transformers/pr_48314/en/_app/immutable/nodes/120.rlWHO4E4.js

import"../chunks/DsnmJJEf.js";import{i as xe,h as Fe,C as Ge,H as d,b as Re,a as c,D as n,E as De,s as Ne}from"../chunks/CD2rhSaz.js";import{p as Ie,o as qe,s as e,f as p,a as l,b as Xe,c as s,d as h,r as a,n as b}from"../chunks/DmjbnfDo.js";import{T as u}from"../chunks/B2suExpn.js";import{E as m}from"../chunks/B5qO6y4i.js";import{H as _e}from"../chunks/D8p7KaJK.js";const Qe='{"title":"DeBERTa-v2","local":"deberta-v2","sections":[{"title":"DebertaV2Config","local":"transformers.DebertaV2Config","sections":[],"depth":2},{"title":"DebertaV2Tokenizer","local":"transformers.DebertaV2Tokenizer","sections":[],"depth":2},{"title":"DebertaV2Model","local":"transformers.DebertaV2Model","sections":[],"depth":2},{"title":"DebertaV2PreTrainedModel","local":"transformers.DebertaV2PreTrainedModel","sections":[],"depth":2},{"title":"DebertaV2ForMaskedLM","local":"transformers.DebertaV2ForMaskedLM","sections":[],"depth":2},{"title":"DebertaV2ForSequenceClassification","local":"transformers.DebertaV2ForSequenceClassification","sections":[],"depth":2},{"title":"DebertaV2ForTokenClassification","local":"transformers.DebertaV2ForTokenClassification","sections":[],"depth":2},{"title":"DebertaV2ForQuestionAnswering","local":"transformers.DebertaV2ForQuestionAnswering","sections":[],"depth":2},{"title":"DebertaV2ForMultipleChoice","local":"transformers.DebertaV2ForMultipleChoice","sections":[],"depth":2}],"depth":1}';var Se=h('<meta name="hf:doc:metadata"/>'),Ye=h("<!> <!>",1),f=h("<p>Example:</p> <!>",1),g=h(`<p>Although the recipe for forward pass needs to be defined within this function, one should call the <code>Module</code> instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.</p>`),Ee=h("<p>Example of single-label classification:</p> <!>",1),Ae=h("<p>Example of multi-label classification:</p> <!>",1),He=h(`<p></p> <p><em>This model was published in HF papers on 2020-06-05 and contributed to Hugging Face Transformers on 2021-02-19.</em></p> <!> <div style="float: right;"><div class="flex flex-wrap space-x-1"><img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&amp;logo=pytorch&amp;logoColor=white"/></div></div> <!> <p><a href="https://huggingface.co/papers/2006.03654" rel="nofollow">DeBERTa-v2</a> improves on the original <a href="./deberta">DeBERTa</a> architecture by using a SentencePiece-based tokenizer and a new vocabulary size of 128K. It also adds an additional convolutional layer within the first transformer layer to better learn local dependencies of input tokens. Finally, the position projection and content projection matrices are shared in the attention layer to reduce the number of parameters.</p> <p>You can find all the original [DeBERTa-v2] checkpoints under the <a href="https://huggingface.co/microsoft?search_models=deberta-v2" rel="nofollow">Microsoft</a> organization.</p> <blockquote class="tip"><p>This model was contributed by <a href="https://huggingface.co/DeBERTa" rel="nofollow">Pengcheng He</a>.</p> <p>Click on the DeBERTa-v2 models in the right sidebar for more examples of how to apply DeBERTa-v2 to different language tasks.</p></blockquote> <p>The example below demonstrates how to classify text with <a href="/docs/transformers/pr_48314/en/main_classes/pipelines#transformers.Pipeline">Pipeline</a> or the <a href="/docs/transformers/pr_48314/en/model_doc/auto#transformers.AutoModel">AutoModel</a> class.</p> <!> <p>Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the <a href="../quantization/overview">Quantization</a> overview for more available quantization backends.</p> <p>The example below uses <a href="../quantization/bitsandbytes">bitsandbytes quantization</a> to only quantize the weights to 4-bit.</p> <!> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>This is the configuration class to store the configuration of a Deberta V2Model. It is used to instantiate a Deberta V2
model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
defaults will yield a similar configuration to that of the <a href="https://huggingface.co/microsoft/deberta-v2-xlarge" rel="nofollow">microsoft/deberta-v2-xlarge</a></p> <p>Configuration objects inherit from <a href="/docs/transformers/pr_48314/en/main_classes/configuration#transformers.PreTrainedConfig">PreTrainedConfig</a> and can be used to control the model outputs. Read the
documentation from <a href="/docs/transformers/pr_48314/en/main_classes/configuration#transformers.PreTrainedConfig">PreTrainedConfig</a> for more information.</p> <!></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>Construct a DeBERTa-v2 tokenizer (backed by HuggingFace’s <em>tokenizers</em> library). Based on Unigram tokenization.</p> <p>This tokenizer inherits from <a href="/docs/transformers/pr_48314/en/main_classes/tokenizer#transformers.TokenizersBackend">TokenizersBackend</a> which contains most of the main methods. Users should
refer to this superclass for more information regarding those methods.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>Retrieve sequence ids from a token list that has no special tokens added.</p> <p>For fast tokenizers, data collators call this with <code>already_has_special_tokens=True</code> to build a mask over an
already-formatted sequence. In that case, we compute the mask by checking membership in <code>all_special_ids</code>.</p></div> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The bare Deberta V2 Model outputting raw hidden-states without any specific head on top.</p> <p>This model inherits from <a href="/docs/transformers/pr_48314/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass.
Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2Model">DebertaV2Model</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>last_hidden_state</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>) — Sequence of hidden-states at the output of the last layer of the model.</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, +
one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
heads.</p></li></ul></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>This model inherits from <a href="/docs/transformers/pr_48314/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass.
Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>A mock value for a dotted path (e.g. <code>torch.float32</code>): attribute access chains,
calls behave as pass-through decorators, <code>repr</code> is the dotted path, and using it
as a base class substitutes a plain-<code>type</code> base (PEP 560 <code>__mro_entries__</code>), so
real subclasses keep a normal metaclass and <code>inspect.signature</code> reads their real <code>__init__</code> instead of a mock’s.</p></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The Deberta V2 Model with a <code>language modeling</code> head on top.”</p> <p>This model inherits from <a href="/docs/transformers/pr_48314/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass.
Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2ForMaskedLM">DebertaV2ForMaskedLM</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>loss</strong> (<code>torch.FloatTensor</code> of shape <code>(1,)</code>, <em>optional</em>, returned when <code>labels</code> is provided) — Masked language modeling (MLM) loss.</p></li> <li><p><strong>logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, config.vocab_size)</code>) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, +
one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
heads.</p></li></ul> <!></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>DeBERTa Model transformer with a sequence classification/regression head on top (a linear layer on top of the
pooled output) e.g. for GLUE tasks.</p> <p>This model inherits from <a href="/docs/transformers/pr_48314/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass.
Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2ForSequenceClassification">DebertaV2ForSequenceClassification</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>loss</strong> (<code>torch.FloatTensor</code> of shape <code>(1,)</code>, <em>optional</em>, returned when <code>labels</code> is provided) — Classification (or regression if config.num_labels==1) loss.</p></li> <li><p><strong>logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, config.num_labels)</code>) — Classification (or regression if config.num_labels==1) scores (before SoftMax).</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, +
one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
heads.</p></li></ul> <!> <!></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The Deberta V2 transformer with a token classification head on top (a linear layer on top of the hidden-states
output) e.g. for Named-Entity-Recognition (NER) tasks.</p> <p>This model inherits from <a href="/docs/transformers/pr_48314/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass.
Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2ForTokenClassification">DebertaV2ForTokenClassification</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>loss</strong> (<code>torch.FloatTensor</code> of shape <code>(1,)</code>, <em>optional</em>, returned when <code>labels</code> is provided) — Classification loss.</p></li> <li><p><strong>logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, config.num_labels)</code>) — Classification scores (before SoftMax).</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, +
one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
heads.</p></li></ul> <!></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The Deberta V2 transformer with a span classification head on top for extractive question-answering tasks like
SQuAD (a linear layer on top of the hidden-states output to compute <code>span start logits</code> and <code>span end logits</code>).</p> <p>This model inherits from <a href="/docs/transformers/pr_48314/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass.
Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2ForQuestionAnswering">DebertaV2ForQuestionAnswering</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>loss</strong> (<code>torch.FloatTensor</code> of shape <code>(1,)</code>, <em>optional</em>, returned when <code>labels</code> is provided) — Total span extraction loss is the sum of a Cross-Entropy for the start and end positions.</p></li> <li><p><strong>start_logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length)</code>) — Span-start scores (before SoftMax).</p></li> <li><p><strong>end_logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length)</code>) — Span-end scores (before SoftMax).</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, +
one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
heads.</p></li></ul> <!></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The Deberta V2 Model with a multiple choice classification head on top (a linear layer on top of the pooled output and a
softmax) e.g. for RocStories/SWAG tasks.</p> <p>This model inherits from <a href="/docs/transformers/pr_48314/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
etc.)</p> <p>This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass.
Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
and behavior.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2ForMultipleChoice">DebertaV2ForMultipleChoice</a> forward method, overrides the <code>__call__</code> special method.</p> <!> <ul><li><p><strong>loss</strong> (<code>torch.FloatTensor</code> of shape <em>(1,)</em>, <em>optional</em>, returned when <code>labels</code> is provided) — Classification loss.</p></li> <li><p><strong>logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, num_choices)</code>) — <em>num_choices</em> is the second dimension of the input tensors. (see <em>input_ids</em> above).</p> <p>Classification scores (before SoftMax).</p></li> <li><p><strong>hidden_states</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_hidden_states=True</code> is passed or when <code>config.output_hidden_states=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for the output of the embeddings, if the model has an embedding layer, +
one for the output of each layer) of shape <code>(batch_size, sequence_length, hidden_size)</code>.</p> <p>Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.</p></li> <li><p><strong>attentions</strong> (<code>tuple(torch.FloatTensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>torch.FloatTensor</code> (one for each layer) of shape <code>(batch_size, num_heads, sequence_length, sequence_length)</code>.</p> <p>Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
heads.</p></li></ul> <!></div></div> <!> <p></p>`,1);function oo(ye,Me){Ie(Me,!1),qe(()=>{new URLSearchParams(window.location.search).get("fw")}),xe();var z=He();Fe("1cnxdeg",t=>{var r=Se();Ne(r,"content",Qe),l(t,r)});var Z=e(p(z),4);Ge(Z,{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"});var C=e(Z,4);d(C,{title:"DeBERTa-v2",local:"deberta-v2",headingTag:"h1"});var W=e(C,10);Re(W,{id:"usage",options:["Pipeline","AutoModel"],children:(t,r)=>{var o=Ye(),i=p(o);_e(i,{id:"usage",option:"Pipeline",children:(U,Be)=>{c(U,{code:"ZnJvbSUyMHRyYW5zZm9ybWVycyUyMGltcG9ydCUyMHBpcGVsaW5lJTBBJTBBJTBBcGlwZWxpbmUlMjAlM0QlMjBwaXBlbGluZSglMEElMjAlMjAlMjAlMjB0YXNrJTNEJTIydGV4dC1jbGFzc2lmaWNhdGlvbiUyMiUyQyUwQSUyMCUyMCUyMCUyMG1vZGVsJTNEJTIybWljcm9zb2Z0JTJGZGViZXJ0YS12Mi14bGFyZ2UtbW5saSUyMiUyQyUwQSUyMCUyMCUyMCUyMGRldmljZSUzRDAlMkMlMEEpJTBBcmVzdWx0JTIwJTNEJTIwcGlwZWxpbmUoJTIyRGVCRVJUYS12MiUyMGlzJTIwZ3JlYXQlMjBhdCUyMHVuZGVyc3RhbmRpbmclMjBjb250ZXh0ISUyMiklMEFwcmludChyZXN1bHQp",highlighted:`<span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> pipeline


pipeline = pipeline(
    task=<span class="hljs-string">&quot;text-classification&quot;</span>,
    model=<span class="hljs-string">&quot;microsoft/deberta-v2-xlarge-mnli&quot;</span>,
    device=<span class="hljs-number">0</span>,
)
result = pipeline(<span class="hljs-string">&quot;DeBERTa-v2 is great at understanding context!&quot;</span>)
<span class="hljs-built_in">print</span>(result)`,lang:"python",wrap:!1})},$$slots:{default:!0}});var We=e(i,2);_e(We,{id:"usage",option:"AutoModel",children:(U,Be)=>{c(U,{code:"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",highlighted:`<span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForSequenceClassification, AutoTokenizer


tokenizer = AutoTokenizer.from_pretrained(
    <span class="hljs-string">&quot;microsoft/deberta-v2-xlarge-mnli&quot;</span>
)
model = AutoModelForSequenceClassification.from_pretrained(
    <span class="hljs-string">&quot;microsoft/deberta-v2-xlarge-mnli&quot;</span>,
    device_map=<span class="hljs-string">&quot;auto&quot;</span>
)

inputs = tokenizer(<span class="hljs-string">&quot;DeBERTa-v2 is great at understanding context!&quot;</span>, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>).to(model.device)
outputs = model(**inputs)

logits = outputs.logits
predicted_class_id = logits.argmax().item()
predicted_label = model.config.id2label[predicted_class_id]
<span class="hljs-built_in">print</span>(<span class="hljs-string">f&quot;Predicted label: <span class="hljs-subst">{predicted_label}</span>&quot;</span>)`,lang:"python",wrap:!1})},$$slots:{default:!0}}),l(t,o)},$$slots:{default:!0}});var B=e(W,6);c(B,{code:"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",highlighted:`<span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForSequenceClassification, AutoTokenizer, BitsAndBytesConfig

model_id = <span class="hljs-string">&quot;microsoft/deberta-v2-xlarge-mnli&quot;</span>
quantization_config = BitsAndBytesConfig(
    load_in_4bit=<span class="hljs-literal">True</span>,
    bnb_4bit_quant_type=<span class="hljs-string">&quot;nf4&quot;</span>,
    bnb_4bit_compute_dtype=<span class="hljs-string">&quot;float16&quot;</span>,
    bnb_4bit_use_double_quant=<span class="hljs-literal">True</span>,
)
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSequenceClassification.from_pretrained(
    model_id,
    quantization_config=quantization_config,
    dtype=<span class="hljs-string">&quot;float16&quot;</span>
 device_map=<span class="hljs-string">&quot;auto&quot;</span>)

inputs = tokenizer(<span class="hljs-string">&quot;DeBERTa-v2 is great at understanding context!&quot;</span>, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>).to(model.device)
outputs = model(**inputs)
logits = outputs.logits
predicted_class_id = logits.argmax().item()
predicted_label = model.config.id2label[predicted_class_id]
<span class="hljs-built_in">print</span>(<span class="hljs-string">f&quot;Predicted label: <span class="hljs-subst">{predicted_label}</span>&quot;</span>)
`,lang:"py",wrap:!1});var x=e(B,2);d(x,{title:"DebertaV2Config",local:"transformers.DebertaV2Config",headingTag:"h2"});var _=e(x,2),F=s(_);n(F,{name:"class transformers.DebertaV2Config",anchor:"transformers.DebertaV2Config",source:"https://github.com/huggingface/transformers/blob/vr_48314/src/transformers/models/deberta_v2/configuration_deberta_v2.py#L24",parameters:[{name:"transformers_version",val:": str | None = None"},{name:"architectures",val:": list[str] | None = None"},{name:"output_hidden_states",val:": bool | None = False"},{name:"return_dict",val:": bool | None = True"},{name:"dtype",val:": typing.Union[str, ForwardRef('torch.dtype'), NoneType] = None"},{name:"chunk_size_feed_forward",val:": int = 0"},{name:"is_encoder_decoder",val:": bool = False"},{name:"id2label",val:": dict[int, str] | dict[str, str] | None = None"},{name:"label2id",val:": dict[str, int] | dict[str, str] | None = None"},{name:"problem_type",val:": typing.Optional[typing.Literal['regression', 'single_label_classification', 'multi_label_classification']] = None"},{name:"vocab_size",val:": int = 128100"},{name:"hidden_size",val:": int = 1536"},{name:"num_hidden_layers",val:": int = 24"},{name:"num_attention_heads",val:": int = 24"},{name:"intermediate_size",val:": int = 6144"},{name:"hidden_act",val:": str = 'gelu'"},{name:"hidden_dropout_prob",val:": float | int = 0.1"},{name:"attention_probs_dropout_prob",val:": float | int = 0.1"},{name:"max_position_embeddings",val:": int = 512"},{name:"type_vocab_size",val:": int = 0"},{name:"initializer_range",val:": float = 0.02"},{name:"layer_norm_eps",val:": float = 1e-07"},{name:"relative_attention",val:": bool = False"},{name:"max_relative_positions",val:": int = -1"},{name:"pad_token_id",val:": int | None = 0"},{name:"bos_token_id",val:": int | None = None"},{name:"eos_token_id",val:": int | list[int] | None = None"},{name:"position_biased_input",val:": bool = True"},{name:"pos_att_type",val:": str | list[str] | None = None"},{name:"pooler_dropout",val:": float | int = 0.0"},{name:"pooler_hidden_act",val:": str = 'gelu'"},{name:"legacy",val:": bool = True"},{name:"tie_word_embeddings",val:": bool = True"}],parametersDescription:[{anchor:"transformers.DebertaV2Config.vocab_size",description:`<strong>vocab_size</strong> (<code>int</code>, <em>optional</em>, defaults to <code>128100</code>) &#x2014;
Vocabulary size of the model. Defines the number of different tokens that can be represented by the <code>input_ids</code>.`,name:"vocab_size"},{anchor:"transformers.DebertaV2Config.hidden_size",description:`<strong>hidden_size</strong> (<code>int</code>, <em>optional</em>, defaults to <code>1536</code>) &#x2014;
Dimension of the hidden representations.`,name:"hidden_size"},{anchor:"transformers.DebertaV2Config.num_hidden_layers",description:`<strong>num_hidden_layers</strong> (<code>int</code>, <em>optional</em>, defaults to <code>24</code>) &#x2014;
Number of hidden layers in the Transformer decoder.`,name:"num_hidden_layers"},{anchor:"transformers.DebertaV2Config.num_attention_heads",description:`<strong>num_attention_heads</strong> (<code>int</code>, <em>optional</em>, defaults to <code>24</code>) &#x2014;
Number of attention heads for each attention layer in the Transformer decoder.`,name:"num_attention_heads"},{anchor:"transformers.DebertaV2Config.intermediate_size",description:`<strong>intermediate_size</strong> (<code>int</code>, <em>optional</em>, defaults to <code>6144</code>) &#x2014;
Dimension of the MLP representations.`,name:"intermediate_size"},{anchor:"transformers.DebertaV2Config.hidden_act",description:`<strong>hidden_act</strong> (<code>str</code>, <em>optional</em>, defaults to <code>gelu</code>) &#x2014;
The non-linear activation function (function or string) in the decoder. For example, <code>&quot;gelu&quot;</code>,
<code>&quot;relu&quot;</code>, <code>&quot;silu&quot;</code>, etc.`,name:"hidden_act"},{anchor:"transformers.DebertaV2Config.hidden_dropout_prob",description:`<strong>hidden_dropout_prob</strong> (<code>Union[float, int]</code>, <em>optional</em>, defaults to <code>0.1</code>) &#x2014;
The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.`,name:"hidden_dropout_prob"},{anchor:"transformers.DebertaV2Config.attention_probs_dropout_prob",description:`<strong>attention_probs_dropout_prob</strong> (<code>Union[float, int]</code>, <em>optional</em>, defaults to <code>0.1</code>) &#x2014;
The dropout ratio for the attention probabilities.`,name:"attention_probs_dropout_prob"},{anchor:"transformers.DebertaV2Config.max_position_embeddings",description:`<strong>max_position_embeddings</strong> (<code>int</code>, <em>optional</em>, defaults to <code>512</code>) &#x2014;
The maximum sequence length that this model might ever be used with.`,name:"max_position_embeddings"},{anchor:"transformers.DebertaV2Config.type_vocab_size",description:`<strong>type_vocab_size</strong> (<code>int</code>, <em>optional</em>, defaults to <code>0</code>) &#x2014;
The vocabulary size of the <code>token_type_ids</code>.`,name:"type_vocab_size"},{anchor:"transformers.DebertaV2Config.initializer_range",description:`<strong>initializer_range</strong> (<code>float</code>, <em>optional</em>, defaults to <code>0.02</code>) &#x2014;
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.`,name:"initializer_range"},{anchor:"transformers.DebertaV2Config.layer_norm_eps",description:`<strong>layer_norm_eps</strong> (<code>float</code>, <em>optional</em>, defaults to <code>1e-07</code>) &#x2014;
The epsilon used by the layer normalization layers.`,name:"layer_norm_eps"},{anchor:"transformers.DebertaV2Config.relative_attention",description:`<strong>relative_attention</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) &#x2014;
Whether use relative position encoding.`,name:"relative_attention"},{anchor:"transformers.DebertaV2Config.max_relative_positions",description:`<strong>max_relative_positions</strong> (<code>int</code>, <em>optional</em>, defaults to -1) &#x2014;
The range of relative positions <code>[-max_position_embeddings, max_position_embeddings]</code>. Use the same value
as <code>max_position_embeddings</code>.`,name:"max_relative_positions"},{anchor:"transformers.DebertaV2Config.pad_token_id",description:`<strong>pad_token_id</strong> (<code>int</code>, <em>optional</em>, defaults to <code>0</code>) &#x2014;
Token id used for padding in the vocabulary.`,name:"pad_token_id"},{anchor:"transformers.DebertaV2Config.bos_token_id",description:`<strong>bos_token_id</strong> (<code>int</code>, <em>optional</em>) &#x2014;
Token id used for beginning-of-stream in the vocabulary.`,name:"bos_token_id"},{anchor:"transformers.DebertaV2Config.eos_token_id",description:`<strong>eos_token_id</strong> (<code>Union[int, list[int]]</code>, <em>optional</em>) &#x2014;
Token id used for end-of-stream in the vocabulary.`,name:"eos_token_id"},{anchor:"transformers.DebertaV2Config.position_biased_input",description:`<strong>position_biased_input</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) &#x2014;
Whether add absolute position embedding to content embedding.`,name:"position_biased_input"},{anchor:"transformers.DebertaV2Config.pos_att_type",description:`<strong>pos_att_type</strong> (<code>list[str]</code>, <em>optional</em>) &#x2014;
The type of relative position attention, it can be a combination of <code>[&quot;p2c&quot;, &quot;c2p&quot;]</code>, e.g. <code>[&quot;p2c&quot;]</code>,
<code>[&quot;p2c&quot;, &quot;c2p&quot;]</code>, <code>[&quot;p2c&quot;, &quot;c2p&quot;]</code>.`,name:"pos_att_type"},{anchor:"transformers.DebertaV2Config.pooler_dropout",description:`<strong>pooler_dropout</strong> (<code>float</code>, <em>optional</em>, defaults to <code>0</code>) &#x2014;
Dropout rate in the pooler module.`,name:"pooler_dropout"},{anchor:"transformers.DebertaV2Config.pooler_hidden_act",description:`<strong>pooler_hidden_act</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;gelu&quot;</code>) &#x2014;
Activation function used in the dropout module.`,name:"pooler_hidden_act"},{anchor:"transformers.DebertaV2Config.legacy",description:`<strong>legacy</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) &#x2014;
Whether or not the model should use the legacy <code>LegacyDebertaOnlyMLMHead</code>, which does not work properly
for mask infilling tasks.`,name:"legacy"},{anchor:"transformers.DebertaV2Config.tie_word_embeddings",description:`<strong>tie_word_embeddings</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) &#x2014;
Whether to tie weight embeddings according to model&#x2019;s <code>tied_weights_keys</code> mapping.`,name:"tie_word_embeddings"}]});var Te=e(F,6);m(Te,{anchor:"transformers.DebertaV2Config.example",children:(t,r)=>{var o=f(),i=e(p(o),2);c(i,{code:"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",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> DebertaV2Config, DebertaV2Model

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># Initializing a DeBERTa-v2 microsoft/deberta-v2-xlarge style configuration</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>configuration = DebertaV2Config()

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># Initializing a model (with random weights) from the microsoft/deberta-v2-xlarge style configuration</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>model = DebertaV2Model(configuration)

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># Accessing the model configuration</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>configuration = model.config`,lang:"python",wrap:!1}),l(t,o)},$$slots:{default:!0}}),a(_);var G=e(_,2);d(G,{title:"DebertaV2Tokenizer",local:"transformers.DebertaV2Tokenizer",headingTag:"h2"});var y=e(G,2),R=s(y);n(R,{name:"class transformers.DebertaV2Tokenizer",anchor:"transformers.DebertaV2Tokenizer",source:"https://github.com/huggingface/transformers/blob/vr_48314/src/transformers/models/deberta_v2/tokenization_deberta_v2.py#L28",parameters:[{name:"vocab",val:": str | dict | list | None = None"},{name:"do_lower_case",val:" = False"},{name:"split_by_punct",val:" = False"},{name:"bos_token",val:" = '[CLS]'"},{name:"eos_token",val:" = '[SEP]'"},{name:"unk_token",val:" = '[UNK]'"},{name:"sep_token",val:" = '[SEP]'"},{name:"pad_token",val:" = '[PAD]'"},{name:"cls_token",val:" = '[CLS]'"},{name:"mask_token",val:" = '[MASK]'"},{name:"add_prefix_space",val:" = True"},{name:"unk_id",val:" = 1"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.DebertaV2Tokenizer.vocab_file",description:`<strong>vocab_file</strong> (<code>str</code>, <em>optional</em>) &#x2014;
Path to the vocabulary file (SentencePiece model file). Not used directly but kept for compatibility.`,name:"vocab_file"},{anchor:"transformers.DebertaV2Tokenizer.vocab",description:`<strong>vocab</strong> (<code>str</code>, <code>dict</code> or <code>list</code>, <em>optional</em>) &#x2014;
List of tuples (piece, score) for the vocabulary.`,name:"vocab"},{anchor:"transformers.DebertaV2Tokenizer.precompiled_charsmap",description:`<strong>precompiled_charsmap</strong> (<code>bytes</code>, <em>optional</em>) &#x2014;
Precompiled character map for normalization.`,name:"precompiled_charsmap"},{anchor:"transformers.DebertaV2Tokenizer.do_lower_case",description:`<strong>do_lower_case</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) &#x2014;
Whether or not to lowercase the input when tokenizing.`,name:"do_lower_case"},{anchor:"transformers.DebertaV2Tokenizer.split_by_punct",description:`<strong>split_by_punct</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) &#x2014;
Whether to split by punctuation.`,name:"split_by_punct"},{anchor:"transformers.DebertaV2Tokenizer.bos_token",description:`<strong>bos_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;[CLS]&quot;</code>) &#x2014;
The beginning of sequence token.`,name:"bos_token"},{anchor:"transformers.DebertaV2Tokenizer.eos_token",description:`<strong>eos_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;[SEP]&quot;</code>) &#x2014;
The end of sequence token.`,name:"eos_token"},{anchor:"transformers.DebertaV2Tokenizer.unk_token",description:`<strong>unk_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;[UNK]&quot;</code>) &#x2014;
The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this
token instead.`,name:"unk_token"},{anchor:"transformers.DebertaV2Tokenizer.sep_token",description:`<strong>sep_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;[SEP]&quot;</code>) &#x2014;
The separator token, which is used when building a sequence from multiple sequences, e.g. two sequences for
sequence classification or for a text and a question for question answering. It is also used as the last
token of a sequence built with special tokens.`,name:"sep_token"},{anchor:"transformers.DebertaV2Tokenizer.pad_token",description:`<strong>pad_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;[PAD]&quot;</code>) &#x2014;
The token used for padding, for example when batching sequences of different lengths.`,name:"pad_token"},{anchor:"transformers.DebertaV2Tokenizer.cls_token",description:`<strong>cls_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;[CLS]&quot;</code>) &#x2014;
The classifier token which is used when doing sequence classification (classification of the whole sequence
instead of per-token classification). It is the first token of the sequence when built with special tokens.`,name:"cls_token"},{anchor:"transformers.DebertaV2Tokenizer.mask_token",description:`<strong>mask_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;[MASK]&quot;</code>) &#x2014;
The token used for masking values. This is the token used when training this model with masked language
modeling. This is the token which the model will try to predict.`,name:"mask_token"},{anchor:"transformers.DebertaV2Tokenizer.add_prefix_space",description:`<strong>add_prefix_space</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) &#x2014;
Whether or not to add an initial space to the input. This allows to treat the leading word just as any
other word.`,name:"add_prefix_space"},{anchor:"transformers.DebertaV2Tokenizer.unk_id",description:`<strong>unk_id</strong> (<code>int</code>, <em>optional</em>, defaults to index of <code>unk_token</code> in vocab) &#x2014;
The ID of the unknown token in the vocabulary.`,name:"unk_id"}]});var M=e(R,6),ke=s(M);n(ke,{name:"get_special_tokens_mask",anchor:"transformers.DebertaV2Tokenizer.get_special_tokens_mask",source:"https://github.com/huggingface/transformers/blob/vr_48314/src/transformers/tokenization_utils_base.py#L1311",parameters:[{name:"token_ids_0",val:": list[int]"},{name:"token_ids_1",val:": list[int] | None = None"},{name:"already_has_special_tokens",val:": bool = False"}],parametersDescription:[{anchor:"transformers.DebertaV2Tokenizer.get_special_tokens_mask.token_ids_0",description:"<strong>token_ids_0</strong> &#x2014; List of IDs for the (possibly already formatted) sequence.",name:"token_ids_0"},{anchor:"transformers.DebertaV2Tokenizer.get_special_tokens_mask.token_ids_1",description:"<strong>token_ids_1</strong> &#x2014; Unused when <code>already_has_special_tokens=True</code>. Must be None in that case.",name:"token_ids_1"},{anchor:"transformers.DebertaV2Tokenizer.get_special_tokens_mask.already_has_special_tokens",description:"<strong>already_has_special_tokens</strong> &#x2014; Whether the sequence is already formatted with special tokens.",name:"already_has_special_tokens"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>1 for a special token, 0 for a sequence token.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A list of integers in the range [0, 1]</p>
`}),b(4),a(M);var D=e(M,2),we=s(D);n(we,{name:"save_vocabulary",anchor:"transformers.DebertaV2Tokenizer.save_vocabulary",source:"https://github.com/huggingface/transformers/blob/vr_48314/src/transformers/tokenization_utils_tokenizers.py#L509",parameters:[{name:"save_directory",val:": str"},{name:"filename_prefix",val:": str | None = None"}]}),a(D),a(y);var N=e(y,2);d(N,{title:"DebertaV2Model",local:"transformers.DebertaV2Model",headingTag:"h2"});var T=e(N,2),I=s(T);n(I,{name:"class transformers.DebertaV2Model",anchor:"transformers.DebertaV2Model",source:"https://github.com/huggingface/transformers/blob/vr_48314/src/transformers/models/deberta_v2/modeling_deberta_v2.py#L702",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.DebertaV2Model.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2Model">DebertaV2Model</a>) &#x2014;
Model configuration class with all the parameters of the model. Initializing with a config file does not
load the weights associated with the model, only the configuration. Check out the
<a href="/docs/transformers/pr_48314/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var q=e(I,8),X=s(q);n(X,{name:"forward",anchor:"transformers.DebertaV2Model.forward",source:"https://github.com/huggingface/transformers/blob/vr_48314/src/transformers/models/deberta_v2/modeling_deberta_v2.py#L719",parameters:[{name:"input_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"position_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": bool | None = None"},{name:"output_hidden_states",val:": bool | None = None"},{name:"return_dict",val:": bool | None = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.DebertaV2Model.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p>
<p>Indices can be obtained using <a href="/docs/transformers/pr_48314/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_48314/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/pr_48314/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p>
<p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.DebertaV2Model.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p>
<ul>
<li>1 for tokens that are <strong>not masked</strong>,</li>
<li>0 for tokens that are <strong>masked</strong>.</li>
</ul>
<p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.DebertaV2Model.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p>
<ul>
<li>0 corresponds to a <em>sentence A</em> token,</li>
<li>1 corresponds to a <em>sentence B</em> token.</li>
</ul>
<p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.DebertaV2Model.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Indices of positions of each input sequence tokens in the position embeddings. Selected in the range <code>[0, config.n_positions - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.DebertaV2Model.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) &#x2014;
Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This
is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.DebertaV2Model.forward.output_attentions",description:`<strong>output_attentions</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return the attentions tensors of all attention layers. See <code>attentions</code> under returned
tensors for more detail.`,name:"output_attentions"},{anchor:"transformers.DebertaV2Model.forward.output_hidden_states",description:`<strong>output_hidden_states</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return the hidden states of all layers. See <code>hidden_states</code> under returned tensors for
more detail.`,name:"output_hidden_states"},{anchor:"transformers.DebertaV2Model.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_48314/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/pr_48314/en/main_classes/output#transformers.modeling_outputs.BaseModelOutput"
>BaseModelOutput</a> or a tuple of
<code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various
elements depending on the configuration (<a
  href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2Config"
>DebertaV2Config</a>) and inputs.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/pr_48314/en/main_classes/output#transformers.modeling_outputs.BaseModelOutput"
>BaseModelOutput</a> or <code>tuple(torch.FloatTensor)</code></p>
`});var ve=e(X,4);u(ve,{children:(t,r)=>{var o=g();l(t,o)},$$slots:{default:!0}}),b(2),a(q),a(T);var Q=e(T,2);d(Q,{title:"DebertaV2PreTrainedModel",local:"transformers.DebertaV2PreTrainedModel",headingTag:"h2"});var k=e(Q,2),S=s(k);n(S,{name:"class transformers.DebertaV2PreTrainedModel",anchor:"transformers.DebertaV2PreTrainedModel",source:"https://github.com/huggingface/transformers/blob/vr_48314/src/transformers/models/deberta_v2/modeling_deberta_v2.py#L684",parameters:[{name:"config",val:": PreTrainedConfig"},{name:"*inputs",val:""},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.DebertaV2PreTrainedModel.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_48314/en/main_classes/configuration#transformers.PreTrainedConfig">PreTrainedConfig</a>) &#x2014;
Model configuration class with all the parameters of the model. Initializing with a config file does not
load the weights associated with the model, only the configuration. Check out the
<a href="/docs/transformers/pr_48314/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var Y=e(S,6),Je=s(Y);n(Je,{name:"forward",anchor:"transformers.DebertaV2PreTrainedModel.forward",parameters:[{name:"*args",val:""},{name:"**kwargs",val:""}]}),b(2),a(Y),a(k);var E=e(k,2);d(E,{title:"DebertaV2ForMaskedLM",local:"transformers.DebertaV2ForMaskedLM",headingTag:"h2"});var w=e(E,2),A=s(w);n(A,{name:"class transformers.DebertaV2ForMaskedLM",anchor:"transformers.DebertaV2ForMaskedLM",source:"https://github.com/huggingface/transformers/blob/vr_48314/src/transformers/models/deberta_v2/modeling_deberta_v2.py#L887",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.DebertaV2ForMaskedLM.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2ForMaskedLM">DebertaV2ForMaskedLM</a>) &#x2014;
Model configuration class with all the parameters of the model. Initializing with a config file does not
load the weights associated with the model, only the configuration. Check out the
<a href="/docs/transformers/pr_48314/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var H=e(A,8),P=s(H);n(P,{name:"forward",anchor:"transformers.DebertaV2ForMaskedLM.forward",source:"https://github.com/huggingface/transformers/blob/vr_48314/src/transformers/models/deberta_v2/modeling_deberta_v2.py#L922",parameters:[{name:"input_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"position_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.Tensor] = None"},{name:"labels",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": bool | None = None"},{name:"output_hidden_states",val:": bool | None = None"},{name:"return_dict",val:": bool | None = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.DebertaV2ForMaskedLM.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p>
<p>Indices can be obtained using <a href="/docs/transformers/pr_48314/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_48314/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/pr_48314/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p>
<p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.DebertaV2ForMaskedLM.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p>
<ul>
<li>1 for tokens that are <strong>not masked</strong>,</li>
<li>0 for tokens that are <strong>masked</strong>.</li>
</ul>
<p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.DebertaV2ForMaskedLM.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p>
<ul>
<li>0 corresponds to a <em>sentence A</em> token,</li>
<li>1 corresponds to a <em>sentence B</em> token.</li>
</ul>
<p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.DebertaV2ForMaskedLM.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Indices of positions of each input sequence tokens in the position embeddings. Selected in the range <code>[0, config.n_positions - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.DebertaV2ForMaskedLM.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) &#x2014;
Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This
is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.DebertaV2ForMaskedLM.forward.labels",description:`<strong>labels</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Labels for computing the masked language modeling loss. Indices should be in <code>[-100, 0, ..., config.vocab_size]</code> (see <code>input_ids</code> docstring) Tokens with indices set to <code>-100</code> are ignored (masked), the
loss is only computed for the tokens with labels in <code>[0, ..., config.vocab_size]</code>`,name:"labels"},{anchor:"transformers.DebertaV2ForMaskedLM.forward.output_attentions",description:`<strong>output_attentions</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return the attentions tensors of all attention layers. See <code>attentions</code> under returned
tensors for more detail.`,name:"output_attentions"},{anchor:"transformers.DebertaV2ForMaskedLM.forward.output_hidden_states",description:`<strong>output_hidden_states</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return the hidden states of all layers. See <code>hidden_states</code> under returned tensors for
more detail.`,name:"output_hidden_states"},{anchor:"transformers.DebertaV2ForMaskedLM.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_48314/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/pr_48314/en/main_classes/output#transformers.modeling_outputs.MaskedLMOutput"
>MaskedLMOutput</a> or a tuple of
<code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various
elements depending on the configuration (<a
  href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2Config"
>DebertaV2Config</a>) and inputs.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/pr_48314/en/main_classes/output#transformers.modeling_outputs.MaskedLMOutput"
>MaskedLMOutput</a> or <code>tuple(torch.FloatTensor)</code></p>
`});var L=e(P,4);u(L,{children:(t,r)=>{var o=g();l(t,o)},$$slots:{default:!0}});var je=e(L,4);m(je,{anchor:"transformers.DebertaV2ForMaskedLM.forward.example",children:(t,r)=>{var o=f(),i=e(p(o),2);c(i,{code:"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",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, DebertaV2ForMaskedLM
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> torch

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-v2-xlarge&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = DebertaV2ForMaskedLM.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-v2-xlarge&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = tokenizer(<span class="hljs-string">&quot;The capital of France is &lt;mask&gt;.&quot;</span>, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">with</span> torch.no_grad():
<span class="hljs-meta">... </span>    logits = model(**inputs).logits

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># retrieve index of &lt;mask&gt;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>mask_token_index = (inputs.input_ids == tokenizer.mask_token_id)[<span class="hljs-number">0</span>].nonzero(as_tuple=<span class="hljs-literal">True</span>)[<span class="hljs-number">0</span>]

<span class="hljs-meta">&gt;&gt;&gt; </span>predicted_token_id = logits[<span class="hljs-number">0</span>, mask_token_index].argmax(axis=-<span class="hljs-number">1</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer.decode(predicted_token_id)
...

<span class="hljs-meta">&gt;&gt;&gt; </span>labels = tokenizer(<span class="hljs-string">&quot;The capital of France is Paris.&quot;</span>, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>)[<span class="hljs-string">&quot;input_ids&quot;</span>]
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># mask labels of non-&lt;mask&gt; tokens</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>labels = torch.where(inputs.input_ids == tokenizer.mask_token_id, labels, -<span class="hljs-number">100</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>outputs = model(**inputs, labels=labels)
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-built_in">round</span>(outputs.loss.item(), <span class="hljs-number">2</span>)
...`,lang:"python",wrap:!1}),l(t,o)},$$slots:{default:!0}}),a(H),a(w);var O=e(w,2);d(O,{title:"DebertaV2ForSequenceClassification",local:"transformers.DebertaV2ForSequenceClassification",headingTag:"h2"});var v=e(O,2),K=s(v);n(K,{name:"class transformers.DebertaV2ForSequenceClassification",anchor:"transformers.DebertaV2ForSequenceClassification",source:"https://github.com/huggingface/transformers/blob/vr_48314/src/transformers/models/deberta_v2/modeling_deberta_v2.py#L1009",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.DebertaV2ForSequenceClassification.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2ForSequenceClassification">DebertaV2ForSequenceClassification</a>) &#x2014;
Model configuration class with all the parameters of the model. Initializing with a config file does not
load the weights associated with the model, only the configuration. Check out the
<a href="/docs/transformers/pr_48314/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var $=e(K,8),ee=s($);n(ee,{name:"forward",anchor:"transformers.DebertaV2ForSequenceClassification.forward",source:"https://github.com/huggingface/transformers/blob/vr_48314/src/transformers/models/deberta_v2/modeling_deberta_v2.py#L1034",parameters:[{name:"input_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"position_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.Tensor] = None"},{name:"labels",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": bool | None = None"},{name:"output_hidden_states",val:": bool | None = None"},{name:"return_dict",val:": bool | None = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.DebertaV2ForSequenceClassification.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p>
<p>Indices can be obtained using <a href="/docs/transformers/pr_48314/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_48314/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/pr_48314/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p>
<p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.DebertaV2ForSequenceClassification.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p>
<ul>
<li>1 for tokens that are <strong>not masked</strong>,</li>
<li>0 for tokens that are <strong>masked</strong>.</li>
</ul>
<p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.DebertaV2ForSequenceClassification.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p>
<ul>
<li>0 corresponds to a <em>sentence A</em> token,</li>
<li>1 corresponds to a <em>sentence B</em> token.</li>
</ul>
<p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.DebertaV2ForSequenceClassification.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Indices of positions of each input sequence tokens in the position embeddings. Selected in the range <code>[0, config.n_positions - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.DebertaV2ForSequenceClassification.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) &#x2014;
Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This
is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.DebertaV2ForSequenceClassification.forward.labels",description:`<strong>labels</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size,)</code>, <em>optional</em>) &#x2014;
Labels for computing the sequence classification/regression loss. Indices should be in <code>[0, ..., config.num_labels - 1]</code>. If <code>config.num_labels == 1</code> a regression loss is computed (Mean-Square loss), If
<code>config.num_labels &gt; 1</code> a classification loss is computed (Cross-Entropy).`,name:"labels"},{anchor:"transformers.DebertaV2ForSequenceClassification.forward.output_attentions",description:`<strong>output_attentions</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return the attentions tensors of all attention layers. See <code>attentions</code> under returned
tensors for more detail.`,name:"output_attentions"},{anchor:"transformers.DebertaV2ForSequenceClassification.forward.output_hidden_states",description:`<strong>output_hidden_states</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return the hidden states of all layers. See <code>hidden_states</code> under returned tensors for
more detail.`,name:"output_hidden_states"},{anchor:"transformers.DebertaV2ForSequenceClassification.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_48314/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/pr_48314/en/main_classes/output#transformers.modeling_outputs.SequenceClassifierOutput"
>SequenceClassifierOutput</a> or a tuple of
<code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various
elements depending on the configuration (<a
  href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2Config"
>DebertaV2Config</a>) and inputs.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/pr_48314/en/main_classes/output#transformers.modeling_outputs.SequenceClassifierOutput"
>SequenceClassifierOutput</a> or <code>tuple(torch.FloatTensor)</code></p>
`});var oe=e(ee,4);u(oe,{children:(t,r)=>{var o=g();l(t,o)},$$slots:{default:!0}});var te=e(oe,4);m(te,{anchor:"transformers.DebertaV2ForSequenceClassification.forward.example",children:(t,r)=>{var o=Ee(),i=e(p(o),2);c(i,{code:"aW1wb3J0JTIwdG9yY2glMEFmcm9tJTIwdHJhbnNmb3JtZXJzJTIwaW1wb3J0JTIwQXV0b1Rva2VuaXplciUyQyUyMERlYmVydGFWMkZvclNlcXVlbmNlQ2xhc3NpZmljYXRpb24lMEElMEF0b2tlbml6ZXIlMjAlM0QlMjBBdXRvVG9rZW5pemVyLmZyb21fcHJldHJhaW5lZCglMjJtaWNyb3NvZnQlMkZkZWJlcnRhLXYyLXhsYXJnZSUyMiklMEFtb2RlbCUyMCUzRCUyMERlYmVydGFWMkZvclNlcXVlbmNlQ2xhc3NpZmljYXRpb24uZnJvbV9wcmV0cmFpbmVkKCUyMm1pY3Jvc29mdCUyRmRlYmVydGEtdjIteGxhcmdlJTIyKSUwQSUwQWlucHV0cyUyMCUzRCUyMHRva2VuaXplciglMjJIZWxsbyUyQyUyMG15JTIwZG9nJTIwaXMlMjBjdXRlJTIyJTJDJTIwcmV0dXJuX3RlbnNvcnMlM0QlMjJwdCUyMiklMEElMEF3aXRoJTIwdG9yY2gubm9fZ3JhZCgpJTNBJTBBJTIwJTIwJTIwJTIwbG9naXRzJTIwJTNEJTIwbW9kZWwoKippbnB1dHMpLmxvZ2l0cyUwQSUwQXByZWRpY3RlZF9jbGFzc19pZCUyMCUzRCUyMGxvZ2l0cy5hcmdtYXgoKS5pdGVtKCklMEFtb2RlbC5jb25maWcuaWQybGFiZWwlNUJwcmVkaWN0ZWRfY2xhc3NfaWQlNUQlMEElMEElMjMlMjBUbyUyMHRyYWluJTIwYSUyMG1vZGVsJTIwb24lMjAlNjBudW1fbGFiZWxzJTYwJTIwY2xhc3NlcyUyQyUyMHlvdSUyMGNhbiUyMHBhc3MlMjAlNjBudW1fbGFiZWxzJTNEbnVtX2xhYmVscyU2MCUyMHRvJTIwJTYwLmZyb21fcHJldHJhaW5lZCguLi4pJTYwJTBBbnVtX2xhYmVscyUyMCUzRCUyMGxlbihtb2RlbC5jb25maWcuaWQybGFiZWwpJTBBbW9kZWwlMjAlM0QlMjBEZWJlcnRhVjJGb3JTZXF1ZW5jZUNsYXNzaWZpY2F0aW9uLmZyb21fcHJldHJhaW5lZCglMjJtaWNyb3NvZnQlMkZkZWJlcnRhLXYyLXhsYXJnZSUyMiUyQyUyMG51bV9sYWJlbHMlM0RudW1fbGFiZWxzKSUwQSUwQWxhYmVscyUyMCUzRCUyMHRvcmNoLnRlbnNvciglNUIxJTVEKSUwQWxvc3MlMjAlM0QlMjBtb2RlbCgqKmlucHV0cyUyQyUyMGxhYmVscyUzRGxhYmVscykubG9zcyUwQXJvdW5kKGxvc3MuaXRlbSgpJTJDJTIwMik=",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> torch
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, DebertaV2ForSequenceClassification

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-v2-xlarge&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = DebertaV2ForSequenceClassification.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-v2-xlarge&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = tokenizer(<span class="hljs-string">&quot;Hello, my dog is cute&quot;</span>, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">with</span> torch.no_grad():
<span class="hljs-meta">... </span>    logits = model(**inputs).logits

<span class="hljs-meta">&gt;&gt;&gt; </span>predicted_class_id = logits.argmax().item()
<span class="hljs-meta">&gt;&gt;&gt; </span>model.config.id2label[predicted_class_id]
...

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># To train a model on \`num_labels\` classes, you can pass \`num_labels=num_labels\` to \`.from_pretrained(...)\`</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>num_labels = <span class="hljs-built_in">len</span>(model.config.id2label)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = DebertaV2ForSequenceClassification.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-v2-xlarge&quot;</span>, num_labels=num_labels)

<span class="hljs-meta">&gt;&gt;&gt; </span>labels = torch.tensor([<span class="hljs-number">1</span>])
<span class="hljs-meta">&gt;&gt;&gt; </span>loss = model(**inputs, labels=labels).loss
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-built_in">round</span>(loss.item(), <span class="hljs-number">2</span>)
...`,lang:"python",wrap:!1}),l(t,o)},$$slots:{default:!0}});var Ve=e(te,2);m(Ve,{anchor:"transformers.DebertaV2ForSequenceClassification.forward.example-2",children:(t,r)=>{var o=Ae(),i=e(p(o),2);c(i,{code:"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",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> torch
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, DebertaV2ForSequenceClassification

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-v2-xlarge&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = DebertaV2ForSequenceClassification.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-v2-xlarge&quot;</span>, problem_type=<span class="hljs-string">&quot;multi_label_classification&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = tokenizer(<span class="hljs-string">&quot;Hello, my dog is cute&quot;</span>, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">with</span> torch.no_grad():
<span class="hljs-meta">... </span>    logits = model(**inputs).logits

<span class="hljs-meta">&gt;&gt;&gt; </span>predicted_class_ids = torch.arange(<span class="hljs-number">0</span>, logits.shape[-<span class="hljs-number">1</span>])[torch.sigmoid(logits).squeeze(dim=<span class="hljs-number">0</span>) &gt; <span class="hljs-number">0.5</span>]

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># To train a model on \`num_labels\` classes, you can pass \`num_labels=num_labels\` to \`.from_pretrained(...)\`</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>num_labels = <span class="hljs-built_in">len</span>(model.config.id2label)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = DebertaV2ForSequenceClassification.from_pretrained(
<span class="hljs-meta">... </span>    <span class="hljs-string">&quot;microsoft/deberta-v2-xlarge&quot;</span>, num_labels=num_labels, problem_type=<span class="hljs-string">&quot;multi_label_classification&quot;</span>
<span class="hljs-meta">... </span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>labels = torch.<span class="hljs-built_in">sum</span>(
<span class="hljs-meta">... </span>    torch.nn.functional.one_hot(predicted_class_ids[<span class="hljs-literal">None</span>, :].clone(), num_classes=num_labels), dim=<span class="hljs-number">1</span>
<span class="hljs-meta">... </span>).to(torch.<span class="hljs-built_in">float</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>loss = model(**inputs, labels=labels).loss`,lang:"python",wrap:!1}),l(t,o)},$$slots:{default:!0}}),a($),a(v);var ne=e(v,2);d(ne,{title:"DebertaV2ForTokenClassification",local:"transformers.DebertaV2ForTokenClassification",headingTag:"h2"});var J=e(ne,2),se=s(J);n(se,{name:"class transformers.DebertaV2ForTokenClassification",anchor:"transformers.DebertaV2ForTokenClassification",source:"https://github.com/huggingface/transformers/blob/vr_48314/src/transformers/models/deberta_v2/modeling_deberta_v2.py#L1119",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.DebertaV2ForTokenClassification.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2ForTokenClassification">DebertaV2ForTokenClassification</a>) &#x2014;
Model configuration class with all the parameters of the model. Initializing with a config file does not
load the weights associated with the model, only the configuration. Check out the
<a href="/docs/transformers/pr_48314/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var ae=e(se,8),re=s(ae);n(re,{name:"forward",anchor:"transformers.DebertaV2ForTokenClassification.forward",source:"https://github.com/huggingface/transformers/blob/vr_48314/src/transformers/models/deberta_v2/modeling_deberta_v2.py#L1131",parameters:[{name:"input_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"position_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.Tensor] = None"},{name:"labels",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": bool | None = None"},{name:"output_hidden_states",val:": bool | None = None"},{name:"return_dict",val:": bool | None = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.DebertaV2ForTokenClassification.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p>
<p>Indices can be obtained using <a href="/docs/transformers/pr_48314/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_48314/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/pr_48314/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p>
<p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.DebertaV2ForTokenClassification.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p>
<ul>
<li>1 for tokens that are <strong>not masked</strong>,</li>
<li>0 for tokens that are <strong>masked</strong>.</li>
</ul>
<p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.DebertaV2ForTokenClassification.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p>
<ul>
<li>0 corresponds to a <em>sentence A</em> token,</li>
<li>1 corresponds to a <em>sentence B</em> token.</li>
</ul>
<p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.DebertaV2ForTokenClassification.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Indices of positions of each input sequence tokens in the position embeddings. Selected in the range <code>[0, config.n_positions - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.DebertaV2ForTokenClassification.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) &#x2014;
Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This
is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.DebertaV2ForTokenClassification.forward.labels",description:`<strong>labels</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Labels for computing the token classification loss. Indices should be in <code>[0, ..., config.num_labels - 1]</code>.`,name:"labels"},{anchor:"transformers.DebertaV2ForTokenClassification.forward.output_attentions",description:`<strong>output_attentions</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return the attentions tensors of all attention layers. See <code>attentions</code> under returned
tensors for more detail.`,name:"output_attentions"},{anchor:"transformers.DebertaV2ForTokenClassification.forward.output_hidden_states",description:`<strong>output_hidden_states</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return the hidden states of all layers. See <code>hidden_states</code> under returned tensors for
more detail.`,name:"output_hidden_states"},{anchor:"transformers.DebertaV2ForTokenClassification.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_48314/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/pr_48314/en/main_classes/output#transformers.modeling_outputs.TokenClassifierOutput"
>TokenClassifierOutput</a> or a tuple of
<code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various
elements depending on the configuration (<a
  href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2Config"
>DebertaV2Config</a>) and inputs.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/pr_48314/en/main_classes/output#transformers.modeling_outputs.TokenClassifierOutput"
>TokenClassifierOutput</a> or <code>tuple(torch.FloatTensor)</code></p>
`});var ie=e(re,4);u(ie,{children:(t,r)=>{var o=g();l(t,o)},$$slots:{default:!0}});var Ue=e(ie,4);m(Ue,{anchor:"transformers.DebertaV2ForTokenClassification.forward.example",children:(t,r)=>{var o=f(),i=e(p(o),2);c(i,{code:"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",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, DebertaV2ForTokenClassification
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> torch

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-v2-xlarge&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = DebertaV2ForTokenClassification.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-v2-xlarge&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = tokenizer(
<span class="hljs-meta">... </span>    <span class="hljs-string">&quot;HuggingFace is a company based in Paris and New York&quot;</span>, add_special_tokens=<span class="hljs-literal">False</span>, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>
<span class="hljs-meta">... </span>)

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">with</span> torch.no_grad():
<span class="hljs-meta">... </span>    logits = model(**inputs).logits

<span class="hljs-meta">&gt;&gt;&gt; </span>predicted_token_class_ids = logits.argmax(-<span class="hljs-number">1</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># Note that tokens are classified rather then input words which means that</span>
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># there might be more predicted token classes than words.</span>
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># Multiple token classes might account for the same word</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>predicted_tokens_classes = [model.config.id2label[t.item()] <span class="hljs-keyword">for</span> t <span class="hljs-keyword">in</span> predicted_token_class_ids[<span class="hljs-number">0</span>]]
<span class="hljs-meta">&gt;&gt;&gt; </span>predicted_tokens_classes
...

<span class="hljs-meta">&gt;&gt;&gt; </span>labels = predicted_token_class_ids
<span class="hljs-meta">&gt;&gt;&gt; </span>loss = model(**inputs, labels=labels).loss
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-built_in">round</span>(loss.item(), <span class="hljs-number">2</span>)
...`,lang:"python",wrap:!1}),l(t,o)},$$slots:{default:!0}}),a(ae),a(J);var le=e(J,2);d(le,{title:"DebertaV2ForQuestionAnswering",local:"transformers.DebertaV2ForQuestionAnswering",headingTag:"h2"});var j=e(le,2),de=s(j);n(de,{name:"class transformers.DebertaV2ForQuestionAnswering",anchor:"transformers.DebertaV2ForQuestionAnswering",source:"https://github.com/huggingface/transformers/blob/vr_48314/src/transformers/models/deberta_v2/modeling_deberta_v2.py#L1182",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.DebertaV2ForQuestionAnswering.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2ForQuestionAnswering">DebertaV2ForQuestionAnswering</a>) &#x2014;
Model configuration class with all the parameters of the model. Initializing with a config file does not
load the weights associated with the model, only the configuration. Check out the
<a href="/docs/transformers/pr_48314/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var ce=e(de,8),pe=s(ce);n(pe,{name:"forward",anchor:"transformers.DebertaV2ForQuestionAnswering.forward",source:"https://github.com/huggingface/transformers/blob/vr_48314/src/transformers/models/deberta_v2/modeling_deberta_v2.py#L1193",parameters:[{name:"input_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"position_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.Tensor] = None"},{name:"start_positions",val:": typing.Optional[torch.Tensor] = None"},{name:"end_positions",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": bool | None = None"},{name:"output_hidden_states",val:": bool | None = None"},{name:"return_dict",val:": bool | None = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.DebertaV2ForQuestionAnswering.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p>
<p>Indices can be obtained using <a href="/docs/transformers/pr_48314/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_48314/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/pr_48314/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p>
<p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.DebertaV2ForQuestionAnswering.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p>
<ul>
<li>1 for tokens that are <strong>not masked</strong>,</li>
<li>0 for tokens that are <strong>masked</strong>.</li>
</ul>
<p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.DebertaV2ForQuestionAnswering.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p>
<ul>
<li>0 corresponds to a <em>sentence A</em> token,</li>
<li>1 corresponds to a <em>sentence B</em> token.</li>
</ul>
<p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.DebertaV2ForQuestionAnswering.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Indices of positions of each input sequence tokens in the position embeddings. Selected in the range <code>[0, config.n_positions - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.DebertaV2ForQuestionAnswering.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) &#x2014;
Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This
is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.DebertaV2ForQuestionAnswering.forward.start_positions",description:`<strong>start_positions</strong> (<code>torch.Tensor</code> of shape <code>(batch_size,)</code>, <em>optional</em>) &#x2014;
Labels for position (index) of the start of the labelled span for computing the token classification loss.
Positions are clamped to the length of the sequence (<code>sequence_length</code>). Position outside of the sequence
are not taken into account for computing the loss.`,name:"start_positions"},{anchor:"transformers.DebertaV2ForQuestionAnswering.forward.end_positions",description:`<strong>end_positions</strong> (<code>torch.Tensor</code> of shape <code>(batch_size,)</code>, <em>optional</em>) &#x2014;
Labels for position (index) of the end of the labelled span for computing the token classification loss.
Positions are clamped to the length of the sequence (<code>sequence_length</code>). Position outside of the sequence
are not taken into account for computing the loss.`,name:"end_positions"},{anchor:"transformers.DebertaV2ForQuestionAnswering.forward.output_attentions",description:`<strong>output_attentions</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return the attentions tensors of all attention layers. See <code>attentions</code> under returned
tensors for more detail.`,name:"output_attentions"},{anchor:"transformers.DebertaV2ForQuestionAnswering.forward.output_hidden_states",description:`<strong>output_hidden_states</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return the hidden states of all layers. See <code>hidden_states</code> under returned tensors for
more detail.`,name:"output_hidden_states"},{anchor:"transformers.DebertaV2ForQuestionAnswering.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_48314/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/pr_48314/en/main_classes/output#transformers.modeling_outputs.QuestionAnsweringModelOutput"
>QuestionAnsweringModelOutput</a> or a tuple of
<code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various
elements depending on the configuration (<a
  href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2Config"
>DebertaV2Config</a>) and inputs.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/pr_48314/en/main_classes/output#transformers.modeling_outputs.QuestionAnsweringModelOutput"
>QuestionAnsweringModelOutput</a> or <code>tuple(torch.FloatTensor)</code></p>
`});var me=e(pe,4);u(me,{children:(t,r)=>{var o=g();l(t,o)},$$slots:{default:!0}});var ze=e(me,4);m(ze,{anchor:"transformers.DebertaV2ForQuestionAnswering.forward.example",children:(t,r)=>{var o=f(),i=e(p(o),2);c(i,{code:"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",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, DebertaV2ForQuestionAnswering
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> torch

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-v2-xlarge&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = DebertaV2ForQuestionAnswering.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-v2-xlarge&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>question, text = <span class="hljs-string">&quot;Who was Jim Henson?&quot;</span>, <span class="hljs-string">&quot;Jim Henson was a nice puppet&quot;</span>

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = tokenizer(question, text, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">with</span> torch.no_grad():
<span class="hljs-meta">... </span>    outputs = model(**inputs)

<span class="hljs-meta">&gt;&gt;&gt; </span>answer_start_index = outputs.start_logits.argmax()
<span class="hljs-meta">&gt;&gt;&gt; </span>answer_end_index = outputs.end_logits.argmax()

<span class="hljs-meta">&gt;&gt;&gt; </span>predict_answer_tokens = inputs.input_ids[<span class="hljs-number">0</span>, answer_start_index : answer_end_index + <span class="hljs-number">1</span>]
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer.decode(predict_answer_tokens, skip_special_tokens=<span class="hljs-literal">True</span>)
...

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># target is &quot;nice puppet&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>target_start_index = torch.tensor([<span class="hljs-number">14</span>])
<span class="hljs-meta">&gt;&gt;&gt; </span>target_end_index = torch.tensor([<span class="hljs-number">15</span>])

<span class="hljs-meta">&gt;&gt;&gt; </span>outputs = model(**inputs, start_positions=target_start_index, end_positions=target_end_index)
<span class="hljs-meta">&gt;&gt;&gt; </span>loss = outputs.loss
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-built_in">round</span>(loss.item(), <span class="hljs-number">2</span>)
...`,lang:"python",wrap:!1}),l(t,o)},$$slots:{default:!0}}),a(ce),a(j);var he=e(j,2);d(he,{title:"DebertaV2ForMultipleChoice",local:"transformers.DebertaV2ForMultipleChoice",headingTag:"h2"});var V=e(he,2),ue=s(V);n(ue,{name:"class transformers.DebertaV2ForMultipleChoice",anchor:"transformers.DebertaV2ForMultipleChoice",source:"https://github.com/huggingface/transformers/blob/vr_48314/src/transformers/models/deberta_v2/modeling_deberta_v2.py#L1260",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.DebertaV2ForMultipleChoice.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2ForMultipleChoice">DebertaV2ForMultipleChoice</a>) &#x2014;
Model configuration class with all the parameters of the model. Initializing with a config file does not
load the weights associated with the model, only the configuration. Check out the
<a href="/docs/transformers/pr_48314/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}]});var ge=e(ue,8),fe=s(ge);n(fe,{name:"forward",anchor:"transformers.DebertaV2ForMultipleChoice.forward",source:"https://github.com/huggingface/transformers/blob/vr_48314/src/transformers/models/deberta_v2/modeling_deberta_v2.py#L1284",parameters:[{name:"input_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"token_type_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"position_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"inputs_embeds",val:": typing.Optional[torch.Tensor] = None"},{name:"labels",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": bool | None = None"},{name:"output_hidden_states",val:": bool | None = None"},{name:"return_dict",val:": bool | None = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.DebertaV2ForMultipleChoice.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.</p>
<p>Indices can be obtained using <a href="/docs/transformers/pr_48314/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_48314/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/pr_48314/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details.</p>
<p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.DebertaV2ForMultipleChoice.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p>
<ul>
<li>1 for tokens that are <strong>not masked</strong>,</li>
<li>0 for tokens that are <strong>masked</strong>.</li>
</ul>
<p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.DebertaV2ForMultipleChoice.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Segment token indices to indicate first and second portions of the inputs. Indices are selected in <code>[0, 1]</code>:</p>
<ul>
<li>0 corresponds to a <em>sentence A</em> token,</li>
<li>1 corresponds to a <em>sentence B</em> token.</li>
</ul>
<p><a href="../glossary#token-type-ids">What are token type IDs?</a>`,name:"token_type_ids"},{anchor:"transformers.DebertaV2ForMultipleChoice.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Indices of positions of each input sequence tokens in the position embeddings. Selected in the range <code>[0, config.n_positions - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.DebertaV2ForMultipleChoice.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>) &#x2014;
Optionally, instead of passing <code>input_ids</code> you can choose to directly pass an embedded representation. This
is useful if you want more control over how to convert <code>input_ids</code> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.DebertaV2ForMultipleChoice.forward.labels",description:`<strong>labels</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size,)</code>, <em>optional</em>) &#x2014;
Labels for computing the multiple choice classification loss. Indices should be in <code>[0, ..., num_choices-1]</code> where <code>num_choices</code> is the size of the second dimension of the input tensors. (See
<code>input_ids</code> above)`,name:"labels"},{anchor:"transformers.DebertaV2ForMultipleChoice.forward.output_attentions",description:`<strong>output_attentions</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return the attentions tensors of all attention layers. See <code>attentions</code> under returned
tensors for more detail.`,name:"output_attentions"},{anchor:"transformers.DebertaV2ForMultipleChoice.forward.output_hidden_states",description:`<strong>output_hidden_states</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return the hidden states of all layers. See <code>hidden_states</code> under returned tensors for
more detail.`,name:"output_hidden_states"},{anchor:"transformers.DebertaV2ForMultipleChoice.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_48314/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/pr_48314/en/main_classes/output#transformers.modeling_outputs.MultipleChoiceModelOutput"
>MultipleChoiceModelOutput</a> or a tuple of
<code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various
elements depending on the configuration (<a
  href="/docs/transformers/pr_48314/en/model_doc/deberta-v2#transformers.DebertaV2Config"
>DebertaV2Config</a>) and inputs.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/pr_48314/en/main_classes/output#transformers.modeling_outputs.MultipleChoiceModelOutput"
>MultipleChoiceModelOutput</a> or <code>tuple(torch.FloatTensor)</code></p>
`});var be=e(fe,4);u(be,{children:(t,r)=>{var o=g();l(t,o)},$$slots:{default:!0}});var Ze=e(be,4);m(Ze,{anchor:"transformers.DebertaV2ForMultipleChoice.forward.example",children:(t,r)=>{var o=f(),i=e(p(o),2);c(i,{code:"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",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, DebertaV2ForMultipleChoice
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> torch

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-v2-xlarge&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = DebertaV2ForMultipleChoice.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-v2-xlarge&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>prompt = <span class="hljs-string">&quot;In Italy, pizza served in formal settings, such as at a restaurant, is presented unsliced.&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>choice0 = <span class="hljs-string">&quot;It is eaten with a fork and a knife.&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>choice1 = <span class="hljs-string">&quot;It is eaten while held in the hand.&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>labels = torch.tensor(<span class="hljs-number">0</span>).unsqueeze(<span class="hljs-number">0</span>)  <span class="hljs-comment"># choice0 is correct (according to Wikipedia ;)), batch size 1</span>

<span class="hljs-meta">&gt;&gt;&gt; </span>encoding = tokenizer([prompt, prompt], [choice0, choice1], return_tensors=<span class="hljs-string">&quot;pt&quot;</span>, padding=<span class="hljs-literal">True</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>outputs = model(**{k: v.unsqueeze(<span class="hljs-number">0</span>) <span class="hljs-keyword">for</span> k, v <span class="hljs-keyword">in</span> encoding.items()}, labels=labels)  <span class="hljs-comment"># batch size is 1</span>

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># the linear classifier still needs to be trained</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>loss = outputs.loss
<span class="hljs-meta">&gt;&gt;&gt; </span>logits = outputs.logits`,lang:"python",wrap:!1}),l(t,o)},$$slots:{default:!0}}),a(ge),a(V);var Ce=e(V,2);De(Ce,{source:"https://github.com/huggingface/transformers/blob/main/docs/source/en/model_doc/deberta-v2.md"}),b(2),l(ye,z),Xe()}export{oo as component};