*j$dddlmZddlmZmZddlmZddlmZm Z Gddej Z dS)) annotations)Tensornn) CrossEncoder)batch_to_devicefullnamecPeZdZejfdfd Z dddZdZxZS)MSELossmodelr activation_fn nn.ModulereturnNonec t||_||_t jdi||_t|jts1t|j j dt|jd|jj dkr)t|j j d|jj ddS)a Computes the MSE loss between the computed query-document score and a target query-document score. This loss is used to distill a cross-encoder model from a teacher cross-encoder model or gold labels. Args: model (:class:`~sentence_transformers.cross_encoder.model.CrossEncoder`): A CrossEncoder model to be trained. activation_fn (:class:`~torch.nn.Module`): Activation function applied to the logits before computing the loss. **kwargs: Additional keyword arguments passed to the underlying :class:`torch.nn.MSELoss`. .. note:: Be mindful of the magnitude of both the labels and what the model produces. If the teacher model produces logits with Sigmoid to bound them to [0, 1], then you may wish to use a Sigmoid activation function in the loss. References: - Improving Efficient Neural Ranking Models with Cross-Architecture Knowledge Distillation: https://huggingface.co/papers/2010.02666 - `Cross Encoder > Training Examples > Distillation <../../../examples/cross_encoder/training/distillation/README.html>`_ Requirements: 1. Your model must be initialized with `num_labels = 1` (a.k.a. the default) to predict one class. 2. Usually uses a finetuned CrossEncoder teacher M in a knowledge distillation setup. Inputs: +-----------------------------------------+-----------------------------+-------------------------------+ | Inputs | Labels | Number of Model Output Labels | +=========================================+=============================+===============================+ | (input_A, input_B) pairs | similarity score | 1 | +-----------------------------------------+-----------------------------+-------------------------------+ Relations: - :class:`MarginMSELoss` is similar to this loss, but with a margin through a negative pair. Example: :: from sentence_transformers.cross_encoder import CrossEncoder, CrossEncoderTrainer, losses from datasets import Dataset student_model = CrossEncoder("microsoft/mpnet-base") teacher_model = CrossEncoder("cross-encoder/ms-marco-MiniLM-L12-v2") train_dataset = Dataset.from_dict({ "query": ["What are pandas?", "What is the capital of France?"], "answer": ["Pandas are a kind of bear.", "The capital of France is Paris."], }) def compute_labels(batch): return { "label": teacher_model.predict(list(zip(batch["query"], batch["answer"]))) } train_dataset = train_dataset.map(compute_labels, batched=True) loss = losses.MSELoss(student_model) trainer = CrossEncoderTrainer( model=student_model, train_dataset=train_dataset, loss=loss, ) trainer.train() z? expects a model of type CrossEncoder, but got a model of type .z; expects a model with 1 output label, but got a model with z output labels.N)super__init__r r rr loss_fct isinstancer ValueError __class____name__type num_labels)selfr r kwargsrs \C:\Projects\DSCons\.venv\Lib\site-packages\sentence_transformers\cross_encoder\losses\mse.pyrzMSELoss.__init__ sz  * ,,V,, $*l33 >*??+/ +;+;???  : A % %>*OO(, (=OOO  & %Ninputslist[list[str]]labelsrprompt str | Nonetaskc&t|dkr tdt|dtt|d|d}|j|||}t ||jj}||dd}| |}| || }|S) NzMMSELoss expects a dataset with two non-label columns, but got a dataset with z columns.rr)r$r&scores) lenrlistzipr preprocessrdeviceviewr rdetachfloat)rr!r#r$r&pairslogitslosss rforwardzMSELoss.forwardXs v;;!  v`cdj`k`kvvv SF1I..//&&uV$&GG ):;;F##H-22266##F++}}VV]]__%:%:%<%<== r c.dt|jiS)Nr )rr )rs rget_config_dictzMSELoss.get_config_dicths Xd&899  r )r rr r rr)NN) r!r"r#rr$r%r&r%rr) r __module__ __qualname__rIdentityrr6r8 __classcell__)rs@rr r sGRr{}}LLLLLLL^fj        r r N) __future__rtorchrr)sentence_transformers.cross_encoder.modelrsentence_transformers.utilrrModuler rr rrBs""""""BBBBBB@@@@@@@@b b b b b bib b b b b r