenterCropToTensor Normalize)tqdm) build_model)SimpleTokenizer)InterpolationModez1.7.1z.PyTorch version 1.7.1 or higher is recommended)available_modelsloadtokenizezwhttps://openaipublic.azureedge.net/clip/models/afeb0e10f9e5a86da6080e35cf09123aca3b358a0c3e3b6c78a7b63bc04b6762/RN50.ptzxhttps://openaipublic.azureedge.net/clip/models/8fa8567bab74a42d41c5915025a8e4538c3bdbe8804a470a72f30b0d94fab599/RN101.ptzyhttps://openaipublic.azureedge.net/clip/models/7e526bd135e493cef0776de27d5f42653e6b4c8bf9e0f653bb11773263205fdd/RN50x4.ptzzhttps://openaipublic.azureedge.net/clip/models/52378b407f34354e150460fe41077663dd5b39c54cd0bfd2b27167a4a06ec9aa/RN50x16.ptzzhttps://openaipublic.azureedge.net/clip/models/be1cfb55d75a9666199fb2206c106743da0f6468c9d327f3e0d0a543a9919d9c/RN50x64.ptz{https://openaipublic.azureedge.net/clip/models/40d365715913c9da98579312b702a82c18be219cc2a73407c4526f58eba950af/ViT-B-32.ptz{https://openaipublic.azureedge.net/clip/models/5806e77cd80f8b59890b7e101eabd078d9fb84e6937f9e85e4ecb61988df416f/ViT-B-16.ptz{https://openaipublic.azureedge.net/clip/models/b8cca3fd41ae0c99ba7e8951adf17d267cdb84cd88be6f7c2e0eca1737a03836/ViT-L-14.ptzhttps://openaipublic.azureedge.net/clip/models/3035c92b350959924f9f00213499208652fc7ea050643e8b385c2dac08641f02/ViT-L-14-336px.pt) ZRN50ZRN101ZRN50x4ZRN50x16ZRN50x64zViT-B/32zViT-B/16zViT-L/14zViT-L/14@336px)urlrootc Cstj|ddtj|}|dd}tj||}tj|r\tj|s\t|dtj|rt t |d  |kr|St|dtj|}t |d~}tt|d d d dd d <}| d}|sq|||t|qWdn1s0YWdn1s60YWdn1sV0Yt t |d  |krtd|S)NT)exist_ok/z! exists and is not a regular filerbzH exists, but the SHA256 checksum does not match; re-downloading the filewbzContent-LengthPZiBi)totalZncolsZunitZ unit_scaleZ unit_divisori zDModel has been downloaded but the SHA256 checksum does not not match)osmakedirspathbasenamesplitjoinexistsisfile RuntimeErrorhashlibZsha256openread hexdigestwarningswarnurllibrequesturlopenr intinfogetwriteupdatelen) rrfilenameZexpected_sha256Zdownload_targetsourceoutputZloopbufferr9R/var/www/html/stable-diffusion-webui/venv/lib/python3.9/site-packages/clip/clip.py _download+s(  "  pr;cCs |dS)NZRGB)convert)imager9r9r:_convert_image_to_rgbKsr>cCs&tt|tdt|tttddgS)N) interpolation)g3<4'?gwg M?gy{ ?)gB91?gwt.?g U?)rrBICUBICr r>r r )Zn_pxr9r9r: _transformOs rA)returncCs ttS)z*Returns the names of available CLIP models)list_MODELSkeysr9r9r9r:rYsrcudacpuF)namedevicejit download_rootc s|tvr$tt||ptjd}n(tj|r6|}ntd|dtt|dp}z$t j j ||rjndd }d}Wn:ty|rt d|d d }t j |dd}Yn0Wdn1s0Y|st|p|}tdkr||t|jjfSt j jfd d gd }dd|jdDdfdd} || | |j| |jtdkrt j jdd gd } t| jdd} | fdd} || | |j| |j||t|j!fS)aLoad a CLIP model Parameters ---------- name : str A model name listed by `clip.available_models()`, or the path to a model checkpoint containing the state_dict device : Union[str, torch.device] The device to put the loaded model jit : bool Whether to load the optimized JIT model or more hackable non-JIT model (default). download_root: str path to download the model files; by default, it uses "~/.cache/clip" Returns ------- model : torch.nn.Module The CLIP model preprocess : Callable[[PIL.Image], torch.Tensor] A torchvision transform that converts a PIL image into a tensor that the returned model can take as its input z ~/.cache/clipzModel z not found; available models = rrG)Z map_locationNzFile z6 is not a JIT archive. Loading as a state dict insteadFcstgtSN)torchonestorIr9)rIr9r:zload..)Zexample_inputscSsg|]}dt|vr|qS)ZDevice)repr).0nr9r9r: rQzload..prim::Constantcszt|dr|jgng}Wnty0g}Yn0t|drJ||jj|D]<}|dD],}d|vr\t|ddr\| q\qNdS)Ngraphforward1rVvaluerF) hasattrrXr%appendrY findAllNodesZattributeNamesstr startswithcopyAttributes)modulegraphsrXnode) device_noder9r: patch_devices   zload..patch_devicecSstgSrL)rMrNfloatr9r9r9r:rPrQaten::tor cszt|dr|jgng}Wnty0g}Yn0t|drJ||jj|D]P}|dD]@}t|}dD]*}||ddkrp|| qpq\qNdS)NrXrYrg)r rZ) r[rXr%r\rYr]rCinputsrcr`)rarbrXrcrji) float_noder9r: patch_floats    zload..patch_float)"rDr;rr expanduserr$r%rr'rMrJrevalr*r+r state_dictrOr^rfrAZvisualZinput_resolutiontracerXr]ZapplyZ encode_imageZ encode_textrCZfindNoderjrcitem) rHrIrJrKZ model_pathZ opened_filemodelrpZ device_holderreZ float_holderZ float_inputrmr9)rIrdrlr:r^sF   2        rM)textscontext_lengthtruncaterBcst|tr|g}tjdtjdfdd|D}tjtjtjdkrjtj t ||tj d}ntj t ||tj d}t |D]`\}}t ||kr|r|d|}|d<ntd ||d |t|||dt |f<q|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|]}gt|gqSr9) _tokenizerencode)rStextZ eot_tokenZ sot_tokenr9r:rUrQztokenize..z1.8.0)ZdtypeNrWzInput z is too long for context length ) isinstancer^rxencoderrversionparserM __version__zerosr4longr/ enumerater%Ztensor)rurvrwZ all_tokensresultrktokensr9r{r:rs       r)rtF)1r&rr,r*typingrrr pkg_resourcesrrMZPILrZtorchvision.transformsrrr r r r rsrZsimple_tokenizerrZ _Tokenizerrr@ ImportErrorr~rrr+__all__rxrDr^r;r>rArrFZ is_availablerIboolrr/Z IntTensorZ LongTensorrr9r9r9r:sH             6g