o S"g @sJddlmZddlZddlZddlmZddlmZGdddejZdS)) annotationsN)Image)nncseZdZUdZded<d$d%fd d Zd&d d Zed'ddZej d(ddZd)ddZ d*d+ddZ ed,ddZ d-dd Z ed.d"d#ZZS)/ CLIPModelTbool save_in_rootopenai/clip-vit-base-patch32N model_namestrreturnNonecs6t|dur |}tj||_tj||_dSN)super__init__ transformersrfrom_pretrainedmodel CLIPProcessor processor)selfr processor_name __class__h/mnt/skqttb/ctump_chatbot/chatbot/lib/python3.10/site-packages/sentence_transformers/models/CLIPModel.pyr s zCLIPModel.__init__cCsdS)Nz CLIPModel()rrrrr__repr__szCLIPModel.__repr__intcCs |jjjSr r tokenizermodel_max_lengthrrrrmax_seq_length zCLIPModel.max_seq_lengthvaluecCs||jj_dSr r)rr#rrrr!sfeaturesdict[str, torch.Tensor]c Csg}g}d|vr|jj|dd}|j|d}d|vrC|jj|d|dd|dd|dd|ddd }|j|d}g}t|}t|}t|d D]\} } | d krc|t |qS|t |qSt | |d <|S) N pixel_values)r& input_idsattention_mask position_idsoutput_attentionsoutput_hidden_states)r(r)r*r+r,image_text_inforsentence_embedding) r vision_modelvisual_projection text_modelgettext_projectioniter enumerateappendnexttorchstackfloat) rr$ image_embeds text_embedsvision_outputs text_outputsr.image_features text_featuresidx input_typerrrforward s.    zCLIPModel.forwardpadding str | boolc Csg}g}g}t|D]\}}t|tjr|||dq |||dq i}t|r:|jj||ddd}t|rK|jj|dd} | j|d<||d<t |S) Nrr'Tpt)rD truncationreturn_tensors)rHr&r-) r5 isinstancerr6lenrrimage_processorr&dict) rtextsrDimages texts_valuesr-rAdataencodingr?rrrtokenize@s"      zCLIPModel.tokenizetransformers.CLIPProcessorcCs|jSr )rrrrrrXszCLIPModel.tokenizer output_pathcCs|j||j|dSr )rsave_pretrainedr)rrTrrrsave\s zCLIPModel.save input_pathcCs t|dS)N)r )r)rWrrrload`r"zCLIPModel.load)rN)r r r r )r r )r r)r#rr r )r$r%r r%)T)rDrEr r%)r rS)rTr r r )rWr r r)__name__ __module__ __qualname__r__annotations__rrpropertyr!setterrCrRrrV staticmethodrX __classcell__rrrrr s        r) __future__rr8rPILrrModulerrrrrs