UhddlmZddlmZmZmZddlZddlZddlm Z ddl m Z ddl m Z ddlmZmZd d lmZd d lmZej*eZeGd de ZeGdde ZdgZy)) dataclass)ListOptionalUnionN)Cache)$ImageClassifierOutputWithNoAttention)PreTrainedModel)auto_docstringlogging)AutoModelForImageTextToText)ShieldGemma2Configc:eZdZUdZdZeejed<y)0ShieldGemma2ImageClassifierOutputWithNoAttentionz^ShieldGemma2 classifies imags as violative or not relative to a specific policy Args: N probabilities) __name__ __module__ __qualname____doc__rrtorchTensor__annotations__/var/www/catia.catastroantioquia-mas.com/valormas/lib/python3.12/site-packages/transformers/models/shieldgemma2/modeling_shieldgemma2.pyrr$s-1M8ELL)0rrc eZdZeZdeffd ZdZdZdZdZ dZ dZ d Z e dd eej d eej"d eej$d eej deeeej"efdeej deej deej"deej deedeedeedeedeeej$fdefdZxZS)"ShieldGemma2ForImageClassificationconfigct||t|dd|_t|dd|_t j ||_y)N)r yes_token_indexi *no_token_indexi)super__init__getattrr"r#r from_configmodel)selfr __class__s rr%z+ShieldGemma2ForImageClassification.__init__1sI '&v/@&I%f.>E0<>@@rcN|jjj|yr,)r(r-set_output_embeddings)r)new_embeddingss rr6z8ShieldGemma2ForImageClassification.set_output_embeddings@s !!77GrcN|jjj|yr,)r(r- set_decoder)r)decoders rr9z.ShieldGemma2ForImageClassification.set_decoderCs !!--g6rcJ|jjjSr,)r(r- get_decoderr/s rr<z.ShieldGemma2ForImageClassification.get_decoderFzz((4466rcJ|jjjSr,)r(r- tie_weightsr/s rr?z.ShieldGemma2ForImageClassification.tie_weightsIr=r input_ids pixel_valuesattention_mask position_idspast_key_valuestoken_type_idscache_position inputs_embedslabels use_cacheoutput_attentionsoutput_hidden_states return_dictlogits_to_keepreturnc |jd||||||||| | | | | |d|}|j}|ddd|j|jgf}t j |d}t ||S)aY Returns: A `ShieldGemma2ImageClassifierOutputWithNoAttention` instance containing the logits and probabilities associated with the model predicting the `Yes` or `No` token as the response to that prompt, captured in the following properties. * `logits` (`torch.Tensor` of shape `(batch_size, 2)`): The first position along dim=1 is the logits for the `Yes` token and the second position along dim=1 is the logits for the `No` token. * `probabilities` (`torch.Tensor` of shape `(batch_size, 2)`): The first position along dim=1 is the probability of predicting the `Yes` token and the second position along dim=1 is the probability of predicting the `No` token. ShieldGemma prompts are constructed such that predicting the `Yes` token means the content *does violate* the policy as described. If you are only interested in the violative condition, use `violated = outputs.probabilities[:, 1]` to extract that slice from the output tensors. When used with the `ShieldGemma2Processor`, the `batch_size` will be equal to `len(images) * len(policies)`, and the order within the batch will be img1_policy1, ... img1_policyN, ... imgM_policyN. )r@rArBrCrDrErFrGrHrIrJrKrLrMN)dim)logitsrr)r(rRr"r#rsoftmaxr)r)r@rArBrCrDrErFrGrHrIrJrKrLrM lm_kwargsoutputsrRselected_logitsrs rforwardz*ShieldGemma2ForImageClassification.forwardLsN$** %)%+))'/!5#)  " B)=)=t?R?R(S!ST o2> ?"'  r)NNNNNNNNNNNNNr)rrrr config_classr%r.r1r4r6r9r<r?r rr LongTensor FloatTensorrrrrboolintrrW __classcell__)r*s@rrr-s%LL1L @>AH77715481537KO595959-1$(,0/3&*34= E,,-= u001= !. = u//0 = "%U->->(?(F"GH = !!1!12= !!1!12=   1 12= ))*= D>= $D>= 'tn= d^= c5<</0= " :#= = rr) dataclassesrtypingrrrrtorch.utils.checkpoint cache_utilsrmodeling_outputsr modeling_utilsr utilsr r autorconfiguration_shieldgemma2r get_loggerrloggerrr__all__rrrrjs "((  D-/:   H % 17[1 1\ \ \ @) r