a h8@sddlmZddlZddlmZddlmZddlZddlZddl m Z ddlm Z m Z ddl mZeeZGdd d e jZdS) ) annotationsN)Iterable)Callable)version)Tensornn)SentenceTransformerc sbeZdZdddefdddddddddfd d Zd d d dddZeddddZZ S) SoftmaxLossTFrintboolrNone)modelsentence_embedding_dimension num_labelsconcatenation_sent_repconcatenation_sent_difference!concatenation_sent_multiplicationloss_fctreturnc st||_||_||_||_||_d}|r8|d7}|rD|d7}|rP|d7}td|t j ||||j d|_ ||_ ttjtdkrtddS) at This loss was used in our SBERT publication (https://arxiv.org/abs/1908.10084) to train the SentenceTransformer model on NLI data. It adds a softmax classifier on top of the output of two transformer networks. :class:`MultipleNegativesRankingLoss` is an alternative loss function that often yields better results, as per https://arxiv.org/abs/2004.09813. Args: model (SentenceTransformer): The SentenceTransformer model. sentence_embedding_dimension (int): The dimension of the sentence embeddings. num_labels (int): The number of different labels. concatenation_sent_rep (bool): Whether to concatenate vectors u,v for the softmax classifier. Defaults to True. concatenation_sent_difference (bool): Whether to add abs(u-v) for the softmax classifier. Defaults to True. concatenation_sent_multiplication (bool): Whether to add u*v for the softmax classifier. Defaults to False. loss_fct (Callable): Custom pytorch loss function. If not set, uses nn.CrossEntropyLoss(). Defaults to nn.CrossEntropyLoss(). References: - Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks: https://arxiv.org/abs/1908.10084 - `Training Examples > Natural Language Inference <../../../examples/sentence_transformer/training/nli/README.html>`_ Requirements: 1. sentence pairs with a class label Inputs: +---------------------------------------+--------+ | Texts | Labels | +=======================================+========+ | (sentence_A, sentence_B) pairs | class | +---------------------------------------+--------+ Example: :: from sentence_transformers import SentenceTransformer, SentenceTransformerTrainer, losses from datasets import Dataset model = SentenceTransformer("microsoft/mpnet-base") train_dataset = Dataset.from_dict({ "sentence1": [ "A person on a horse jumps over a broken down airplane.", "A person on a horse jumps over a broken down airplane.", "A person on a horse jumps over a broken down airplane.", "Children smiling and waving at camera", ], "sentence2": [ "A person is training his horse for a competition.", "A person is at a diner, ordering an omelette.", "A person is outdoors, on a horse.", "There are children present.", ], "label": [1, 2, 0, 0], }) loss = losses.SoftmaxLoss(model, model.get_sentence_embedding_dimension(), num_labels=3) trainer = SentenceTransformerTrainer( model=model, train_dataset=train_dataset, loss=loss, ) trainer.train() rz%Softmax loss: #Vectors concatenated: )devicez4.43.0zSoftmaxLoss requires transformers >= 4.43.0 to work correctly. Otherwise, the classifier layer that maps embeddings to the labels cannot be updated. Consider updating transformers with `pip install transformers>=4.43.0`.N)super__init__r rrrrloggerinforZLinearr classifierrrparse transformers __version__warning) selfr rrrrrrZnum_vectors_concatenated __class__d/var/www/html/assistant/venv/lib/python3.9/site-packages/sentence_transformers/losses/SoftmaxLoss.pyrs,G  zSoftmaxLoss.__init__zIterable[dict[str, Tensor]]rzTensor | tuple[Tensor, Tensor])sentence_featureslabelsrc sfdd|D}|\}}g}jr8||||jrR|t||jrf|||t|d}|}|dur|| d} | S||fSdS)Ncsg|]}|dqS)Zsentence_embedding)r ).0Zsentence_featurer!r$r% wz'SoftmaxLoss.forward..r) rappendrtorchabsrcatrrview) r!r&r'ZrepsZrep_aZrep_bZvectors_concatfeaturesoutputZlossr$r)r%forwardts     zSoftmaxLoss.forwardstr)rcCsdS)Na @inproceedings{reimers-2019-sentence-bert, title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks", author = "Reimers, Nils and Gurevych, Iryna", booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing", month = "11", year = "2019", publisher = "Association for Computational Linguistics", url = "https://arxiv.org/abs/1908.10084", } r$r)r$r$r%citationszSoftmaxLoss.citation) __name__ __module__ __qualname__rZCrossEntropyLossrr4propertyr6 __classcell__r$r$r"r%r s"br ) __future__rloggingcollections.abcrtypingrr.r packagingrrrZ)sentence_transformers.SentenceTransformerr getLoggerr7rModuler r$r$r$r%s