transformers/main/en/_app/immutable/nodes/112.deae7f69.js

import{s as Bn,o as Zn,n as q}from"../chunks/scheduler.25b97de1.js";import{S as Wn,i as En,g as p,s as r,r as T,A as Ln,h as m,f as a,c as i,j as E,u as y,x as g,k as L,y as l,a as d,v as w,d as k,t as M,w as $}from"../chunks/index.d9030fc9.js";import{T as mt}from"../chunks/Tip.baa67368.js";import{D as V}from"../chunks/Docstring.e257edda.js";import{C as Re}from"../chunks/CodeBlock.e6cd0d95.js";import{F as Gn,M as Rn}from"../chunks/Markdown.7217f838.js";import{E as qe}from"../chunks/ExampleCodeBlock.20db4b6e.js";import{P as Dn}from"../chunks/PipelineTag.5f100392.js";import{H as at,E as Vn}from"../chunks/EditOnGithub.91d95064.js";function Hn(x){let e,u="Example:",n,o,_;return o=new Re({props:{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> DebertaConfig, DebertaModel

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

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># Initializing a model (with random weights) from the microsoft/deberta-base style configuration</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>model = DebertaModel(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`,wrap:!1}}),{c(){e=p("p"),e.textContent=u,n=r(),T(o.$$.fragment)},l(t){e=m(t,"P",{"data-svelte-h":!0}),g(e)!=="svelte-11lpom8"&&(e.textContent=u),n=i(t),y(o.$$.fragment,t)},m(t,b){d(t,e,b),d(t,n,b),w(o,t,b),_=!0},p:q,i(t){_||(k(o.$$.fragment,t),_=!0)},o(t){M(o.$$.fragment,t),_=!1},d(t){t&&(a(e),a(n)),$(o,t)}}}function Nn(x){let e,u="be encoded differently whether it is at the beginning of the sentence (without space) or not:",n,o,_;return o=new Re({props:{code:"ZnJvbSUyMHRyYW5zZm9ybWVycyUyMGltcG9ydCUyMERlYmVydGFUb2tlbml6ZXIlMEElMEF0b2tlbml6ZXIlMjAlM0QlMjBEZWJlcnRhVG9rZW5pemVyLmZyb21fcHJldHJhaW5lZCglMjJtaWNyb3NvZnQlMkZkZWJlcnRhLWJhc2UlMjIpJTBBdG9rZW5pemVyKCUyMkhlbGxvJTIwd29ybGQlMjIpJTVCJTIyaW5wdXRfaWRzJTIyJTVEJTBBJTBBdG9rZW5pemVyKCUyMiUyMEhlbGxvJTIwd29ybGQlMjIpJTVCJTIyaW5wdXRfaWRzJTIyJTVE",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> DebertaTokenizer

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = DebertaTokenizer.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-base&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer(<span class="hljs-string">&quot;Hello world&quot;</span>)[<span class="hljs-string">&quot;input_ids&quot;</span>]
[<span class="hljs-number">1</span>, <span class="hljs-number">31414</span>, <span class="hljs-number">232</span>, <span class="hljs-number">2</span>]

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer(<span class="hljs-string">&quot; Hello world&quot;</span>)[<span class="hljs-string">&quot;input_ids&quot;</span>]
[<span class="hljs-number">1</span>, <span class="hljs-number">20920</span>, <span class="hljs-number">232</span>, <span class="hljs-number">2</span>]`,wrap:!1}}),{c(){e=p("p"),e.textContent=u,n=r(),T(o.$$.fragment)},l(t){e=m(t,"P",{"data-svelte-h":!0}),g(e)!=="svelte-12atnao"&&(e.textContent=u),n=i(t),y(o.$$.fragment,t)},m(t,b){d(t,e,b),d(t,n,b),w(o,t,b),_=!0},p:q,i(t){_||(k(o.$$.fragment,t),_=!0)},o(t){M(o.$$.fragment,t),_=!1},d(t){t&&(a(e),a(n)),$(o,t)}}}function Xn(x){let e,u="When used with <code>is_split_into_words=True</code>, this tokenizer will add a space before each word (even the first one).";return{c(){e=p("p"),e.innerHTML=u},l(n){e=m(n,"P",{"data-svelte-h":!0}),g(e)!=="svelte-jhmxzm"&&(e.innerHTML=u)},m(n,o){d(n,e,o)},p:q,d(n){n&&a(e)}}}function Sn(x){let e,u="sequence pair mask has the following format:",n,o,_;return o=new Re({props:{code:"MCUyMDAlMjAwJTIwMCUyMDAlMjAwJTIwMCUyMDAlMjAwJTIwMCUyMDAlMjAxJTIwMSUyMDElMjAxJTIwMSUyMDElMjAxJTIwMSUyMDElMEElN0MlMjBmaXJzdCUyMHNlcXVlbmNlJTIwJTIwJTIwJTIwJTdDJTIwc2Vjb25kJTIwc2VxdWVuY2UlMjAlN0M=",highlighted:`0<span class="hljs-number"> 0 </span>0<span class="hljs-number"> 0 </span>0<span class="hljs-number"> 0 </span>0<span class="hljs-number"> 0 </span>0<span class="hljs-number"> 0 </span>0<span class="hljs-number"> 1 </span>1<span class="hljs-number"> 1 </span>1<span class="hljs-number"> 1 </span>1<span class="hljs-number"> 1 </span>1 1
| first sequence    | second sequence |`,wrap:!1}}),{c(){e=p("p"),e.textContent=u,n=r(),T(o.$$.fragment)},l(t){e=m(t,"P",{"data-svelte-h":!0}),g(e)!=="svelte-16klr56"&&(e.textContent=u),n=i(t),y(o.$$.fragment,t)},m(t,b){d(t,e,b),d(t,n,b),w(o,t,b),_=!0},p:q,i(t){_||(k(o.$$.fragment,t),_=!0)},o(t){M(o.$$.fragment,t),_=!1},d(t){t&&(a(e),a(n)),$(o,t)}}}function Yn(x){let e,u="be encoded differently whether it is at the beginning of the sentence (without space) or not:",n,o,_;return o=new Re({props:{code:"ZnJvbSUyMHRyYW5zZm9ybWVycyUyMGltcG9ydCUyMERlYmVydGFUb2tlbml6ZXJGYXN0JTBBJTBBdG9rZW5pemVyJTIwJTNEJTIwRGViZXJ0YVRva2VuaXplckZhc3QuZnJvbV9wcmV0cmFpbmVkKCUyMm1pY3Jvc29mdCUyRmRlYmVydGEtYmFzZSUyMiklMEF0b2tlbml6ZXIoJTIySGVsbG8lMjB3b3JsZCUyMiklNUIlMjJpbnB1dF9pZHMlMjIlNUQlMEElMEF0b2tlbml6ZXIoJTIyJTIwSGVsbG8lMjB3b3JsZCUyMiklNUIlMjJpbnB1dF9pZHMlMjIlNUQ=",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> DebertaTokenizerFast

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = DebertaTokenizerFast.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-base&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer(<span class="hljs-string">&quot;Hello world&quot;</span>)[<span class="hljs-string">&quot;input_ids&quot;</span>]
[<span class="hljs-number">1</span>, <span class="hljs-number">31414</span>, <span class="hljs-number">232</span>, <span class="hljs-number">2</span>]

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer(<span class="hljs-string">&quot; Hello world&quot;</span>)[<span class="hljs-string">&quot;input_ids&quot;</span>]
[<span class="hljs-number">1</span>, <span class="hljs-number">20920</span>, <span class="hljs-number">232</span>, <span class="hljs-number">2</span>]`,wrap:!1}}),{c(){e=p("p"),e.textContent=u,n=r(),T(o.$$.fragment)},l(t){e=m(t,"P",{"data-svelte-h":!0}),g(e)!=="svelte-12atnao"&&(e.textContent=u),n=i(t),y(o.$$.fragment,t)},m(t,b){d(t,e,b),d(t,n,b),w(o,t,b),_=!0},p:q,i(t){_||(k(o.$$.fragment,t),_=!0)},o(t){M(o.$$.fragment,t),_=!1},d(t){t&&(a(e),a(n)),$(o,t)}}}function Qn(x){let e,u="When used with <code>is_split_into_words=True</code>, this tokenizer needs to be instantiated with <code>add_prefix_space=True</code>.";return{c(){e=p("p"),e.innerHTML=u},l(n){e=m(n,"P",{"data-svelte-h":!0}),g(e)!=="svelte-9gg91e"&&(e.innerHTML=u)},m(n,o){d(n,e,o)},p:q,d(n){n&&a(e)}}}function Pn(x){let e,u="sequence pair mask has the following format:",n,o,_;return o=new Re({props:{code:"MCUyMDAlMjAwJTIwMCUyMDAlMjAwJTIwMCUyMDAlMjAwJTIwMCUyMDAlMjAxJTIwMSUyMDElMjAxJTIwMSUyMDElMjAxJTIwMSUyMDElMEElN0MlMjBmaXJzdCUyMHNlcXVlbmNlJTIwJTIwJTIwJTIwJTdDJTIwc2Vjb25kJTIwc2VxdWVuY2UlMjAlN0M=",highlighted:`0<span class="hljs-number"> 0 </span>0<span class="hljs-number"> 0 </span>0<span class="hljs-number"> 0 </span>0<span class="hljs-number"> 0 </span>0<span class="hljs-number"> 0 </span>0<span class="hljs-number"> 1 </span>1<span class="hljs-number"> 1 </span>1<span class="hljs-number"> 1 </span>1<span class="hljs-number"> 1 </span>1 1
| first sequence    | second sequence |`,wrap:!1}}),{c(){e=p("p"),e.textContent=u,n=r(),T(o.$$.fragment)},l(t){e=m(t,"P",{"data-svelte-h":!0}),g(e)!=="svelte-16klr56"&&(e.textContent=u),n=i(t),y(o.$$.fragment,t)},m(t,b){d(t,e,b),d(t,n,b),w(o,t,b),_=!0},p:q,i(t){_||(k(o.$$.fragment,t),_=!0)},o(t){M(o.$$.fragment,t),_=!1},d(t){t&&(a(e),a(n)),$(o,t)}}}function An(x){let e,u=`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.`;return{c(){e=p("p"),e.innerHTML=u},l(n){e=m(n,"P",{"data-svelte-h":!0}),g(e)!=="svelte-fincs2"&&(e.innerHTML=u)},m(n,o){d(n,e,o)},p:q,d(n){n&&a(e)}}}function On(x){let e,u="Example:",n,o,_;return o=new Re({props:{code:"ZnJvbSUyMHRyYW5zZm9ybWVycyUyMGltcG9ydCUyMEF1dG9Ub2tlbml6ZXIlMkMlMjBEZWJlcnRhTW9kZWwlMEFpbXBvcnQlMjB0b3JjaCUwQSUwQXRva2VuaXplciUyMCUzRCUyMEF1dG9Ub2tlbml6ZXIuZnJvbV9wcmV0cmFpbmVkKCUyMm1pY3Jvc29mdCUyRmRlYmVydGEtYmFzZSUyMiklMEFtb2RlbCUyMCUzRCUyMERlYmVydGFNb2RlbC5mcm9tX3ByZXRyYWluZWQoJTIybWljcm9zb2Z0JTJGZGViZXJ0YS1iYXNlJTIyKSUwQSUwQWlucHV0cyUyMCUzRCUyMHRva2VuaXplciglMjJIZWxsbyUyQyUyMG15JTIwZG9nJTIwaXMlMjBjdXRlJTIyJTJDJTIwcmV0dXJuX3RlbnNvcnMlM0QlMjJwdCUyMiklMEFvdXRwdXRzJTIwJTNEJTIwbW9kZWwoKippbnB1dHMpJTBBJTBBbGFzdF9oaWRkZW5fc3RhdGVzJTIwJTNEJTIwb3V0cHV0cy5sYXN0X2hpZGRlbl9zdGF0ZQ==",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, DebertaModel
<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-base&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = DebertaModel.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-base&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>outputs = model(**inputs)

<span class="hljs-meta">&gt;&gt;&gt; </span>last_hidden_states = outputs.last_hidden_state`,wrap:!1}}),{c(){e=p("p"),e.textContent=u,n=r(),T(o.$$.fragment)},l(t){e=m(t,"P",{"data-svelte-h":!0}),g(e)!=="svelte-11lpom8"&&(e.textContent=u),n=i(t),y(o.$$.fragment,t)},m(t,b){d(t,e,b),d(t,n,b),w(o,t,b),_=!0},p:q,i(t){_||(k(o.$$.fragment,t),_=!0)},o(t){M(o.$$.fragment,t),_=!1},d(t){t&&(a(e),a(n)),$(o,t)}}}function Kn(x){let e,u=`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.`;return{c(){e=p("p"),e.innerHTML=u},l(n){e=m(n,"P",{"data-svelte-h":!0}),g(e)!=="svelte-fincs2"&&(e.innerHTML=u)},m(n,o){d(n,e,o)},p:q,d(n){n&&a(e)}}}function eo(x){let e,u="Example:",n,o,_;return o=new Re({props:{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, DebertaForMaskedLM
<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;lsanochkin/deberta-large-feedback&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = DebertaForMaskedLM.from_pretrained(<span class="hljs-string">&quot;lsanochkin/deberta-large-feedback&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = tokenizer(<span class="hljs-string">&quot;The capital of France is [MASK].&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 [MASK]</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-string">&#x27; Paris&#x27;</span>

<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-[MASK] 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>)
<span class="hljs-number">0.54</span>`,wrap:!1}}),{c(){e=p("p"),e.textContent=u,n=r(),T(o.$$.fragment)},l(t){e=m(t,"P",{"data-svelte-h":!0}),g(e)!=="svelte-11lpom8"&&(e.textContent=u),n=i(t),y(o.$$.fragment,t)},m(t,b){d(t,e,b),d(t,n,b),w(o,t,b),_=!0},p:q,i(t){_||(k(o.$$.fragment,t),_=!0)},o(t){M(o.$$.fragment,t),_=!1},d(t){t&&(a(e),a(n)),$(o,t)}}}function to(x){let e,u=`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.`;return{c(){e=p("p"),e.innerHTML=u},l(n){e=m(n,"P",{"data-svelte-h":!0}),g(e)!=="svelte-fincs2"&&(e.innerHTML=u)},m(n,o){d(n,e,o)},p:q,d(n){n&&a(e)}}}function no(x){let e,u="Example of single-label classification:",n,o,_;return o=new Re({props:{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, DebertaForSequenceClassification

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-base&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = DebertaForSequenceClassification.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-base&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><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 = DebertaForSequenceClassification.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-base&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`,wrap:!1}}),{c(){e=p("p"),e.textContent=u,n=r(),T(o.$$.fragment)},l(t){e=m(t,"P",{"data-svelte-h":!0}),g(e)!=="svelte-ykxpe4"&&(e.textContent=u),n=i(t),y(o.$$.fragment,t)},m(t,b){d(t,e,b),d(t,n,b),w(o,t,b),_=!0},p:q,i(t){_||(k(o.$$.fragment,t),_=!0)},o(t){M(o.$$.fragment,t),_=!1},d(t){t&&(a(e),a(n)),$(o,t)}}}function oo(x){let e,u="Example of multi-label classification:",n,o,_;return o=new Re({props:{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, DebertaForSequenceClassification

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-base&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = DebertaForSequenceClassification.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-base&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 = DebertaForSequenceClassification.from_pretrained(
<span class="hljs-meta">... </span>    <span class="hljs-string">&quot;microsoft/deberta-base&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`,wrap:!1}}),{c(){e=p("p"),e.textContent=u,n=r(),T(o.$$.fragment)},l(t){e=m(t,"P",{"data-svelte-h":!0}),g(e)!=="svelte-1l8e32d"&&(e.textContent=u),n=i(t),y(o.$$.fragment,t)},m(t,b){d(t,e,b),d(t,n,b),w(o,t,b),_=!0},p:q,i(t){_||(k(o.$$.fragment,t),_=!0)},o(t){M(o.$$.fragment,t),_=!1},d(t){t&&(a(e),a(n)),$(o,t)}}}function so(x){let e,u=`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.`;return{c(){e=p("p"),e.innerHTML=u},l(n){e=m(n,"P",{"data-svelte-h":!0}),g(e)!=="svelte-fincs2"&&(e.innerHTML=u)},m(n,o){d(n,e,o)},p:q,d(n){n&&a(e)}}}function ao(x){let e,u="Example:",n,o,_;return o=new Re({props:{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, DebertaForTokenClassification
<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-base&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = DebertaForTokenClassification.from_pretrained(<span class="hljs-string">&quot;microsoft/deberta-base&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>labels = predicted_token_class_ids
<span class="hljs-meta">&gt;&gt;&gt; </span>loss = model(**inputs, labels=labels).loss`,wrap:!1}}),{c(){e=p("p"),e.textContent=u,n=r(),T(o.$$.fragment)},l(t){e=m(t,"P",{"data-svelte-h":!0}),g(e)!=="svelte-11lpom8"&&(e.textContent=u),n=i(t),y(o.$$.fragment,t)},m(t,b){d(t,e,b),d(t,n,b),w(o,t,b),_=!0},p:q,i(t){_||(k(o.$$.fragment,t),_=!0)},o(t){M(o.$$.fragment,t),_=!1},d(t){t&&(a(e),a(n)),$(o,t)}}}function ro(x){let e,u=`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.`;return{c(){e=p("p"),e.innerHTML=u},l(n){e=m(n,"P",{"data-svelte-h":!0}),g(e)!=="svelte-fincs2"&&(e.innerHTML=u)},m(n,o){d(n,e,o)},p:q,d(n){n&&a(e)}}}function io(x){let e,u="Example:",n,o,_;return o=new Re({props:{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, DebertaForQuestionAnswering
<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;Palak/microsoft_deberta-large_squad&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = DebertaForQuestionAnswering.from_pretrained(<span class="hljs-string">&quot;Palak/microsoft_deberta-large_squad&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-string">&#x27; a nice puppet&#x27;</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">12</span>])
<span class="hljs-meta">&gt;&gt;&gt; </span>target_end_index = torch.tensor([<span class="hljs-number">14</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>)
<span class="hljs-number">0.14</span>`,wrap:!1}}),{c(){e=p("p"),e.textContent=u,n=r(),T(o.$$.fragment)},l(t){e=m(t,"P",{"data-svelte-h":!0}),g(e)!=="svelte-11lpom8"&&(e.textContent=u),n=i(t),y(o.$$.fragment,t)},m(t,b){d(t,e,b),d(t,n,b),w(o,t,b),_=!0},p:q,i(t){_||(k(o.$$.fragment,t),_=!0)},o(t){M(o.$$.fragment,t),_=!1},d(t){t&&(a(e),a(n)),$(o,t)}}}function lo(x){let e,u,n,o,_,t,b=`The bare DeBERTa Model transformer outputting raw hidden-states without any specific head on top.
The DeBERTa model was proposed in <a href="https://arxiv.org/abs/2006.03654" rel="nofollow">DeBERTa: Decoding-enhanced BERT with Disentangled
Attention</a> by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen. It’s build
on top of BERT/RoBERTa with two improvements, i.e. disentangled attention and enhanced mask decoder. With those two
improvements, it out perform BERT/RoBERTa on a majority of tasks with 80GB pretraining data.`,Y,U,z=`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.`,Q,C,D,R,h,F='The <a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaModel">DebertaModel</a> forward method, overrides the <code>__call__</code> special method.',yt,Be,gn,ge,ut,Ze,Ct,K,rt,pe,it,ln=`An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
models.`,We,wt,ft,X,je,kt,Xe,Nt=`DeBERTa Model with a <code>language modeling</code> head on top.
The DeBERTa model was proposed in <a href="https://arxiv.org/abs/2006.03654" rel="nofollow">DeBERTa: Decoding-enhanced BERT with Disentangled
Attention</a> by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen. It’s build
on top of BERT/RoBERTa with two improvements, i.e. disentangled attention and enhanced mask decoder. With those two
improvements, it out perform BERT/RoBERTa on a majority of tasks with 80GB pretraining data.`,bn,Se,Xt=`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.`,Ot,P,St,Yt,be,Kt='The <a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaForMaskedLM">DebertaForMaskedLM</a> forward method, overrides the <code>__call__</code> special method.',A,Ee,en,me,tn,ee,Le,_e,Ye,Mt,Dt,ie=`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.`,zt,Ut,Qe=`The DeBERTa model was proposed in <a href="https://arxiv.org/abs/2006.03654" rel="nofollow">DeBERTa: Decoding-enhanced BERT with Disentangled
Attention</a> by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen. It’s build
on top of BERT/RoBERTa with two improvements, i.e. disentangled attention and enhanced mask decoder. With those two
improvements, it out perform BERT/RoBERTa on a majority of tasks with 80GB pretraining data.`,dn,Pe,ht=`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.`,B,H,$t,Ge,It,qt='The <a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaForSequenceClassification">DebertaForSequenceClassification</a> forward method, overrides the <code>__call__</code> special method.',vt,S,xt,Me,Rt,Ae,cn,Oe,gt,ue,$e,kn,Bt,Te=`DeBERTa Model 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.`,Jt,te,lt=`The DeBERTa model was proposed in <a href="https://arxiv.org/abs/2006.03654" rel="nofollow">DeBERTa: Decoding-enhanced BERT with Disentangled
Attention</a> by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen. It’s build
on top of BERT/RoBERTa with two improvements, i.e. disentangled attention and enhanced mask decoder. With those two
improvements, it out perform BERT/RoBERTa on a majority of tasks with 80GB pretraining data.`,_n,dt,nn=`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.`,Tn,ne,Fe,Ce,Zt,Ke='The <a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaForTokenClassification">DebertaForTokenClassification</a> forward method, overrides the <code>__call__</code> special method.',pn,De,ve,W,bt,et,Wt,ye,ze,Et,_t,Mn=`DeBERTa Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear
layers on top of the hidden-states output to compute <code>span start logits</code> and <code>span end logits</code>).`,Lt,Tt,Gt=`The DeBERTa model was proposed in <a href="https://arxiv.org/abs/2006.03654" rel="nofollow">DeBERTa: Decoding-enhanced BERT with Disentangled
Attention</a> by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen. It’s build
on top of BERT/RoBERTa with two improvements, i.e. disentangled attention and enhanced mask decoder. With those two
improvements, it out perform BERT/RoBERTa on a majority of tasks with 80GB pretraining data.`,ct,tt,we=`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.`,Z,le,nt,Qt,Pt,jt='The <a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaForQuestionAnswering">DebertaForQuestionAnswering</a> forward method, overrides the <code>__call__</code> special method.',Vt,Ue,mn,O,At;return e=new at({props:{title:"DebertaModel",local:"transformers.DebertaModel",headingTag:"h2"}}),o=new V({props:{name:"class transformers.DebertaModel",anchor:"transformers.DebertaModel",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.DebertaModel.config",description:`<strong>config</strong> (<a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaConfig">DebertaConfig</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/main/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_deberta.py#L888"}}),D=new V({props:{name:"forward",anchor:"transformers.DebertaModel.forward",parameters:[{name:"input_ids",val:": Optional = None"},{name:"attention_mask",val:": Optional = None"},{name:"token_type_ids",val:": Optional = None"},{name:"position_ids",val:": Optional = None"},{name:"inputs_embeds",val:": Optional = None"},{name:"output_attentions",val:": Optional = None"},{name:"output_hidden_states",val:": Optional = None"},{name:"return_dict",val:": Optional = None"}],parametersDescription:[{anchor:"transformers.DebertaModel.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>) &#x2014;
Indices of input sequence tokens in the vocabulary.</p>
<p>Indices can be obtained using <a href="/docs/transformers/main/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/main/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/main/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.DebertaModel.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.FloatTensor</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.DebertaModel.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.LongTensor</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.DebertaModel.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.LongTensor</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.max_position_embeddings - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.DebertaModel.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.FloatTensor</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 <em>input_ids</em> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.DebertaModel.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.DebertaModel.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.DebertaModel.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/main/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_deberta.py#L916",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_outputs.BaseModelOutput"
>transformers.modeling_outputs.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/main/en/model_doc/deberta#transformers.DebertaConfig"
>DebertaConfig</a>) and inputs.</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>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_outputs.BaseModelOutput"
>transformers.modeling_outputs.BaseModelOutput</a> or <code>tuple(torch.FloatTensor)</code></p>
`}}),Be=new mt({props:{$$slots:{default:[An]},$$scope:{ctx:x}}}),ge=new qe({props:{anchor:"transformers.DebertaModel.forward.example",$$slots:{default:[On]},$$scope:{ctx:x}}}),Ze=new at({props:{title:"DebertaPreTrainedModel",local:"transformers.DebertaPreTrainedModel",headingTag:"h2"}}),rt=new V({props:{name:"class transformers.DebertaPreTrainedModel",anchor:"transformers.DebertaPreTrainedModel",parameters:[{name:"config",val:": PretrainedConfig"},{name:"*inputs",val:""},{name:"**kwargs",val:""}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_deberta.py#L802"}}),wt=new at({props:{title:"DebertaForMaskedLM",local:"transformers.DebertaForMaskedLM",headingTag:"h2"}}),je=new V({props:{name:"class transformers.DebertaForMaskedLM",anchor:"transformers.DebertaForMaskedLM",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.DebertaForMaskedLM.config",description:`<strong>config</strong> (<a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaConfig">DebertaConfig</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/main/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_deberta.py#L1003"}}),St=new V({props:{name:"forward",anchor:"transformers.DebertaForMaskedLM.forward",parameters:[{name:"input_ids",val:": Optional = None"},{name:"attention_mask",val:": Optional = None"},{name:"token_type_ids",val:": Optional = None"},{name:"position_ids",val:": Optional = None"},{name:"inputs_embeds",val:": Optional = None"},{name:"labels",val:": Optional = None"},{name:"output_attentions",val:": Optional = None"},{name:"output_hidden_states",val:": Optional = None"},{name:"return_dict",val:": Optional = None"}],parametersDescription:[{anchor:"transformers.DebertaForMaskedLM.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>) &#x2014;
Indices of input sequence tokens in the vocabulary.</p>
<p>Indices can be obtained using <a href="/docs/transformers/main/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/main/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/main/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.DebertaForMaskedLM.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.FloatTensor</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.DebertaForMaskedLM.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.LongTensor</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.DebertaForMaskedLM.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.LongTensor</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.max_position_embeddings - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.DebertaForMaskedLM.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.FloatTensor</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 <em>input_ids</em> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.DebertaForMaskedLM.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.DebertaForMaskedLM.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.DebertaForMaskedLM.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/main/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"},{anchor:"transformers.DebertaForMaskedLM.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"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_deberta.py#L1023",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_outputs.MaskedLMOutput"
>transformers.modeling_outputs.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/main/en/model_doc/deberta#transformers.DebertaConfig"
>DebertaConfig</a>) and inputs.</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>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_outputs.MaskedLMOutput"
>transformers.modeling_outputs.MaskedLMOutput</a> or <code>tuple(torch.FloatTensor)</code></p>
`}}),Ee=new mt({props:{$$slots:{default:[Kn]},$$scope:{ctx:x}}}),me=new qe({props:{anchor:"transformers.DebertaForMaskedLM.forward.example",$$slots:{default:[eo]},$$scope:{ctx:x}}}),ee=new at({props:{title:"DebertaForSequenceClassification",local:"transformers.DebertaForSequenceClassification",headingTag:"h2"}}),Ye=new V({props:{name:"class transformers.DebertaForSequenceClassification",anchor:"transformers.DebertaForSequenceClassification",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.DebertaForSequenceClassification.config",description:`<strong>config</strong> (<a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaConfig">DebertaConfig</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/main/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_deberta.py#L1138"}}),$t=new V({props:{name:"forward",anchor:"transformers.DebertaForSequenceClassification.forward",parameters:[{name:"input_ids",val:": Optional = None"},{name:"attention_mask",val:": Optional = None"},{name:"token_type_ids",val:": Optional = None"},{name:"position_ids",val:": Optional = None"},{name:"inputs_embeds",val:": Optional = None"},{name:"labels",val:": Optional = None"},{name:"output_attentions",val:": Optional = None"},{name:"output_hidden_states",val:": Optional = None"},{name:"return_dict",val:": Optional = None"}],parametersDescription:[{anchor:"transformers.DebertaForSequenceClassification.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>) &#x2014;
Indices of input sequence tokens in the vocabulary.</p>
<p>Indices can be obtained using <a href="/docs/transformers/main/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/main/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/main/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.DebertaForSequenceClassification.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.FloatTensor</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.DebertaForSequenceClassification.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.LongTensor</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.DebertaForSequenceClassification.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.LongTensor</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.max_position_embeddings - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.DebertaForSequenceClassification.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.FloatTensor</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 <em>input_ids</em> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.DebertaForSequenceClassification.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.DebertaForSequenceClassification.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.DebertaForSequenceClassification.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/main/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"},{anchor:"transformers.DebertaForSequenceClassification.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"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_deberta.py#L1170",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_outputs.SequenceClassifierOutput"
>transformers.modeling_outputs.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/main/en/model_doc/deberta#transformers.DebertaConfig"
>DebertaConfig</a>) and inputs.</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>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_outputs.SequenceClassifierOutput"
>transformers.modeling_outputs.SequenceClassifierOutput</a> or <code>tuple(torch.FloatTensor)</code></p>
`}}),S=new mt({props:{$$slots:{default:[to]},$$scope:{ctx:x}}}),Me=new qe({props:{anchor:"transformers.DebertaForSequenceClassification.forward.example",$$slots:{default:[no]},$$scope:{ctx:x}}}),Ae=new qe({props:{anchor:"transformers.DebertaForSequenceClassification.forward.example-2",$$slots:{default:[oo]},$$scope:{ctx:x}}}),Oe=new at({props:{title:"DebertaForTokenClassification",local:"transformers.DebertaForTokenClassification",headingTag:"h2"}}),$e=new V({props:{name:"class transformers.DebertaForTokenClassification",anchor:"transformers.DebertaForTokenClassification",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.DebertaForTokenClassification.config",description:`<strong>config</strong> (<a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaConfig">DebertaConfig</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/main/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_deberta.py#L1256"}}),Fe=new V({props:{name:"forward",anchor:"transformers.DebertaForTokenClassification.forward",parameters:[{name:"input_ids",val:": Optional = None"},{name:"attention_mask",val:": Optional = None"},{name:"token_type_ids",val:": Optional = None"},{name:"position_ids",val:": Optional = None"},{name:"inputs_embeds",val:": Optional = None"},{name:"labels",val:": Optional = None"},{name:"output_attentions",val:": Optional = None"},{name:"output_hidden_states",val:": Optional = None"},{name:"return_dict",val:": Optional = None"}],parametersDescription:[{anchor:"transformers.DebertaForTokenClassification.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>) &#x2014;
Indices of input sequence tokens in the vocabulary.</p>
<p>Indices can be obtained using <a href="/docs/transformers/main/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/main/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/main/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.DebertaForTokenClassification.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.FloatTensor</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.DebertaForTokenClassification.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.LongTensor</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.DebertaForTokenClassification.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.LongTensor</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.max_position_embeddings - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.DebertaForTokenClassification.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.FloatTensor</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 <em>input_ids</em> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.DebertaForTokenClassification.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.DebertaForTokenClassification.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.DebertaForTokenClassification.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/main/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"},{anchor:"transformers.DebertaForTokenClassification.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"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_deberta.py#L1275",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_outputs.TokenClassifierOutput"
>transformers.modeling_outputs.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/main/en/model_doc/deberta#transformers.DebertaConfig"
>DebertaConfig</a>) and inputs.</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>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_outputs.TokenClassifierOutput"
>transformers.modeling_outputs.TokenClassifierOutput</a> or <code>tuple(torch.FloatTensor)</code></p>
`}}),De=new mt({props:{$$slots:{default:[so]},$$scope:{ctx:x}}}),W=new qe({props:{anchor:"transformers.DebertaForTokenClassification.forward.example",$$slots:{default:[ao]},$$scope:{ctx:x}}}),et=new at({props:{title:"DebertaForQuestionAnswering",local:"transformers.DebertaForQuestionAnswering",headingTag:"h2"}}),ze=new V({props:{name:"class transformers.DebertaForQuestionAnswering",anchor:"transformers.DebertaForQuestionAnswering",parameters:[{name:"config",val:""}],parametersDescription:[{anchor:"transformers.DebertaForQuestionAnswering.config",description:`<strong>config</strong> (<a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaConfig">DebertaConfig</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/main/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_deberta.py#L1329"}}),nt=new V({props:{name:"forward",anchor:"transformers.DebertaForQuestionAnswering.forward",parameters:[{name:"input_ids",val:": Optional = None"},{name:"attention_mask",val:": Optional = None"},{name:"token_type_ids",val:": Optional = None"},{name:"position_ids",val:": Optional = None"},{name:"inputs_embeds",val:": Optional = None"},{name:"start_positions",val:": Optional = None"},{name:"end_positions",val:": Optional = None"},{name:"output_attentions",val:": Optional = None"},{name:"output_hidden_states",val:": Optional = None"},{name:"return_dict",val:": Optional = None"}],parametersDescription:[{anchor:"transformers.DebertaForQuestionAnswering.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>) &#x2014;
Indices of input sequence tokens in the vocabulary.</p>
<p>Indices can be obtained using <a href="/docs/transformers/main/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/main/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/main/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.DebertaForQuestionAnswering.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.FloatTensor</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.DebertaForQuestionAnswering.forward.token_type_ids",description:`<strong>token_type_ids</strong> (<code>torch.LongTensor</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.DebertaForQuestionAnswering.forward.position_ids",description:`<strong>position_ids</strong> (<code>torch.LongTensor</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.max_position_embeddings - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.DebertaForQuestionAnswering.forward.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>torch.FloatTensor</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 <em>input_ids</em> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.DebertaForQuestionAnswering.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.DebertaForQuestionAnswering.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.DebertaForQuestionAnswering.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/main/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"},{anchor:"transformers.DebertaForQuestionAnswering.forward.start_positions",description:`<strong>start_positions</strong> (<code>torch.LongTensor</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.DebertaForQuestionAnswering.forward.end_positions",description:`<strong>end_positions</strong> (<code>torch.LongTensor</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"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_deberta.py#L1347",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_outputs.QuestionAnsweringModelOutput"
>transformers.modeling_outputs.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/main/en/model_doc/deberta#transformers.DebertaConfig"
>DebertaConfig</a>) and inputs.</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>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_outputs.QuestionAnsweringModelOutput"
>transformers.modeling_outputs.QuestionAnsweringModelOutput</a> or <code>tuple(torch.FloatTensor)</code></p>
`}}),Ue=new mt({props:{$$slots:{default:[ro]},$$scope:{ctx:x}}}),O=new qe({props:{anchor:"transformers.DebertaForQuestionAnswering.forward.example",$$slots:{default:[io]},$$scope:{ctx:x}}}),{c(){T(e.$$.fragment),u=r(),n=p("div"),T(o.$$.fragment),_=r(),t=p("p"),t.innerHTML=b,Y=r(),U=p("p"),U.innerHTML=z,Q=r(),C=p("div"),T(D.$$.fragment),R=r(),h=p("p"),h.innerHTML=F,yt=r(),T(Be.$$.fragment),gn=r(),T(ge.$$.fragment),ut=r(),T(Ze.$$.fragment),Ct=r(),K=p("div"),T(rt.$$.fragment),pe=r(),it=p("p"),it.textContent=ln,We=r(),T(wt.$$.fragment),ft=r(),X=p("div"),T(je.$$.fragment),kt=r(),Xe=p("p"),Xe.innerHTML=Nt,bn=r(),Se=p("p"),Se.innerHTML=Xt,Ot=r(),P=p("div"),T(St.$$.fragment),Yt=r(),be=p("p"),be.innerHTML=Kt,A=r(),T(Ee.$$.fragment),en=r(),T(me.$$.fragment),tn=r(),T(ee.$$.fragment),Le=r(),_e=p("div"),T(Ye.$$.fragment),Mt=r(),Dt=p("p"),Dt.textContent=ie,zt=r(),Ut=p("p"),Ut.innerHTML=Qe,dn=r(),Pe=p("p"),Pe.innerHTML=ht,B=r(),H=p("div"),T($t.$$.fragment),Ge=r(),It=p("p"),It.innerHTML=qt,vt=r(),T(S.$$.fragment),xt=r(),T(Me.$$.fragment),Rt=r(),T(Ae.$$.fragment),cn=r(),T(Oe.$$.fragment),gt=r(),ue=p("div"),T($e.$$.fragment),kn=r(),Bt=p("p"),Bt.textContent=Te,Jt=r(),te=p("p"),te.innerHTML=lt,_n=r(),dt=p("p"),dt.innerHTML=nn,Tn=r(),ne=p("div"),T(Fe.$$.fragment),Ce=r(),Zt=p("p"),Zt.innerHTML=Ke,pn=r(),T(De.$$.fragment),ve=r(),T(W.$$.fragment),bt=r(),T(et.$$.fragment),Wt=r(),ye=p("div"),T(ze.$$.fragment),Et=r(),_t=p("p"),_t.innerHTML=Mn,Lt=r(),Tt=p("p"),Tt.innerHTML=Gt,ct=r(),tt=p("p"),tt.innerHTML=we,Z=r(),le=p("div"),T(nt.$$.fragment),Qt=r(),Pt=p("p"),Pt.innerHTML=jt,Vt=r(),T(Ue.$$.fragment),mn=r(),T(O.$$.fragment),this.h()},l(f){y(e.$$.fragment,f),u=i(f),n=m(f,"DIV",{class:!0});var j=E(n);y(o.$$.fragment,j),_=i(j),t=m(j,"P",{"data-svelte-h":!0}),g(t)!=="svelte-1bcviae"&&(t.innerHTML=b),Y=i(j),U=m(j,"P",{"data-svelte-h":!0}),g(U)!=="svelte-hswkmf"&&(U.innerHTML=z),Q=i(j),C=m(j,"DIV",{class:!0});var G=E(C);y(D.$$.fragment,G),R=i(G),h=m(G,"P",{"data-svelte-h":!0}),g(h)!=="svelte-bw2rgg"&&(h.innerHTML=F),yt=i(G),y(Be.$$.fragment,G),gn=i(G),y(ge.$$.fragment,G),G.forEach(a),j.forEach(a),ut=i(f),y(Ze.$$.fragment,f),Ct=i(f),K=m(f,"DIV",{class:!0});var de=E(K);y(rt.$$.fragment,de),pe=i(de),it=m(de,"P",{"data-svelte-h":!0}),g(it)!=="svelte-nj1b0k"&&(it.textContent=ln),de.forEach(a),We=i(f),y(wt.$$.fragment,f),ft=i(f),X=m(f,"DIV",{class:!0});var Ve=E(X);y(je.$$.fragment,Ve),kt=i(Ve),Xe=m(Ve,"P",{"data-svelte-h":!0}),g(Xe)!=="svelte-qd1vq5"&&(Xe.innerHTML=Nt),bn=i(Ve),Se=m(Ve,"P",{"data-svelte-h":!0}),g(Se)!=="svelte-hswkmf"&&(Se.innerHTML=Xt),Ot=i(Ve),P=m(Ve,"DIV",{class:!0});var oe=E(P);y(St.$$.fragment,oe),Yt=i(oe),be=m(oe,"P",{"data-svelte-h":!0}),g(be)!=="svelte-f90wc0"&&(be.innerHTML=Kt),A=i(oe),y(Ee.$$.fragment,oe),en=i(oe),y(me.$$.fragment,oe),oe.forEach(a),Ve.forEach(a),tn=i(f),y(ee.$$.fragment,f),Le=i(f),_e=m(f,"DIV",{class:!0});var He=E(_e);y(Ye.$$.fragment,He),Mt=i(He),Dt=m(He,"P",{"data-svelte-h":!0}),g(Dt)!=="svelte-1xhb56d"&&(Dt.textContent=ie),zt=i(He),Ut=m(He,"P",{"data-svelte-h":!0}),g(Ut)!=="svelte-1bqto0e"&&(Ut.innerHTML=Qe),dn=i(He),Pe=m(He,"P",{"data-svelte-h":!0}),g(Pe)!=="svelte-hswkmf"&&(Pe.innerHTML=ht),B=i(He),H=m(He,"DIV",{class:!0});var xe=E(H);y($t.$$.fragment,xe),Ge=i(xe),It=m(xe,"P",{"data-svelte-h":!0}),g(It)!=="svelte-1bmj70q"&&(It.innerHTML=qt),vt=i(xe),y(S.$$.fragment,xe),xt=i(xe),y(Me.$$.fragment,xe),Rt=i(xe),y(Ae.$$.fragment,xe),xe.forEach(a),He.forEach(a),cn=i(f),y(Oe.$$.fragment,f),gt=i(f),ue=m(f,"DIV",{class:!0});var N=E(ue);y($e.$$.fragment,N),kn=i(N),Bt=m(N,"P",{"data-svelte-h":!0}),g(Bt)!=="svelte-1t75rxs"&&(Bt.textContent=Te),Jt=i(N),te=m(N,"P",{"data-svelte-h":!0}),g(te)!=="svelte-1bqto0e"&&(te.innerHTML=lt),_n=i(N),dt=m(N,"P",{"data-svelte-h":!0}),g(dt)!=="svelte-hswkmf"&&(dt.innerHTML=nn),Tn=i(N),ne=m(N,"DIV",{class:!0});var ke=E(ne);y(Fe.$$.fragment,ke),Ce=i(ke),Zt=m(ke,"P",{"data-svelte-h":!0}),g(Zt)!=="svelte-dl0e3s"&&(Zt.innerHTML=Ke),pn=i(ke),y(De.$$.fragment,ke),ve=i(ke),y(W.$$.fragment,ke),ke.forEach(a),N.forEach(a),bt=i(f),y(et.$$.fragment,f),Wt=i(f),ye=m(f,"DIV",{class:!0});var Je=E(ye);y(ze.$$.fragment,Je),Et=i(Je),_t=m(Je,"P",{"data-svelte-h":!0}),g(_t)!=="svelte-1nbcy0m"&&(_t.innerHTML=Mn),Lt=i(Je),Tt=m(Je,"P",{"data-svelte-h":!0}),g(Tt)!=="svelte-1bqto0e"&&(Tt.innerHTML=Gt),ct=i(Je),tt=m(Je,"P",{"data-svelte-h":!0}),g(tt)!=="svelte-hswkmf"&&(tt.innerHTML=we),Z=i(Je),le=m(Je,"DIV",{class:!0});var se=E(le);y(nt.$$.fragment,se),Qt=i(se),Pt=m(se,"P",{"data-svelte-h":!0}),g(Pt)!=="svelte-1rk6xj4"&&(Pt.innerHTML=jt),Vt=i(se),y(Ue.$$.fragment,se),mn=i(se),y(O.$$.fragment,se),se.forEach(a),Je.forEach(a),this.h()},h(){L(C,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(n,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(K,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(P,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(X,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(H,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(_e,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(ne,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(ue,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(le,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(ye,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8")},m(f,j){w(e,f,j),d(f,u,j),d(f,n,j),w(o,n,null),l(n,_),l(n,t),l(n,Y),l(n,U),l(n,Q),l(n,C),w(D,C,null),l(C,R),l(C,h),l(C,yt),w(Be,C,null),l(C,gn),w(ge,C,null),d(f,ut,j),w(Ze,f,j),d(f,Ct,j),d(f,K,j),w(rt,K,null),l(K,pe),l(K,it),d(f,We,j),w(wt,f,j),d(f,ft,j),d(f,X,j),w(je,X,null),l(X,kt),l(X,Xe),l(X,bn),l(X,Se),l(X,Ot),l(X,P),w(St,P,null),l(P,Yt),l(P,be),l(P,A),w(Ee,P,null),l(P,en),w(me,P,null),d(f,tn,j),w(ee,f,j),d(f,Le,j),d(f,_e,j),w(Ye,_e,null),l(_e,Mt),l(_e,Dt),l(_e,zt),l(_e,Ut),l(_e,dn),l(_e,Pe),l(_e,B),l(_e,H),w($t,H,null),l(H,Ge),l(H,It),l(H,vt),w(S,H,null),l(H,xt),w(Me,H,null),l(H,Rt),w(Ae,H,null),d(f,cn,j),w(Oe,f,j),d(f,gt,j),d(f,ue,j),w($e,ue,null),l(ue,kn),l(ue,Bt),l(ue,Jt),l(ue,te),l(ue,_n),l(ue,dt),l(ue,Tn),l(ue,ne),w(Fe,ne,null),l(ne,Ce),l(ne,Zt),l(ne,pn),w(De,ne,null),l(ne,ve),w(W,ne,null),d(f,bt,j),w(et,f,j),d(f,Wt,j),d(f,ye,j),w(ze,ye,null),l(ye,Et),l(ye,_t),l(ye,Lt),l(ye,Tt),l(ye,ct),l(ye,tt),l(ye,Z),l(ye,le),w(nt,le,null),l(le,Qt),l(le,Pt),l(le,Vt),w(Ue,le,null),l(le,mn),w(O,le,null),At=!0},p(f,j){const G={};j&2&&(G.$$scope={dirty:j,ctx:f}),Be.$set(G);const de={};j&2&&(de.$$scope={dirty:j,ctx:f}),ge.$set(de);const Ve={};j&2&&(Ve.$$scope={dirty:j,ctx:f}),Ee.$set(Ve);const oe={};j&2&&(oe.$$scope={dirty:j,ctx:f}),me.$set(oe);const He={};j&2&&(He.$$scope={dirty:j,ctx:f}),S.$set(He);const xe={};j&2&&(xe.$$scope={dirty:j,ctx:f}),Me.$set(xe);const N={};j&2&&(N.$$scope={dirty:j,ctx:f}),Ae.$set(N);const ke={};j&2&&(ke.$$scope={dirty:j,ctx:f}),De.$set(ke);const Je={};j&2&&(Je.$$scope={dirty:j,ctx:f}),W.$set(Je);const se={};j&2&&(se.$$scope={dirty:j,ctx:f}),Ue.$set(se);const xn={};j&2&&(xn.$$scope={dirty:j,ctx:f}),O.$set(xn)},i(f){At||(k(e.$$.fragment,f),k(o.$$.fragment,f),k(D.$$.fragment,f),k(Be.$$.fragment,f),k(ge.$$.fragment,f),k(Ze.$$.fragment,f),k(rt.$$.fragment,f),k(wt.$$.fragment,f),k(je.$$.fragment,f),k(St.$$.fragment,f),k(Ee.$$.fragment,f),k(me.$$.fragment,f),k(ee.$$.fragment,f),k(Ye.$$.fragment,f),k($t.$$.fragment,f),k(S.$$.fragment,f),k(Me.$$.fragment,f),k(Ae.$$.fragment,f),k(Oe.$$.fragment,f),k($e.$$.fragment,f),k(Fe.$$.fragment,f),k(De.$$.fragment,f),k(W.$$.fragment,f),k(et.$$.fragment,f),k(ze.$$.fragment,f),k(nt.$$.fragment,f),k(Ue.$$.fragment,f),k(O.$$.fragment,f),At=!0)},o(f){M(e.$$.fragment,f),M(o.$$.fragment,f),M(D.$$.fragment,f),M(Be.$$.fragment,f),M(ge.$$.fragment,f),M(Ze.$$.fragment,f),M(rt.$$.fragment,f),M(wt.$$.fragment,f),M(je.$$.fragment,f),M(St.$$.fragment,f),M(Ee.$$.fragment,f),M(me.$$.fragment,f),M(ee.$$.fragment,f),M(Ye.$$.fragment,f),M($t.$$.fragment,f),M(S.$$.fragment,f),M(Me.$$.fragment,f),M(Ae.$$.fragment,f),M(Oe.$$.fragment,f),M($e.$$.fragment,f),M(Fe.$$.fragment,f),M(De.$$.fragment,f),M(W.$$.fragment,f),M(et.$$.fragment,f),M(ze.$$.fragment,f),M(nt.$$.fragment,f),M(Ue.$$.fragment,f),M(O.$$.fragment,f),At=!1},d(f){f&&(a(u),a(n),a(ut),a(Ct),a(K),a(We),a(ft),a(X),a(tn),a(Le),a(_e),a(cn),a(gt),a(ue),a(bt),a(Wt),a(ye)),$(e,f),$(o),$(D),$(Be),$(ge),$(Ze,f),$(rt),$(wt,f),$(je),$(St),$(Ee),$(me),$(ee,f),$(Ye),$($t),$(S),$(Me),$(Ae),$(Oe,f),$($e),$(Fe),$(De),$(W),$(et,f),$(ze),$(nt),$(Ue),$(O)}}}function co(x){let e,u;return e=new Rn({props:{$$slots:{default:[lo]},$$scope:{ctx:x}}}),{c(){T(e.$$.fragment)},l(n){y(e.$$.fragment,n)},m(n,o){w(e,n,o),u=!0},p(n,o){const _={};o&2&&(_.$$scope={dirty:o,ctx:n}),e.$set(_)},i(n){u||(k(e.$$.fragment,n),u=!0)},o(n){M(e.$$.fragment,n),u=!1},d(n){$(e,n)}}}function po(x){let e,u="TensorFlow models and layers in <code>transformers</code> accept two formats as input:",n,o,_="<li>having all inputs as keyword arguments (like PyTorch models), or</li> <li>having all inputs as a list, tuple or dict in the first positional argument.</li>",t,b,Y=`The reason the second format is supported is that Keras methods prefer this format when passing inputs to models
and layers. Because of this support, when using methods like <code>model.fit()</code> things should “just work” for you - just
pass your inputs and labels in any format that <code>model.fit()</code> supports! If, however, you want to use the second
format outside of Keras methods like <code>fit()</code> and <code>predict()</code>, such as when creating your own layers or models with
the Keras <code>Functional</code> API, there are three possibilities you can use to gather all the input Tensors in the first
positional argument:`,U,z,Q=`<li>a single Tensor with <code>input_ids</code> only and nothing else: <code>model(input_ids)</code></li> <li>a list of varying length with one or several input Tensors IN THE ORDER given in the docstring:
<code>model([input_ids, attention_mask])</code> or <code>model([input_ids, attention_mask, token_type_ids])</code></li> <li>a dictionary with one or several input Tensors associated to the input names given in the docstring:
<code>model({&quot;input_ids&quot;: input_ids, &quot;token_type_ids&quot;: token_type_ids})</code></li>`,C,D,R=`Note that when creating models and layers with
<a href="https://keras.io/guides/making_new_layers_and_models_via_subclassing/" rel="nofollow">subclassing</a> then you don’t need to worry
about any of this, as you can just pass inputs like you would to any other Python function!`;return{c(){e=p("p"),e.innerHTML=u,n=r(),o=p("ul"),o.innerHTML=_,t=r(),b=p("p"),b.innerHTML=Y,U=r(),z=p("ul"),z.innerHTML=Q,C=r(),D=p("p"),D.innerHTML=R},l(h){e=m(h,"P",{"data-svelte-h":!0}),g(e)!=="svelte-1ajbfxg"&&(e.innerHTML=u),n=i(h),o=m(h,"UL",{"data-svelte-h":!0}),g(o)!=="svelte-qm1t26"&&(o.innerHTML=_),t=i(h),b=m(h,"P",{"data-svelte-h":!0}),g(b)!=="svelte-1v9qsc5"&&(b.innerHTML=Y),U=i(h),z=m(h,"UL",{"data-svelte-h":!0}),g(z)!=="svelte-15scerc"&&(z.innerHTML=Q),C=i(h),D=m(h,"P",{"data-svelte-h":!0}),g(D)!=="svelte-1an3odd"&&(D.innerHTML=R)},m(h,F){d(h,e,F),d(h,n,F),d(h,o,F),d(h,t,F),d(h,b,F),d(h,U,F),d(h,z,F),d(h,C,F),d(h,D,F)},p:q,d(h){h&&(a(e),a(n),a(o),a(t),a(b),a(U),a(z),a(C),a(D))}}}function mo(x){let e,u=`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.`;return{c(){e=p("p"),e.innerHTML=u},l(n){e=m(n,"P",{"data-svelte-h":!0}),g(e)!=="svelte-fincs2"&&(e.innerHTML=u)},m(n,o){d(n,e,o)},p:q,d(n){n&&a(e)}}}function uo(x){let e,u="Example:",n,o,_;return o=new Re({props:{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, TFDebertaModel
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> tensorflow <span class="hljs-keyword">as</span> tf

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;kamalkraj/deberta-base&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = TFDebertaModel.from_pretrained(<span class="hljs-string">&quot;kamalkraj/deberta-base&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;tf&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>outputs = model(inputs)

<span class="hljs-meta">&gt;&gt;&gt; </span>last_hidden_states = outputs.last_hidden_state`,wrap:!1}}),{c(){e=p("p"),e.textContent=u,n=r(),T(o.$$.fragment)},l(t){e=m(t,"P",{"data-svelte-h":!0}),g(e)!=="svelte-11lpom8"&&(e.textContent=u),n=i(t),y(o.$$.fragment,t)},m(t,b){d(t,e,b),d(t,n,b),w(o,t,b),_=!0},p:q,i(t){_||(k(o.$$.fragment,t),_=!0)},o(t){M(o.$$.fragment,t),_=!1},d(t){t&&(a(e),a(n)),$(o,t)}}}function fo(x){let e,u="TensorFlow models and layers in <code>transformers</code> accept two formats as input:",n,o,_="<li>having all inputs as keyword arguments (like PyTorch models), or</li> <li>having all inputs as a list, tuple or dict in the first positional argument.</li>",t,b,Y=`The reason the second format is supported is that Keras methods prefer this format when passing inputs to models
and layers. Because of this support, when using methods like <code>model.fit()</code> things should “just work” for you - just
pass your inputs and labels in any format that <code>model.fit()</code> supports! If, however, you want to use the second
format outside of Keras methods like <code>fit()</code> and <code>predict()</code>, such as when creating your own layers or models with
the Keras <code>Functional</code> API, there are three possibilities you can use to gather all the input Tensors in the first
positional argument:`,U,z,Q=`<li>a single Tensor with <code>input_ids</code> only and nothing else: <code>model(input_ids)</code></li> <li>a list of varying length with one or several input Tensors IN THE ORDER given in the docstring:
<code>model([input_ids, attention_mask])</code> or <code>model([input_ids, attention_mask, token_type_ids])</code></li> <li>a dictionary with one or several input Tensors associated to the input names given in the docstring:
<code>model({&quot;input_ids&quot;: input_ids, &quot;token_type_ids&quot;: token_type_ids})</code></li>`,C,D,R=`Note that when creating models and layers with
<a href="https://keras.io/guides/making_new_layers_and_models_via_subclassing/" rel="nofollow">subclassing</a> then you don’t need to worry
about any of this, as you can just pass inputs like you would to any other Python function!`;return{c(){e=p("p"),e.innerHTML=u,n=r(),o=p("ul"),o.innerHTML=_,t=r(),b=p("p"),b.innerHTML=Y,U=r(),z=p("ul"),z.innerHTML=Q,C=r(),D=p("p"),D.innerHTML=R},l(h){e=m(h,"P",{"data-svelte-h":!0}),g(e)!=="svelte-1ajbfxg"&&(e.innerHTML=u),n=i(h),o=m(h,"UL",{"data-svelte-h":!0}),g(o)!=="svelte-qm1t26"&&(o.innerHTML=_),t=i(h),b=m(h,"P",{"data-svelte-h":!0}),g(b)!=="svelte-1v9qsc5"&&(b.innerHTML=Y),U=i(h),z=m(h,"UL",{"data-svelte-h":!0}),g(z)!=="svelte-15scerc"&&(z.innerHTML=Q),C=i(h),D=m(h,"P",{"data-svelte-h":!0}),g(D)!=="svelte-1an3odd"&&(D.innerHTML=R)},m(h,F){d(h,e,F),d(h,n,F),d(h,o,F),d(h,t,F),d(h,b,F),d(h,U,F),d(h,z,F),d(h,C,F),d(h,D,F)},p:q,d(h){h&&(a(e),a(n),a(o),a(t),a(b),a(U),a(z),a(C),a(D))}}}function ho(x){let e,u=`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.`;return{c(){e=p("p"),e.innerHTML=u},l(n){e=m(n,"P",{"data-svelte-h":!0}),g(e)!=="svelte-fincs2"&&(e.innerHTML=u)},m(n,o){d(n,e,o)},p:q,d(n){n&&a(e)}}}function go(x){let e,u="Example:",n,o,_;return o=new Re({props:{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, TFDebertaForMaskedLM
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> tensorflow <span class="hljs-keyword">as</span> tf

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;kamalkraj/deberta-base&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = TFDebertaForMaskedLM.from_pretrained(<span class="hljs-string">&quot;kamalkraj/deberta-base&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = tokenizer(<span class="hljs-string">&quot;The capital of France is [MASK].&quot;</span>, return_tensors=<span class="hljs-string">&quot;tf&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>logits = model(**inputs).logits

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># retrieve index of [MASK]</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>mask_token_index = tf.where((inputs.input_ids == tokenizer.mask_token_id)[<span class="hljs-number">0</span>])
<span class="hljs-meta">&gt;&gt;&gt; </span>selected_logits = tf.gather_nd(logits[<span class="hljs-number">0</span>], indices=mask_token_index)

<span class="hljs-meta">&gt;&gt;&gt; </span>predicted_token_id = tf.math.argmax(selected_logits, axis=-<span class="hljs-number">1</span>)`,wrap:!1}}),{c(){e=p("p"),e.textContent=u,n=r(),T(o.$$.fragment)},l(t){e=m(t,"P",{"data-svelte-h":!0}),g(e)!=="svelte-11lpom8"&&(e.textContent=u),n=i(t),y(o.$$.fragment,t)},m(t,b){d(t,e,b),d(t,n,b),w(o,t,b),_=!0},p:q,i(t){_||(k(o.$$.fragment,t),_=!0)},o(t){M(o.$$.fragment,t),_=!1},d(t){t&&(a(e),a(n)),$(o,t)}}}function bo(x){let e,u;return e=new Re({props:{code:"bGFiZWxzJTIwJTNEJTIwdG9rZW5pemVyKCUyMlRoZSUyMGNhcGl0YWwlMjBvZiUyMEZyYW5jZSUyMGlzJTIwUGFyaXMuJTIyJTJDJTIwcmV0dXJuX3RlbnNvcnMlM0QlMjJ0ZiUyMiklNUIlMjJpbnB1dF9pZHMlMjIlNUQlMEElMjMlMjBtYXNrJTIwbGFiZWxzJTIwb2YlMjBub24tJTVCTUFTSyU1RCUyMHRva2VucyUwQWxhYmVscyUyMCUzRCUyMHRmLndoZXJlKGlucHV0cy5pbnB1dF9pZHMlMjAlM0QlM0QlMjB0b2tlbml6ZXIubWFza190b2tlbl9pZCUyQyUyMGxhYmVscyUyQyUyMC0xMDApJTBBJTBBb3V0cHV0cyUyMCUzRCUyMG1vZGVsKCoqaW5wdXRzJTJDJTIwbGFiZWxzJTNEbGFiZWxzKQ==",highlighted:`<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;tf&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-[MASK] tokens</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>labels = tf.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)`,wrap:!1}}),{c(){T(e.$$.fragment)},l(n){y(e.$$.fragment,n)},m(n,o){w(e,n,o),u=!0},p:q,i(n){u||(k(e.$$.fragment,n),u=!0)},o(n){M(e.$$.fragment,n),u=!1},d(n){$(e,n)}}}function _o(x){let e,u="TensorFlow models and layers in <code>transformers</code> accept two formats as input:",n,o,_="<li>having all inputs as keyword arguments (like PyTorch models), or</li> <li>having all inputs as a list, tuple or dict in the first positional argument.</li>",t,b,Y=`The reason the second format is supported is that Keras methods prefer this format when passing inputs to models
and layers. Because of this support, when using methods like <code>model.fit()</code> things should “just work” for you - just
pass your inputs and labels in any format that <code>model.fit()</code> supports! If, however, you want to use the second
format outside of Keras methods like <code>fit()</code> and <code>predict()</code>, such as when creating your own layers or models with
the Keras <code>Functional</code> API, there are three possibilities you can use to gather all the input Tensors in the first
positional argument:`,U,z,Q=`<li>a single Tensor with <code>input_ids</code> only and nothing else: <code>model(input_ids)</code></li> <li>a list of varying length with one or several input Tensors IN THE ORDER given in the docstring:
<code>model([input_ids, attention_mask])</code> or <code>model([input_ids, attention_mask, token_type_ids])</code></li> <li>a dictionary with one or several input Tensors associated to the input names given in the docstring:
<code>model({&quot;input_ids&quot;: input_ids, &quot;token_type_ids&quot;: token_type_ids})</code></li>`,C,D,R=`Note that when creating models and layers with
<a href="https://keras.io/guides/making_new_layers_and_models_via_subclassing/" rel="nofollow">subclassing</a> then you don’t need to worry
about any of this, as you can just pass inputs like you would to any other Python function!`;return{c(){e=p("p"),e.innerHTML=u,n=r(),o=p("ul"),o.innerHTML=_,t=r(),b=p("p"),b.innerHTML=Y,U=r(),z=p("ul"),z.innerHTML=Q,C=r(),D=p("p"),D.innerHTML=R},l(h){e=m(h,"P",{"data-svelte-h":!0}),g(e)!=="svelte-1ajbfxg"&&(e.innerHTML=u),n=i(h),o=m(h,"UL",{"data-svelte-h":!0}),g(o)!=="svelte-qm1t26"&&(o.innerHTML=_),t=i(h),b=m(h,"P",{"data-svelte-h":!0}),g(b)!=="svelte-1v9qsc5"&&(b.innerHTML=Y),U=i(h),z=m(h,"UL",{"data-svelte-h":!0}),g(z)!=="svelte-15scerc"&&(z.innerHTML=Q),C=i(h),D=m(h,"P",{"data-svelte-h":!0}),g(D)!=="svelte-1an3odd"&&(D.innerHTML=R)},m(h,F){d(h,e,F),d(h,n,F),d(h,o,F),d(h,t,F),d(h,b,F),d(h,U,F),d(h,z,F),d(h,C,F),d(h,D,F)},p:q,d(h){h&&(a(e),a(n),a(o),a(t),a(b),a(U),a(z),a(C),a(D))}}}function To(x){let e,u=`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.`;return{c(){e=p("p"),e.innerHTML=u},l(n){e=m(n,"P",{"data-svelte-h":!0}),g(e)!=="svelte-fincs2"&&(e.innerHTML=u)},m(n,o){d(n,e,o)},p:q,d(n){n&&a(e)}}}function yo(x){let e,u="Example:",n,o,_;return o=new Re({props:{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, TFDebertaForSequenceClassification
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> tensorflow <span class="hljs-keyword">as</span> tf

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;kamalkraj/deberta-base&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = TFDebertaForSequenceClassification.from_pretrained(<span class="hljs-string">&quot;kamalkraj/deberta-base&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;tf&quot;</span>)

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

<span class="hljs-meta">&gt;&gt;&gt; </span>predicted_class_id = <span class="hljs-built_in">int</span>(tf.math.argmax(logits, axis=-<span class="hljs-number">1</span>)[<span class="hljs-number">0</span>])`,wrap:!1}}),{c(){e=p("p"),e.textContent=u,n=r(),T(o.$$.fragment)},l(t){e=m(t,"P",{"data-svelte-h":!0}),g(e)!=="svelte-11lpom8"&&(e.textContent=u),n=i(t),y(o.$$.fragment,t)},m(t,b){d(t,e,b),d(t,n,b),w(o,t,b),_=!0},p:q,i(t){_||(k(o.$$.fragment,t),_=!0)},o(t){M(o.$$.fragment,t),_=!1},d(t){t&&(a(e),a(n)),$(o,t)}}}function wo(x){let e,u;return e=new Re({props:{code:"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",highlighted:'<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>\n<span class="hljs-meta">&gt;&gt;&gt; </span>num_labels = <span class="hljs-built_in">len</span>(model.config.id2label)\n<span class="hljs-meta">&gt;&gt;&gt; </span>model = TFDebertaForSequenceClassification.from_pretrained(<span class="hljs-string">&quot;kamalkraj/deberta-base&quot;</span>, num_labels=num_labels)\n\n<span class="hljs-meta">&gt;&gt;&gt; </span>labels = tf.constant(<span class="hljs-number">1</span>)\n<span class="hljs-meta">&gt;&gt;&gt; </span>loss = model(**inputs, labels=labels).loss',wrap:!1}}),{c(){T(e.$$.fragment)},l(n){y(e.$$.fragment,n)},m(n,o){w(e,n,o),u=!0},p:q,i(n){u||(k(e.$$.fragment,n),u=!0)},o(n){M(e.$$.fragment,n),u=!1},d(n){$(e,n)}}}function ko(x){let e,u="TensorFlow models and layers in <code>transformers</code> accept two formats as input:",n,o,_="<li>having all inputs as keyword arguments (like PyTorch models), or</li> <li>having all inputs as a list, tuple or dict in the first positional argument.</li>",t,b,Y=`The reason the second format is supported is that Keras methods prefer this format when passing inputs to models
and layers. Because of this support, when using methods like <code>model.fit()</code> things should “just work” for you - just
pass your inputs and labels in any format that <code>model.fit()</code> supports! If, however, you want to use the second
format outside of Keras methods like <code>fit()</code> and <code>predict()</code>, such as when creating your own layers or models with
the Keras <code>Functional</code> API, there are three possibilities you can use to gather all the input Tensors in the first
positional argument:`,U,z,Q=`<li>a single Tensor with <code>input_ids</code> only and nothing else: <code>model(input_ids)</code></li> <li>a list of varying length with one or several input Tensors IN THE ORDER given in the docstring:
<code>model([input_ids, attention_mask])</code> or <code>model([input_ids, attention_mask, token_type_ids])</code></li> <li>a dictionary with one or several input Tensors associated to the input names given in the docstring:
<code>model({&quot;input_ids&quot;: input_ids, &quot;token_type_ids&quot;: token_type_ids})</code></li>`,C,D,R=`Note that when creating models and layers with
<a href="https://keras.io/guides/making_new_layers_and_models_via_subclassing/" rel="nofollow">subclassing</a> then you don’t need to worry
about any of this, as you can just pass inputs like you would to any other Python function!`;return{c(){e=p("p"),e.innerHTML=u,n=r(),o=p("ul"),o.innerHTML=_,t=r(),b=p("p"),b.innerHTML=Y,U=r(),z=p("ul"),z.innerHTML=Q,C=r(),D=p("p"),D.innerHTML=R},l(h){e=m(h,"P",{"data-svelte-h":!0}),g(e)!=="svelte-1ajbfxg"&&(e.innerHTML=u),n=i(h),o=m(h,"UL",{"data-svelte-h":!0}),g(o)!=="svelte-qm1t26"&&(o.innerHTML=_),t=i(h),b=m(h,"P",{"data-svelte-h":!0}),g(b)!=="svelte-1v9qsc5"&&(b.innerHTML=Y),U=i(h),z=m(h,"UL",{"data-svelte-h":!0}),g(z)!=="svelte-15scerc"&&(z.innerHTML=Q),C=i(h),D=m(h,"P",{"data-svelte-h":!0}),g(D)!=="svelte-1an3odd"&&(D.innerHTML=R)},m(h,F){d(h,e,F),d(h,n,F),d(h,o,F),d(h,t,F),d(h,b,F),d(h,U,F),d(h,z,F),d(h,C,F),d(h,D,F)},p:q,d(h){h&&(a(e),a(n),a(o),a(t),a(b),a(U),a(z),a(C),a(D))}}}function Mo(x){let e,u=`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.`;return{c(){e=p("p"),e.innerHTML=u},l(n){e=m(n,"P",{"data-svelte-h":!0}),g(e)!=="svelte-fincs2"&&(e.innerHTML=u)},m(n,o){d(n,e,o)},p:q,d(n){n&&a(e)}}}function $o(x){let e,u="Example:",n,o,_;return o=new Re({props:{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, TFDebertaForTokenClassification
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> tensorflow <span class="hljs-keyword">as</span> tf

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;kamalkraj/deberta-base&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = TFDebertaForTokenClassification.from_pretrained(<span class="hljs-string">&quot;kamalkraj/deberta-base&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;tf&quot;</span>
<span class="hljs-meta">... </span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>logits = model(**inputs).logits
<span class="hljs-meta">&gt;&gt;&gt; </span>predicted_token_class_ids = tf.math.argmax(logits, axis=-<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] <span class="hljs-keyword">for</span> t <span class="hljs-keyword">in</span> predicted_token_class_ids[<span class="hljs-number">0</span>].numpy().tolist()]`,wrap:!1}}),{c(){e=p("p"),e.textContent=u,n=r(),T(o.$$.fragment)},l(t){e=m(t,"P",{"data-svelte-h":!0}),g(e)!=="svelte-11lpom8"&&(e.textContent=u),n=i(t),y(o.$$.fragment,t)},m(t,b){d(t,e,b),d(t,n,b),w(o,t,b),_=!0},p:q,i(t){_||(k(o.$$.fragment,t),_=!0)},o(t){M(o.$$.fragment,t),_=!1},d(t){t&&(a(e),a(n)),$(o,t)}}}function vo(x){let e,u;return e=new Re({props:{code:"bGFiZWxzJTIwJTNEJTIwcHJlZGljdGVkX3Rva2VuX2NsYXNzX2lkcyUwQWxvc3MlMjAlM0QlMjB0Zi5tYXRoLnJlZHVjZV9tZWFuKG1vZGVsKCoqaW5wdXRzJTJDJTIwbGFiZWxzJTNEbGFiZWxzKS5sb3NzKQ==",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span>labels = predicted_token_class_ids
<span class="hljs-meta">&gt;&gt;&gt; </span>loss = tf.math.reduce_mean(model(**inputs, labels=labels).loss)`,wrap:!1}}),{c(){T(e.$$.fragment)},l(n){y(e.$$.fragment,n)},m(n,o){w(e,n,o),u=!0},p:q,i(n){u||(k(e.$$.fragment,n),u=!0)},o(n){M(e.$$.fragment,n),u=!1},d(n){$(e,n)}}}function xo(x){let e,u="TensorFlow models and layers in <code>transformers</code> accept two formats as input:",n,o,_="<li>having all inputs as keyword arguments (like PyTorch models), or</li> <li>having all inputs as a list, tuple or dict in the first positional argument.</li>",t,b,Y=`The reason the second format is supported is that Keras methods prefer this format when passing inputs to models
and layers. Because of this support, when using methods like <code>model.fit()</code> things should “just work” for you - just
pass your inputs and labels in any format that <code>model.fit()</code> supports! If, however, you want to use the second
format outside of Keras methods like <code>fit()</code> and <code>predict()</code>, such as when creating your own layers or models with
the Keras <code>Functional</code> API, there are three possibilities you can use to gather all the input Tensors in the first
positional argument:`,U,z,Q=`<li>a single Tensor with <code>input_ids</code> only and nothing else: <code>model(input_ids)</code></li> <li>a list of varying length with one or several input Tensors IN THE ORDER given in the docstring:
<code>model([input_ids, attention_mask])</code> or <code>model([input_ids, attention_mask, token_type_ids])</code></li> <li>a dictionary with one or several input Tensors associated to the input names given in the docstring:
<code>model({&quot;input_ids&quot;: input_ids, &quot;token_type_ids&quot;: token_type_ids})</code></li>`,C,D,R=`Note that when creating models and layers with
<a href="https://keras.io/guides/making_new_layers_and_models_via_subclassing/" rel="nofollow">subclassing</a> then you don’t need to worry
about any of this, as you can just pass inputs like you would to any other Python function!`;return{c(){e=p("p"),e.innerHTML=u,n=r(),o=p("ul"),o.innerHTML=_,t=r(),b=p("p"),b.innerHTML=Y,U=r(),z=p("ul"),z.innerHTML=Q,C=r(),D=p("p"),D.innerHTML=R},l(h){e=m(h,"P",{"data-svelte-h":!0}),g(e)!=="svelte-1ajbfxg"&&(e.innerHTML=u),n=i(h),o=m(h,"UL",{"data-svelte-h":!0}),g(o)!=="svelte-qm1t26"&&(o.innerHTML=_),t=i(h),b=m(h,"P",{"data-svelte-h":!0}),g(b)!=="svelte-1v9qsc5"&&(b.innerHTML=Y),U=i(h),z=m(h,"UL",{"data-svelte-h":!0}),g(z)!=="svelte-15scerc"&&(z.innerHTML=Q),C=i(h),D=m(h,"P",{"data-svelte-h":!0}),g(D)!=="svelte-1an3odd"&&(D.innerHTML=R)},m(h,F){d(h,e,F),d(h,n,F),d(h,o,F),d(h,t,F),d(h,b,F),d(h,U,F),d(h,z,F),d(h,C,F),d(h,D,F)},p:q,d(h){h&&(a(e),a(n),a(o),a(t),a(b),a(U),a(z),a(C),a(D))}}}function Jo(x){let e,u=`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.`;return{c(){e=p("p"),e.innerHTML=u},l(n){e=m(n,"P",{"data-svelte-h":!0}),g(e)!=="svelte-fincs2"&&(e.innerHTML=u)},m(n,o){d(n,e,o)},p:q,d(n){n&&a(e)}}}function jo(x){let e,u="Example:",n,o,_;return o=new Re({props:{code:"ZnJvbSUyMHRyYW5zZm9ybWVycyUyMGltcG9ydCUyMEF1dG9Ub2tlbml6ZXIlMkMlMjBURkRlYmVydGFGb3JRdWVzdGlvbkFuc3dlcmluZyUwQWltcG9ydCUyMHRlbnNvcmZsb3clMjBhcyUyMHRmJTBBJTBBdG9rZW5pemVyJTIwJTNEJTIwQXV0b1Rva2VuaXplci5mcm9tX3ByZXRyYWluZWQoJTIya2FtYWxrcmFqJTJGZGViZXJ0YS1iYXNlJTIyKSUwQW1vZGVsJTIwJTNEJTIwVEZEZWJlcnRhRm9yUXVlc3Rpb25BbnN3ZXJpbmcuZnJvbV9wcmV0cmFpbmVkKCUyMmthbWFsa3JhaiUyRmRlYmVydGEtYmFzZSUyMiklMEElMEFxdWVzdGlvbiUyQyUyMHRleHQlMjAlM0QlMjAlMjJXaG8lMjB3YXMlMjBKaW0lMjBIZW5zb24lM0YlMjIlMkMlMjAlMjJKaW0lMjBIZW5zb24lMjB3YXMlMjBhJTIwbmljZSUyMHB1cHBldCUyMiUwQSUwQWlucHV0cyUyMCUzRCUyMHRva2VuaXplcihxdWVzdGlvbiUyQyUyMHRleHQlMkMlMjByZXR1cm5fdGVuc29ycyUzRCUyMnRmJTIyKSUwQW91dHB1dHMlMjAlM0QlMjBtb2RlbCgqKmlucHV0cyklMEElMEFhbnN3ZXJfc3RhcnRfaW5kZXglMjAlM0QlMjBpbnQodGYubWF0aC5hcmdtYXgob3V0cHV0cy5zdGFydF9sb2dpdHMlMkMlMjBheGlzJTNELTEpJTVCMCU1RCklMEFhbnN3ZXJfZW5kX2luZGV4JTIwJTNEJTIwaW50KHRmLm1hdGguYXJnbWF4KG91dHB1dHMuZW5kX2xvZ2l0cyUyQyUyMGF4aXMlM0QtMSklNUIwJTVEKSUwQSUwQXByZWRpY3RfYW5zd2VyX3Rva2VucyUyMCUzRCUyMGlucHV0cy5pbnB1dF9pZHMlNUIwJTJDJTIwYW5zd2VyX3N0YXJ0X2luZGV4JTIwJTNBJTIwYW5zd2VyX2VuZF9pbmRleCUyMCUyQiUyMDElNUQ=",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, TFDebertaForQuestionAnswering
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> tensorflow <span class="hljs-keyword">as</span> tf

<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;kamalkraj/deberta-base&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = TFDebertaForQuestionAnswering.from_pretrained(<span class="hljs-string">&quot;kamalkraj/deberta-base&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;tf&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>outputs = model(**inputs)

<span class="hljs-meta">&gt;&gt;&gt; </span>answer_start_index = <span class="hljs-built_in">int</span>(tf.math.argmax(outputs.start_logits, axis=-<span class="hljs-number">1</span>)[<span class="hljs-number">0</span>])
<span class="hljs-meta">&gt;&gt;&gt; </span>answer_end_index = <span class="hljs-built_in">int</span>(tf.math.argmax(outputs.end_logits, axis=-<span class="hljs-number">1</span>)[<span class="hljs-number">0</span>])

<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>]`,wrap:!1}}),{c(){e=p("p"),e.textContent=u,n=r(),T(o.$$.fragment)},l(t){e=m(t,"P",{"data-svelte-h":!0}),g(e)!=="svelte-11lpom8"&&(e.textContent=u),n=i(t),y(o.$$.fragment,t)},m(t,b){d(t,e,b),d(t,n,b),w(o,t,b),_=!0},p:q,i(t){_||(k(o.$$.fragment,t),_=!0)},o(t){M(o.$$.fragment,t),_=!1},d(t){t&&(a(e),a(n)),$(o,t)}}}function Fo(x){let e,u;return e=new Re({props:{code:"JTIzJTIwdGFyZ2V0JTIwaXMlMjAlMjJuaWNlJTIwcHVwcGV0JTIyJTBBdGFyZ2V0X3N0YXJ0X2luZGV4JTIwJTNEJTIwdGYuY29uc3RhbnQoJTVCMTQlNUQpJTBBdGFyZ2V0X2VuZF9pbmRleCUyMCUzRCUyMHRmLmNvbnN0YW50KCU1QjE1JTVEKSUwQSUwQW91dHB1dHMlMjAlM0QlMjBtb2RlbCgqKmlucHV0cyUyQyUyMHN0YXJ0X3Bvc2l0aW9ucyUzRHRhcmdldF9zdGFydF9pbmRleCUyQyUyMGVuZF9wb3NpdGlvbnMlM0R0YXJnZXRfZW5kX2luZGV4KSUwQWxvc3MlMjAlM0QlMjB0Zi5tYXRoLnJlZHVjZV9tZWFuKG91dHB1dHMubG9zcyk=",highlighted:`<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 = tf.constant([<span class="hljs-number">14</span>])
<span class="hljs-meta">&gt;&gt;&gt; </span>target_end_index = tf.constant([<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 = tf.math.reduce_mean(outputs.loss)`,wrap:!1}}),{c(){T(e.$$.fragment)},l(n){y(e.$$.fragment,n)},m(n,o){w(e,n,o),u=!0},p:q,i(n){u||(k(e.$$.fragment,n),u=!0)},o(n){M(e.$$.fragment,n),u=!1},d(n){$(e,n)}}}function Co(x){let e,u,n,o,_,t,b=`The bare DeBERTa Model transformer outputting raw hidden-states without any specific head on top.
The DeBERTa model was proposed in <a href="https://arxiv.org/abs/2006.03654" rel="nofollow">DeBERTa: Decoding-enhanced BERT with Disentangled
Attention</a> by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen. It’s build
on top of BERT/RoBERTa with two improvements, i.e. disentangled attention and enhanced mask decoder. With those two
improvements, it out perform BERT/RoBERTa on a majority of tasks with 80GB pretraining data.`,Y,U,z=`This model is also a <a href="https://www.tensorflow.org/api_docs/python/tf/keras/Model" rel="nofollow">keras.Model</a> subclass. Use it
as a regular TF 2.0 Keras Model and refer to the TF 2.0 documentation for all matter related to general usage and
behavior.`,Q,C,D,R,h,F,yt,Be='The <a href="/docs/transformers/main/en/model_doc/deberta#transformers.TFDebertaModel">TFDebertaModel</a> forward method, overrides the <code>__call__</code> special method.',gn,ge,ut,Ze,Ct,K,rt,pe,it,ln,We,wt=`An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
models.`,ft,X,je,kt,Xe,Nt="Calls the model on new inputs and returns the outputs as tensors.",bn,Se,Xt=`In this case <code>call()</code> just reapplies
all ops in the graph to the new inputs
(e.g. build a new computational graph from the provided inputs).`,Ot,P,St=`Note: This method should not be called directly. It is only meant to be
overridden when subclassing <code>tf.keras.Model</code>.
To call a model on an input, always use the <code>__call__()</code> method,
i.e. <code>model(inputs)</code>, which relies on the underlying <code>call()</code> method.`,Yt,be,Kt,A,Ee,en,me,tn=`DeBERTa Model with a <code>language modeling</code> head on top.
The DeBERTa model was proposed in <a href="https://arxiv.org/abs/2006.03654" rel="nofollow">DeBERTa: Decoding-enhanced BERT with Disentangled
Attention</a> by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen. It’s build
on top of BERT/RoBERTa with two improvements, i.e. disentangled attention and enhanced mask decoder. With those two
improvements, it out perform BERT/RoBERTa on a majority of tasks with 80GB pretraining data.`,ee,Le,_e=`This model is also a <a href="https://www.tensorflow.org/api_docs/python/tf/keras/Model" rel="nofollow">keras.Model</a> subclass. Use it
as a regular TF 2.0 Keras Model and refer to the TF 2.0 documentation for all matter related to general usage and
behavior.`,Ye,Mt,Dt,ie,zt,Ut,Qe,dn='The <a href="/docs/transformers/main/en/model_doc/deberta#transformers.TFDebertaForMaskedLM">TFDebertaForMaskedLM</a> forward method, overrides the <code>__call__</code> special method.',Pe,ht,B,H,$t,Ge,It,qt,vt,S,xt,Me,Rt,Ae=`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.`,cn,Oe,gt=`The DeBERTa model was proposed in <a href="https://arxiv.org/abs/2006.03654" rel="nofollow">DeBERTa: Decoding-enhanced BERT with Disentangled
Attention</a> by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen. It’s build
on top of BERT/RoBERTa with two improvements, i.e. disentangled attention and enhanced mask decoder. With those two
improvements, it out perform BERT/RoBERTa on a majority of tasks with 80GB pretraining data.`,ue,$e,kn=`This model is also a <a href="https://www.tensorflow.org/api_docs/python/tf/keras/Model" rel="nofollow">keras.Model</a> subclass. Use it
as a regular TF 2.0 Keras Model and refer to the TF 2.0 documentation for all matter related to general usage and
behavior.`,Bt,Te,Jt,te,lt,_n,dt,nn='The <a href="/docs/transformers/main/en/model_doc/deberta#transformers.TFDebertaForSequenceClassification">TFDebertaForSequenceClassification</a> forward method, overrides the <code>__call__</code> special method.',Tn,ne,Fe,Ce,Zt,Ke,pn,De,ve,W,bt,et,Wt,ye=`DeBERTa Model 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.`,ze,Et,_t=`The DeBERTa model was proposed in <a href="https://arxiv.org/abs/2006.03654" rel="nofollow">DeBERTa: Decoding-enhanced BERT with Disentangled
Attention</a> by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen. It’s build
on top of BERT/RoBERTa with two improvements, i.e. disentangled attention and enhanced mask decoder. With those two
improvements, it out perform BERT/RoBERTa on a majority of tasks with 80GB pretraining data.`,Mn,Lt,Tt=`This model is also a <a href="https://www.tensorflow.org/api_docs/python/tf/keras/Model" rel="nofollow">keras.Model</a> subclass. Use it
as a regular TF 2.0 Keras Model and refer to the TF 2.0 documentation for all matter related to general usage and
behavior.`,Gt,ct,tt,we,Z,le,nt,Qt='The <a href="/docs/transformers/main/en/model_doc/deberta#transformers.TFDebertaForTokenClassification">TFDebertaForTokenClassification</a> forward method, overrides the <code>__call__</code> special method.',Pt,jt,Vt,Ue,mn,O,At,f,j,G,de,Ve,oe,He=`DeBERTa Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear
layers on top of the hidden-states output to compute <code>span start logits</code> and <code>span end logits</code>).`,xe,N,ke=`The DeBERTa model was proposed in <a href="https://arxiv.org/abs/2006.03654" rel="nofollow">DeBERTa: Decoding-enhanced BERT with Disentangled
Attention</a> by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen. It’s build
on top of BERT/RoBERTa with two improvements, i.e. disentangled attention and enhanced mask decoder. With those two
improvements, it out perform BERT/RoBERTa on a majority of tasks with 80GB pretraining data.`,Je,se,xn=`This model is also a <a href="https://www.tensorflow.org/api_docs/python/tf/keras/Model" rel="nofollow">keras.Model</a> subclass. Use it
as a regular TF 2.0 Keras Model and refer to the TF 2.0 documentation for all matter related to general usage and
behavior.`,jn,Ft,Fn,ot,Ie,un,yn,$n='The <a href="/docs/transformers/main/en/model_doc/deberta#transformers.TFDebertaForQuestionAnswering">TFDebertaForQuestionAnswering</a> forward method, overrides the <code>__call__</code> special method.',Cn,on,sn,an,wn,fn,vn;return e=new at({props:{title:"TFDebertaModel",local:"transformers.TFDebertaModel",headingTag:"h2"}}),o=new V({props:{name:"class transformers.TFDebertaModel",anchor:"transformers.TFDebertaModel",parameters:[{name:"config",val:": DebertaConfig"},{name:"*inputs",val:""},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.TFDebertaModel.config",description:`<strong>config</strong> (<a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaConfig">DebertaConfig</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/main/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_tf_deberta.py#L1233"}}),C=new mt({props:{$$slots:{default:[po]},$$scope:{ctx:x}}}),h=new V({props:{name:"call",anchor:"transformers.TFDebertaModel.call",parameters:[{name:"input_ids",val:": TFModelInputType | None = None"},{name:"attention_mask",val:": np.ndarray | tf.Tensor | None = None"},{name:"token_type_ids",val:": np.ndarray | tf.Tensor | None = None"},{name:"position_ids",val:": np.ndarray | tf.Tensor | None = None"},{name:"inputs_embeds",val:": np.ndarray | tf.Tensor | None = None"},{name:"output_attentions",val:": Optional[bool] = None"},{name:"output_hidden_states",val:": Optional[bool] = None"},{name:"return_dict",val:": Optional[bool] = None"},{name:"training",val:": Optional[bool] = False"}],parametersDescription:[{anchor:"transformers.TFDebertaModel.call.input_ids",description:`<strong>input_ids</strong> (<code>np.ndarray</code>, <code>tf.Tensor</code>, <code>List[tf.Tensor]</code> \`<code>Dict[str, tf.Tensor]</code> or <code>Dict[str, np.ndarray]</code> and each example must have the shape <code>(batch_size, sequence_length)</code>) &#x2014;
Indices of input sequence tokens in the vocabulary.</p>
<p>Indices can be obtained using <a href="/docs/transformers/main/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/main/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/main/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.TFDebertaModel.call.attention_mask",description:`<strong>attention_mask</strong> (<code>np.ndarray</code> or <code>tf.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.TFDebertaModel.call.token_type_ids",description:`<strong>token_type_ids</strong> (<code>np.ndarray</code> or <code>tf.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.TFDebertaModel.call.position_ids",description:`<strong>position_ids</strong> (<code>np.ndarray</code> or <code>tf.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.max_position_embeddings - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.TFDebertaModel.call.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>np.ndarray</code> or <code>tf.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 <em>input_ids</em> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.TFDebertaModel.call.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.TFDebertaModel.call.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.TFDebertaModel.call.return_dict",description:"<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;\nWhether or not to return a [`~utils.ModelOutput&#x201C;] instead of a plain tuple.",name:"return_dict"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_tf_deberta.py#L1243",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_tf_outputs.TFBaseModelOutput"
>transformers.modeling_tf_outputs.TFBaseModelOutput</a> or a tuple of <code>tf.Tensor</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/main/en/model_doc/deberta#transformers.DebertaConfig"
>DebertaConfig</a>) and inputs.</p>
<ul>
<li>
<p><strong>last_hidden_state</strong> (<code>tf.Tensor</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(tf.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>tf.Tensor</code> (one for the output of the embeddings + 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 initial embedding outputs.</p>
</li>
<li>
<p><strong>attentions</strong> (<code>tuple(tf.Tensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>tf.Tensor</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>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_tf_outputs.TFBaseModelOutput"
>transformers.modeling_tf_outputs.TFBaseModelOutput</a> or <code>tuple(tf.Tensor)</code></p>
`}}),ge=new mt({props:{$$slots:{default:[mo]},$$scope:{ctx:x}}}),Ze=new qe({props:{anchor:"transformers.TFDebertaModel.call.example",$$slots:{default:[uo]},$$scope:{ctx:x}}}),K=new at({props:{title:"TFDebertaPreTrainedModel",local:"transformers.TFDebertaPreTrainedModel",headingTag:"h2"}}),it=new V({props:{name:"class transformers.TFDebertaPreTrainedModel",anchor:"transformers.TFDebertaPreTrainedModel",parameters:[{name:"*args",val:""},{name:"**kwargs",val:""}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_tf_deberta.py#L1137"}}),je=new V({props:{name:"call",anchor:"transformers.TFDebertaPreTrainedModel.call",parameters:[{name:"inputs",val:""},{name:"training",val:" = None"},{name:"mask",val:" = None"}],source:"https://github.com/huggingface/transformers/blob/main/src/tf_keras/src/engine/training.py#L590",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A tensor if there is a single output, or
a list of tensors if there are more than one outputs.</p>
`}}),be=new at({props:{title:"TFDebertaForMaskedLM",local:"transformers.TFDebertaForMaskedLM",headingTag:"h2"}}),Ee=new V({props:{name:"class transformers.TFDebertaForMaskedLM",anchor:"transformers.TFDebertaForMaskedLM",parameters:[{name:"config",val:": DebertaConfig"},{name:"*inputs",val:""},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.TFDebertaForMaskedLM.config",description:`<strong>config</strong> (<a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaConfig">DebertaConfig</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/main/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_tf_deberta.py#L1285"}}),Mt=new mt({props:{$$slots:{default:[fo]},$$scope:{ctx:x}}}),zt=new V({props:{name:"call",anchor:"transformers.TFDebertaForMaskedLM.call",parameters:[{name:"input_ids",val:": TFModelInputType | None = None"},{name:"attention_mask",val:": np.ndarray | tf.Tensor | None = None"},{name:"token_type_ids",val:": np.ndarray | tf.Tensor | None = None"},{name:"position_ids",val:": np.ndarray | tf.Tensor | None = None"},{name:"inputs_embeds",val:": np.ndarray | tf.Tensor | None = None"},{name:"output_attentions",val:": Optional[bool] = None"},{name:"output_hidden_states",val:": Optional[bool] = None"},{name:"return_dict",val:": Optional[bool] = None"},{name:"labels",val:": np.ndarray | tf.Tensor | None = None"},{name:"training",val:": Optional[bool] = False"}],parametersDescription:[{anchor:"transformers.TFDebertaForMaskedLM.call.input_ids",description:`<strong>input_ids</strong> (<code>np.ndarray</code>, <code>tf.Tensor</code>, <code>List[tf.Tensor]</code> \`<code>Dict[str, tf.Tensor]</code> or <code>Dict[str, np.ndarray]</code> and each example must have the shape <code>(batch_size, sequence_length)</code>) &#x2014;
Indices of input sequence tokens in the vocabulary.</p>
<p>Indices can be obtained using <a href="/docs/transformers/main/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/main/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/main/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.TFDebertaForMaskedLM.call.attention_mask",description:`<strong>attention_mask</strong> (<code>np.ndarray</code> or <code>tf.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.TFDebertaForMaskedLM.call.token_type_ids",description:`<strong>token_type_ids</strong> (<code>np.ndarray</code> or <code>tf.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.TFDebertaForMaskedLM.call.position_ids",description:`<strong>position_ids</strong> (<code>np.ndarray</code> or <code>tf.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.max_position_embeddings - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.TFDebertaForMaskedLM.call.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>np.ndarray</code> or <code>tf.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 <em>input_ids</em> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.TFDebertaForMaskedLM.call.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.TFDebertaForMaskedLM.call.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.TFDebertaForMaskedLM.call.return_dict",description:"<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;\nWhether or not to return a [`~utils.ModelOutput&#x201C;] instead of a plain tuple.",name:"return_dict"},{anchor:"transformers.TFDebertaForMaskedLM.call.labels",description:`<strong>labels</strong> (<code>tf.Tensor</code> or <code>np.ndarray</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"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_tf_deberta.py#L1302",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_tf_outputs.TFMaskedLMOutput"
>transformers.modeling_tf_outputs.TFMaskedLMOutput</a> or a tuple of <code>tf.Tensor</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/main/en/model_doc/deberta#transformers.DebertaConfig"
>DebertaConfig</a>) and inputs.</p>
<ul>
<li>
<p><strong>loss</strong> (<code>tf.Tensor</code> of shape <code>(n,)</code>, <em>optional</em>, where n is the number of non-masked labels, returned when <code>labels</code> is provided) — Masked language modeling (MLM) loss.</p>
</li>
<li>
<p><strong>logits</strong> (<code>tf.Tensor</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(tf.Tensor)</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>tf.Tensor</code> (one for the output of the embeddings + 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 initial embedding outputs.</p>
</li>
<li>
<p><strong>attentions</strong> (<code>tuple(tf.Tensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>tf.Tensor</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>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_tf_outputs.TFMaskedLMOutput"
>transformers.modeling_tf_outputs.TFMaskedLMOutput</a> or <code>tuple(tf.Tensor)</code></p>
`}}),ht=new mt({props:{$$slots:{default:[ho]},$$scope:{ctx:x}}}),H=new qe({props:{anchor:"transformers.TFDebertaForMaskedLM.call.example",$$slots:{default:[go]},$$scope:{ctx:x}}}),Ge=new qe({props:{anchor:"transformers.TFDebertaForMaskedLM.call.example-2",$$slots:{default:[bo]},$$scope:{ctx:x}}}),qt=new at({props:{title:"TFDebertaForSequenceClassification",local:"transformers.TFDebertaForSequenceClassification",headingTag:"h2"}}),xt=new V({props:{name:"class transformers.TFDebertaForSequenceClassification",anchor:"transformers.TFDebertaForSequenceClassification",parameters:[{name:"config",val:": DebertaConfig"},{name:"*inputs",val:""},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.TFDebertaForSequenceClassification.config",description:`<strong>config</strong> (<a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaConfig">DebertaConfig</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/main/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_tf_deberta.py#L1366"}}),Te=new mt({props:{$$slots:{default:[_o]},$$scope:{ctx:x}}}),lt=new V({props:{name:"call",anchor:"transformers.TFDebertaForSequenceClassification.call",parameters:[{name:"input_ids",val:": TFModelInputType | None = None"},{name:"attention_mask",val:": np.ndarray | tf.Tensor | None = None"},{name:"token_type_ids",val:": np.ndarray | tf.Tensor | None = None"},{name:"position_ids",val:": np.ndarray | tf.Tensor | None = None"},{name:"inputs_embeds",val:": np.ndarray | tf.Tensor | None = None"},{name:"output_attentions",val:": Optional[bool] = None"},{name:"output_hidden_states",val:": Optional[bool] = None"},{name:"return_dict",val:": Optional[bool] = None"},{name:"labels",val:": np.ndarray | tf.Tensor | None = None"},{name:"training",val:": Optional[bool] = False"}],parametersDescription:[{anchor:"transformers.TFDebertaForSequenceClassification.call.input_ids",description:`<strong>input_ids</strong> (<code>np.ndarray</code>, <code>tf.Tensor</code>, <code>List[tf.Tensor]</code> \`<code>Dict[str, tf.Tensor]</code> or <code>Dict[str, np.ndarray]</code> and each example must have the shape <code>(batch_size, sequence_length)</code>) &#x2014;
Indices of input sequence tokens in the vocabulary.</p>
<p>Indices can be obtained using <a href="/docs/transformers/main/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/main/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/main/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.TFDebertaForSequenceClassification.call.attention_mask",description:`<strong>attention_mask</strong> (<code>np.ndarray</code> or <code>tf.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.TFDebertaForSequenceClassification.call.token_type_ids",description:`<strong>token_type_ids</strong> (<code>np.ndarray</code> or <code>tf.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.TFDebertaForSequenceClassification.call.position_ids",description:`<strong>position_ids</strong> (<code>np.ndarray</code> or <code>tf.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.max_position_embeddings - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.TFDebertaForSequenceClassification.call.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>np.ndarray</code> or <code>tf.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 <em>input_ids</em> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.TFDebertaForSequenceClassification.call.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.TFDebertaForSequenceClassification.call.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.TFDebertaForSequenceClassification.call.return_dict",description:"<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;\nWhether or not to return a [`~utils.ModelOutput&#x201C;] instead of a plain tuple.",name:"return_dict"},{anchor:"transformers.TFDebertaForSequenceClassification.call.labels",description:`<strong>labels</strong> (<code>tf.Tensor</code> or <code>np.ndarray</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"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_tf_deberta.py#L1392",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_tf_outputs.TFSequenceClassifierOutput"
>transformers.modeling_tf_outputs.TFSequenceClassifierOutput</a> or a tuple of <code>tf.Tensor</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/main/en/model_doc/deberta#transformers.DebertaConfig"
>DebertaConfig</a>) and inputs.</p>
<ul>
<li>
<p><strong>loss</strong> (<code>tf.Tensor</code> of shape <code>(batch_size, )</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>tf.Tensor</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(tf.Tensor)</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>tf.Tensor</code> (one for the output of the embeddings + 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 initial embedding outputs.</p>
</li>
<li>
<p><strong>attentions</strong> (<code>tuple(tf.Tensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>tf.Tensor</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>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_tf_outputs.TFSequenceClassifierOutput"
>transformers.modeling_tf_outputs.TFSequenceClassifierOutput</a> or <code>tuple(tf.Tensor)</code></p>
`}}),ne=new mt({props:{$$slots:{default:[To]},$$scope:{ctx:x}}}),Ce=new qe({props:{anchor:"transformers.TFDebertaForSequenceClassification.call.example",$$slots:{default:[yo]},$$scope:{ctx:x}}}),Ke=new qe({props:{anchor:"transformers.TFDebertaForSequenceClassification.call.example-2",$$slots:{default:[wo]},$$scope:{ctx:x}}}),De=new at({props:{title:"TFDebertaForTokenClassification",local:"transformers.TFDebertaForTokenClassification",headingTag:"h2"}}),bt=new V({props:{name:"class transformers.TFDebertaForTokenClassification",anchor:"transformers.TFDebertaForTokenClassification",parameters:[{name:"config",val:": DebertaConfig"},{name:"*inputs",val:""},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.TFDebertaForTokenClassification.config",description:`<strong>config</strong> (<a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaConfig">DebertaConfig</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/main/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_tf_deberta.py#L1465"}}),ct=new mt({props:{$$slots:{default:[ko]},$$scope:{ctx:x}}}),Z=new V({props:{name:"call",anchor:"transformers.TFDebertaForTokenClassification.call",parameters:[{name:"input_ids",val:": TFModelInputType | None = None"},{name:"attention_mask",val:": np.ndarray | tf.Tensor | None = None"},{name:"token_type_ids",val:": np.ndarray | tf.Tensor | None = None"},{name:"position_ids",val:": np.ndarray | tf.Tensor | None = None"},{name:"inputs_embeds",val:": np.ndarray | tf.Tensor | None = None"},{name:"output_attentions",val:": Optional[bool] = None"},{name:"output_hidden_states",val:": Optional[bool] = None"},{name:"return_dict",val:": Optional[bool] = None"},{name:"labels",val:": np.ndarray | tf.Tensor | None = None"},{name:"training",val:": Optional[bool] = False"}],parametersDescription:[{anchor:"transformers.TFDebertaForTokenClassification.call.input_ids",description:`<strong>input_ids</strong> (<code>np.ndarray</code>, <code>tf.Tensor</code>, <code>List[tf.Tensor]</code> \`<code>Dict[str, tf.Tensor]</code> or <code>Dict[str, np.ndarray]</code> and each example must have the shape <code>(batch_size, sequence_length)</code>) &#x2014;
Indices of input sequence tokens in the vocabulary.</p>
<p>Indices can be obtained using <a href="/docs/transformers/main/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/main/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/main/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.TFDebertaForTokenClassification.call.attention_mask",description:`<strong>attention_mask</strong> (<code>np.ndarray</code> or <code>tf.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.TFDebertaForTokenClassification.call.token_type_ids",description:`<strong>token_type_ids</strong> (<code>np.ndarray</code> or <code>tf.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.TFDebertaForTokenClassification.call.position_ids",description:`<strong>position_ids</strong> (<code>np.ndarray</code> or <code>tf.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.max_position_embeddings - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.TFDebertaForTokenClassification.call.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>np.ndarray</code> or <code>tf.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 <em>input_ids</em> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.TFDebertaForTokenClassification.call.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.TFDebertaForTokenClassification.call.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.TFDebertaForTokenClassification.call.return_dict",description:"<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;\nWhether or not to return a [`~utils.ModelOutput&#x201C;] instead of a plain tuple.",name:"return_dict"},{anchor:"transformers.TFDebertaForTokenClassification.call.labels",description:`<strong>labels</strong> (<code>tf.Tensor</code> or <code>np.ndarray</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"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_tf_deberta.py#L1485",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_tf_outputs.TFTokenClassifierOutput"
>transformers.modeling_tf_outputs.TFTokenClassifierOutput</a> or a tuple of <code>tf.Tensor</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/main/en/model_doc/deberta#transformers.DebertaConfig"
>DebertaConfig</a>) and inputs.</p>
<ul>
<li>
<p><strong>loss</strong> (<code>tf.Tensor</code> of shape <code>(n,)</code>, <em>optional</em>, where n is the number of unmasked labels, returned when <code>labels</code> is provided)  — Classification loss.</p>
</li>
<li>
<p><strong>logits</strong> (<code>tf.Tensor</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(tf.Tensor)</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>tf.Tensor</code> (one for the output of the embeddings + 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 initial embedding outputs.</p>
</li>
<li>
<p><strong>attentions</strong> (<code>tuple(tf.Tensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>tf.Tensor</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>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_tf_outputs.TFTokenClassifierOutput"
>transformers.modeling_tf_outputs.TFTokenClassifierOutput</a> or <code>tuple(tf.Tensor)</code></p>
`}}),jt=new mt({props:{$$slots:{default:[Mo]},$$scope:{ctx:x}}}),Ue=new qe({props:{anchor:"transformers.TFDebertaForTokenClassification.call.example",$$slots:{default:[$o]},$$scope:{ctx:x}}}),O=new qe({props:{anchor:"transformers.TFDebertaForTokenClassification.call.example-2",$$slots:{default:[vo]},$$scope:{ctx:x}}}),f=new at({props:{title:"TFDebertaForQuestionAnswering",local:"transformers.TFDebertaForQuestionAnswering",headingTag:"h2"}}),de=new V({props:{name:"class transformers.TFDebertaForQuestionAnswering",anchor:"transformers.TFDebertaForQuestionAnswering",parameters:[{name:"config",val:": DebertaConfig"},{name:"*inputs",val:""},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.TFDebertaForQuestionAnswering.config",description:`<strong>config</strong> (<a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaConfig">DebertaConfig</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/main/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_tf_deberta.py#L1548"}}),Ft=new mt({props:{$$slots:{default:[xo]},$$scope:{ctx:x}}}),Ie=new V({props:{name:"call",anchor:"transformers.TFDebertaForQuestionAnswering.call",parameters:[{name:"input_ids",val:": TFModelInputType | None = None"},{name:"attention_mask",val:": np.ndarray | tf.Tensor | None = None"},{name:"token_type_ids",val:": np.ndarray | tf.Tensor | None = None"},{name:"position_ids",val:": np.ndarray | tf.Tensor | None = None"},{name:"inputs_embeds",val:": np.ndarray | tf.Tensor | None = None"},{name:"output_attentions",val:": Optional[bool] = None"},{name:"output_hidden_states",val:": Optional[bool] = None"},{name:"return_dict",val:": Optional[bool] = None"},{name:"start_positions",val:": np.ndarray | tf.Tensor | None = None"},{name:"end_positions",val:": np.ndarray | tf.Tensor | None = None"},{name:"training",val:": Optional[bool] = False"}],parametersDescription:[{anchor:"transformers.TFDebertaForQuestionAnswering.call.input_ids",description:`<strong>input_ids</strong> (<code>np.ndarray</code>, <code>tf.Tensor</code>, <code>List[tf.Tensor]</code> \`<code>Dict[str, tf.Tensor]</code> or <code>Dict[str, np.ndarray]</code> and each example must have the shape <code>(batch_size, sequence_length)</code>) &#x2014;
Indices of input sequence tokens in the vocabulary.</p>
<p>Indices can be obtained using <a href="/docs/transformers/main/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/main/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/main/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.TFDebertaForQuestionAnswering.call.attention_mask",description:`<strong>attention_mask</strong> (<code>np.ndarray</code> or <code>tf.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.TFDebertaForQuestionAnswering.call.token_type_ids",description:`<strong>token_type_ids</strong> (<code>np.ndarray</code> or <code>tf.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.TFDebertaForQuestionAnswering.call.position_ids",description:`<strong>position_ids</strong> (<code>np.ndarray</code> or <code>tf.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.max_position_embeddings - 1]</code>.</p>
<p><a href="../glossary#position-ids">What are position IDs?</a>`,name:"position_ids"},{anchor:"transformers.TFDebertaForQuestionAnswering.call.inputs_embeds",description:`<strong>inputs_embeds</strong> (<code>np.ndarray</code> or <code>tf.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 <em>input_ids</em> indices into associated vectors than the
model&#x2019;s internal embedding lookup matrix.`,name:"inputs_embeds"},{anchor:"transformers.TFDebertaForQuestionAnswering.call.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.TFDebertaForQuestionAnswering.call.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.TFDebertaForQuestionAnswering.call.return_dict",description:"<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;\nWhether or not to return a [`~utils.ModelOutput&#x201C;] instead of a plain tuple.",name:"return_dict"},{anchor:"transformers.TFDebertaForQuestionAnswering.call.start_positions",description:`<strong>start_positions</strong> (<code>tf.Tensor</code> or <code>np.ndarray</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.TFDebertaForQuestionAnswering.call.end_positions",description:`<strong>end_positions</strong> (<code>tf.Tensor</code> or <code>np.ndarray</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"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/modeling_tf_deberta.py#L1567",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_tf_outputs.TFQuestionAnsweringModelOutput"
>transformers.modeling_tf_outputs.TFQuestionAnsweringModelOutput</a> or a tuple of <code>tf.Tensor</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/main/en/model_doc/deberta#transformers.DebertaConfig"
>DebertaConfig</a>) and inputs.</p>
<ul>
<li>
<p><strong>loss</strong> (<code>tf.Tensor</code> of shape <code>(batch_size, )</code>, <em>optional</em>, returned when <code>start_positions</code> and <code>end_positions</code> are 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>tf.Tensor</code> of shape <code>(batch_size, sequence_length)</code>) — Span-start scores (before SoftMax).</p>
</li>
<li>
<p><strong>end_logits</strong> (<code>tf.Tensor</code> of shape <code>(batch_size, sequence_length)</code>) — Span-end scores (before SoftMax).</p>
</li>
<li>
<p><strong>hidden_states</strong> (<code>tuple(tf.Tensor)</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>tf.Tensor</code> (one for the output of the embeddings + 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 initial embedding outputs.</p>
</li>
<li>
<p><strong>attentions</strong> (<code>tuple(tf.Tensor)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>tf.Tensor</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>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/main/en/main_classes/output#transformers.modeling_tf_outputs.TFQuestionAnsweringModelOutput"
>transformers.modeling_tf_outputs.TFQuestionAnsweringModelOutput</a> or <code>tuple(tf.Tensor)</code></p>
`}}),on=new mt({props:{$$slots:{default:[Jo]},$$scope:{ctx:x}}}),an=new qe({props:{anchor:"transformers.TFDebertaForQuestionAnswering.call.example",$$slots:{default:[jo]},$$scope:{ctx:x}}}),fn=new qe({props:{anchor:"transformers.TFDebertaForQuestionAnswering.call.example-2",$$slots:{default:[Fo]},$$scope:{ctx:x}}}),{c(){T(e.$$.fragment),u=r(),n=p("div"),T(o.$$.fragment),_=r(),t=p("p"),t.innerHTML=b,Y=r(),U=p("p"),U.innerHTML=z,Q=r(),T(C.$$.fragment),D=r(),R=p("div"),T(h.$$.fragment),F=r(),yt=p("p"),yt.innerHTML=Be,gn=r(),T(ge.$$.fragment),ut=r(),T(Ze.$$.fragment),Ct=r(),T(K.$$.fragment),rt=r(),pe=p("div"),T(it.$$.fragment),ln=r(),We=p("p"),We.textContent=wt,ft=r(),X=p("div"),T(je.$$.fragment),kt=r(),Xe=p("p"),Xe.textContent=Nt,bn=r(),Se=p("p"),Se.innerHTML=Xt,Ot=r(),P=p("p"),P.innerHTML=St,Yt=r(),T(be.$$.fragment),Kt=r(),A=p("div"),T(Ee.$$.fragment),en=r(),me=p("p"),me.innerHTML=tn,ee=r(),Le=p("p"),Le.innerHTML=_e,Ye=r(),T(Mt.$$.fragment),Dt=r(),ie=p("div"),T(zt.$$.fragment),Ut=r(),Qe=p("p"),Qe.innerHTML=dn,Pe=r(),T(ht.$$.fragment),B=r(),T(H.$$.fragment),$t=r(),T(Ge.$$.fragment),It=r(),T(qt.$$.fragment),vt=r(),S=p("div"),T(xt.$$.fragment),Me=r(),Rt=p("p"),Rt.textContent=Ae,cn=r(),Oe=p("p"),Oe.innerHTML=gt,ue=r(),$e=p("p"),$e.innerHTML=kn,Bt=r(),T(Te.$$.fragment),Jt=r(),te=p("div"),T(lt.$$.fragment),_n=r(),dt=p("p"),dt.innerHTML=nn,Tn=r(),T(ne.$$.fragment),Fe=r(),T(Ce.$$.fragment),Zt=r(),T(Ke.$$.fragment),pn=r(),T(De.$$.fragment),ve=r(),W=p("div"),T(bt.$$.fragment),et=r(),Wt=p("p"),Wt.textContent=ye,ze=r(),Et=p("p"),Et.innerHTML=_t,Mn=r(),Lt=p("p"),Lt.innerHTML=Tt,Gt=r(),T(ct.$$.fragment),tt=r(),we=p("div"),T(Z.$$.fragment),le=r(),nt=p("p"),nt.innerHTML=Qt,Pt=r(),T(jt.$$.fragment),Vt=r(),T(Ue.$$.fragment),mn=r(),T(O.$$.fragment),At=r(),T(f.$$.fragment),j=r(),G=p("div"),T(de.$$.fragment),Ve=r(),oe=p("p"),oe.innerHTML=He,xe=r(),N=p("p"),N.innerHTML=ke,Je=r(),se=p("p"),se.innerHTML=xn,jn=r(),T(Ft.$$.fragment),Fn=r(),ot=p("div"),T(Ie.$$.fragment),un=r(),yn=p("p"),yn.innerHTML=$n,Cn=r(),T(on.$$.fragment),sn=r(),T(an.$$.fragment),wn=r(),T(fn.$$.fragment),this.h()},l(c){y(e.$$.fragment,c),u=i(c),n=m(c,"DIV",{class:!0});var J=E(n);y(o.$$.fragment,J),_=i(J),t=m(J,"P",{"data-svelte-h":!0}),g(t)!=="svelte-1bcviae"&&(t.innerHTML=b),Y=i(J),U=m(J,"P",{"data-svelte-h":!0}),g(U)!=="svelte-1be7e3c"&&(U.innerHTML=z),Q=i(J),y(C.$$.fragment,J),D=i(J),R=m(J,"DIV",{class:!0});var Ne=E(R);y(h.$$.fragment,Ne),F=i(Ne),yt=m(Ne,"P",{"data-svelte-h":!0}),g(yt)!=="svelte-nh55d8"&&(yt.innerHTML=Be),gn=i(Ne),y(ge.$$.fragment,Ne),ut=i(Ne),y(Ze.$$.fragment,Ne),Ne.forEach(a),J.forEach(a),Ct=i(c),y(K.$$.fragment,c),rt=i(c),pe=m(c,"DIV",{class:!0});var Ht=E(pe);y(it.$$.fragment,Ht),ln=i(Ht),We=m(Ht,"P",{"data-svelte-h":!0}),g(We)!=="svelte-nj1b0k"&&(We.textContent=wt),ft=i(Ht),X=m(Ht,"DIV",{class:!0});var pt=E(X);y(je.$$.fragment,pt),kt=i(pt),Xe=m(pt,"P",{"data-svelte-h":!0}),g(Xe)!=="svelte-4lztt8"&&(Xe.textContent=Nt),bn=i(pt),Se=m(pt,"P",{"data-svelte-h":!0}),g(Se)!=="svelte-bk2n17"&&(Se.innerHTML=Xt),Ot=i(pt),P=m(pt,"P",{"data-svelte-h":!0}),g(P)!=="svelte-18vwu1n"&&(P.innerHTML=St),pt.forEach(a),Ht.forEach(a),Yt=i(c),y(be.$$.fragment,c),Kt=i(c),A=m(c,"DIV",{class:!0});var st=E(A);y(Ee.$$.fragment,st),en=i(st),me=m(st,"P",{"data-svelte-h":!0}),g(me)!=="svelte-qd1vq5"&&(me.innerHTML=tn),ee=i(st),Le=m(st,"P",{"data-svelte-h":!0}),g(Le)!=="svelte-1be7e3c"&&(Le.innerHTML=_e),Ye=i(st),y(Mt.$$.fragment,st),Dt=i(st),ie=m(st,"DIV",{class:!0});var s=E(ie);y(zt.$$.fragment,s),Ut=i(s),Qe=m(s,"P",{"data-svelte-h":!0}),g(Qe)!=="svelte-1qyqmuk"&&(Qe.innerHTML=dn),Pe=i(s),y(ht.$$.fragment,s),B=i(s),y(H.$$.fragment,s),$t=i(s),y(Ge.$$.fragment,s),s.forEach(a),st.forEach(a),It=i(c),y(qt.$$.fragment,c),vt=i(c),S=m(c,"DIV",{class:!0});var v=E(S);y(xt.$$.fragment,v),Me=i(v),Rt=m(v,"P",{"data-svelte-h":!0}),g(Rt)!=="svelte-1xhb56d"&&(Rt.textContent=Ae),cn=i(v),Oe=m(v,"P",{"data-svelte-h":!0}),g(Oe)!=="svelte-1bqto0e"&&(Oe.innerHTML=gt),ue=i(v),$e=m(v,"P",{"data-svelte-h":!0}),g($e)!=="svelte-1be7e3c"&&($e.innerHTML=kn),Bt=i(v),y(Te.$$.fragment,v),Jt=i(v),te=m(v,"DIV",{class:!0});var ae=E(te);y(lt.$$.fragment,ae),_n=i(ae),dt=m(ae,"P",{"data-svelte-h":!0}),g(dt)!=="svelte-1xpz5uy"&&(dt.innerHTML=nn),Tn=i(ae),y(ne.$$.fragment,ae),Fe=i(ae),y(Ce.$$.fragment,ae),Zt=i(ae),y(Ke.$$.fragment,ae),ae.forEach(a),v.forEach(a),pn=i(c),y(De.$$.fragment,c),ve=i(c),W=m(c,"DIV",{class:!0});var I=E(W);y(bt.$$.fragment,I),et=i(I),Wt=m(I,"P",{"data-svelte-h":!0}),g(Wt)!=="svelte-1t75rxs"&&(Wt.textContent=ye),ze=i(I),Et=m(I,"P",{"data-svelte-h":!0}),g(Et)!=="svelte-1bqto0e"&&(Et.innerHTML=_t),Mn=i(I),Lt=m(I,"P",{"data-svelte-h":!0}),g(Lt)!=="svelte-1be7e3c"&&(Lt.innerHTML=Tt),Gt=i(I),y(ct.$$.fragment,I),tt=i(I),we=m(I,"DIV",{class:!0});var fe=E(we);y(Z.$$.fragment,fe),le=i(fe),nt=m(fe,"P",{"data-svelte-h":!0}),g(nt)!=="svelte-1lrtwr4"&&(nt.innerHTML=Qt),Pt=i(fe),y(jt.$$.fragment,fe),Vt=i(fe),y(Ue.$$.fragment,fe),mn=i(fe),y(O.$$.fragment,fe),fe.forEach(a),I.forEach(a),At=i(c),y(f.$$.fragment,c),j=i(c),G=m(c,"DIV",{class:!0});var he=E(G);y(de.$$.fragment,he),Ve=i(he),oe=m(he,"P",{"data-svelte-h":!0}),g(oe)!=="svelte-1nbcy0m"&&(oe.innerHTML=He),xe=i(he),N=m(he,"P",{"data-svelte-h":!0}),g(N)!=="svelte-1bqto0e"&&(N.innerHTML=ke),Je=i(he),se=m(he,"P",{"data-svelte-h":!0}),g(se)!=="svelte-1be7e3c"&&(se.innerHTML=xn),jn=i(he),y(Ft.$$.fragment,he),Fn=i(he),ot=m(he,"DIV",{class:!0});var re=E(ot);y(Ie.$$.fragment,re),un=i(re),yn=m(re,"P",{"data-svelte-h":!0}),g(yn)!=="svelte-1olndow"&&(yn.innerHTML=$n),Cn=i(re),y(on.$$.fragment,re),sn=i(re),y(an.$$.fragment,re),wn=i(re),y(fn.$$.fragment,re),re.forEach(a),he.forEach(a),this.h()},h(){L(R,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(n,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(X,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(pe,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(ie,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(A,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(te,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(S,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(we,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(W,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(ot,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(G,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8")},m(c,J){w(e,c,J),d(c,u,J),d(c,n,J),w(o,n,null),l(n,_),l(n,t),l(n,Y),l(n,U),l(n,Q),w(C,n,null),l(n,D),l(n,R),w(h,R,null),l(R,F),l(R,yt),l(R,gn),w(ge,R,null),l(R,ut),w(Ze,R,null),d(c,Ct,J),w(K,c,J),d(c,rt,J),d(c,pe,J),w(it,pe,null),l(pe,ln),l(pe,We),l(pe,ft),l(pe,X),w(je,X,null),l(X,kt),l(X,Xe),l(X,bn),l(X,Se),l(X,Ot),l(X,P),d(c,Yt,J),w(be,c,J),d(c,Kt,J),d(c,A,J),w(Ee,A,null),l(A,en),l(A,me),l(A,ee),l(A,Le),l(A,Ye),w(Mt,A,null),l(A,Dt),l(A,ie),w(zt,ie,null),l(ie,Ut),l(ie,Qe),l(ie,Pe),w(ht,ie,null),l(ie,B),w(H,ie,null),l(ie,$t),w(Ge,ie,null),d(c,It,J),w(qt,c,J),d(c,vt,J),d(c,S,J),w(xt,S,null),l(S,Me),l(S,Rt),l(S,cn),l(S,Oe),l(S,ue),l(S,$e),l(S,Bt),w(Te,S,null),l(S,Jt),l(S,te),w(lt,te,null),l(te,_n),l(te,dt),l(te,Tn),w(ne,te,null),l(te,Fe),w(Ce,te,null),l(te,Zt),w(Ke,te,null),d(c,pn,J),w(De,c,J),d(c,ve,J),d(c,W,J),w(bt,W,null),l(W,et),l(W,Wt),l(W,ze),l(W,Et),l(W,Mn),l(W,Lt),l(W,Gt),w(ct,W,null),l(W,tt),l(W,we),w(Z,we,null),l(we,le),l(we,nt),l(we,Pt),w(jt,we,null),l(we,Vt),w(Ue,we,null),l(we,mn),w(O,we,null),d(c,At,J),w(f,c,J),d(c,j,J),d(c,G,J),w(de,G,null),l(G,Ve),l(G,oe),l(G,xe),l(G,N),l(G,Je),l(G,se),l(G,jn),w(Ft,G,null),l(G,Fn),l(G,ot),w(Ie,ot,null),l(ot,un),l(ot,yn),l(ot,Cn),w(on,ot,null),l(ot,sn),w(an,ot,null),l(ot,wn),w(fn,ot,null),vn=!0},p(c,J){const Ne={};J&2&&(Ne.$$scope={dirty:J,ctx:c}),C.$set(Ne);const Ht={};J&2&&(Ht.$$scope={dirty:J,ctx:c}),ge.$set(Ht);const pt={};J&2&&(pt.$$scope={dirty:J,ctx:c}),Ze.$set(pt);const st={};J&2&&(st.$$scope={dirty:J,ctx:c}),Mt.$set(st);const s={};J&2&&(s.$$scope={dirty:J,ctx:c}),ht.$set(s);const v={};J&2&&(v.$$scope={dirty:J,ctx:c}),H.$set(v);const ae={};J&2&&(ae.$$scope={dirty:J,ctx:c}),Ge.$set(ae);const I={};J&2&&(I.$$scope={dirty:J,ctx:c}),Te.$set(I);const fe={};J&2&&(fe.$$scope={dirty:J,ctx:c}),ne.$set(fe);const he={};J&2&&(he.$$scope={dirty:J,ctx:c}),Ce.$set(he);const re={};J&2&&(re.$$scope={dirty:J,ctx:c}),Ke.$set(re);const Jn={};J&2&&(Jn.$$scope={dirty:J,ctx:c}),ct.$set(Jn);const ce={};J&2&&(ce.$$scope={dirty:J,ctx:c}),jt.$set(ce);const rn={};J&2&&(rn.$$scope={dirty:J,ctx:c}),Ue.$set(rn);const hn={};J&2&&(hn.$$scope={dirty:J,ctx:c}),O.$set(hn);const zn={};J&2&&(zn.$$scope={dirty:J,ctx:c}),Ft.$set(zn);const Un={};J&2&&(Un.$$scope={dirty:J,ctx:c}),on.$set(Un);const In={};J&2&&(In.$$scope={dirty:J,ctx:c}),an.$set(In);const qn={};J&2&&(qn.$$scope={dirty:J,ctx:c}),fn.$set(qn)},i(c){vn||(k(e.$$.fragment,c),k(o.$$.fragment,c),k(C.$$.fragment,c),k(h.$$.fragment,c),k(ge.$$.fragment,c),k(Ze.$$.fragment,c),k(K.$$.fragment,c),k(it.$$.fragment,c),k(je.$$.fragment,c),k(be.$$.fragment,c),k(Ee.$$.fragment,c),k(Mt.$$.fragment,c),k(zt.$$.fragment,c),k(ht.$$.fragment,c),k(H.$$.fragment,c),k(Ge.$$.fragment,c),k(qt.$$.fragment,c),k(xt.$$.fragment,c),k(Te.$$.fragment,c),k(lt.$$.fragment,c),k(ne.$$.fragment,c),k(Ce.$$.fragment,c),k(Ke.$$.fragment,c),k(De.$$.fragment,c),k(bt.$$.fragment,c),k(ct.$$.fragment,c),k(Z.$$.fragment,c),k(jt.$$.fragment,c),k(Ue.$$.fragment,c),k(O.$$.fragment,c),k(f.$$.fragment,c),k(de.$$.fragment,c),k(Ft.$$.fragment,c),k(Ie.$$.fragment,c),k(on.$$.fragment,c),k(an.$$.fragment,c),k(fn.$$.fragment,c),vn=!0)},o(c){M(e.$$.fragment,c),M(o.$$.fragment,c),M(C.$$.fragment,c),M(h.$$.fragment,c),M(ge.$$.fragment,c),M(Ze.$$.fragment,c),M(K.$$.fragment,c),M(it.$$.fragment,c),M(je.$$.fragment,c),M(be.$$.fragment,c),M(Ee.$$.fragment,c),M(Mt.$$.fragment,c),M(zt.$$.fragment,c),M(ht.$$.fragment,c),M(H.$$.fragment,c),M(Ge.$$.fragment,c),M(qt.$$.fragment,c),M(xt.$$.fragment,c),M(Te.$$.fragment,c),M(lt.$$.fragment,c),M(ne.$$.fragment,c),M(Ce.$$.fragment,c),M(Ke.$$.fragment,c),M(De.$$.fragment,c),M(bt.$$.fragment,c),M(ct.$$.fragment,c),M(Z.$$.fragment,c),M(jt.$$.fragment,c),M(Ue.$$.fragment,c),M(O.$$.fragment,c),M(f.$$.fragment,c),M(de.$$.fragment,c),M(Ft.$$.fragment,c),M(Ie.$$.fragment,c),M(on.$$.fragment,c),M(an.$$.fragment,c),M(fn.$$.fragment,c),vn=!1},d(c){c&&(a(u),a(n),a(Ct),a(rt),a(pe),a(Yt),a(Kt),a(A),a(It),a(vt),a(S),a(pn),a(ve),a(W),a(At),a(j),a(G)),$(e,c),$(o),$(C),$(h),$(ge),$(Ze),$(K,c),$(it),$(je),$(be,c),$(Ee),$(Mt),$(zt),$(ht),$(H),$(Ge),$(qt,c),$(xt),$(Te),$(lt),$(ne),$(Ce),$(Ke),$(De,c),$(bt),$(ct),$(Z),$(jt),$(Ue),$(O),$(f,c),$(de),$(Ft),$(Ie),$(on),$(an),$(fn)}}}function Do(x){let e,u;return e=new Rn({props:{$$slots:{default:[Co]},$$scope:{ctx:x}}}),{c(){T(e.$$.fragment)},l(n){y(e.$$.fragment,n)},m(n,o){w(e,n,o),u=!0},p(n,o){const _={};o&2&&(_.$$scope={dirty:o,ctx:n}),e.$set(_)},i(n){u||(k(e.$$.fragment,n),u=!0)},o(n){M(e.$$.fragment,n),u=!1},d(n){$(e,n)}}}function zo(x){let e,u,n,o,_,t,b,Y,U,z=`The DeBERTa model was proposed in <a href="https://arxiv.org/abs/2006.03654" rel="nofollow">DeBERTa: Decoding-enhanced BERT with Disentangled Attention</a> by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen It is based on Google’s
BERT model released in 2018 and Facebook’s RoBERTa model released in 2019.`,Q,C,D=`It builds on RoBERTa with disentangled attention and enhanced mask decoder training with half of the data used in
RoBERTa.`,R,h,F="The abstract from the paper is the following:",yt,Be,gn=`<em>Recent progress in pre-trained neural language models has significantly improved the performance of many natural
language processing (NLP) tasks. In this paper we propose a new model architecture DeBERTa (Decoding-enhanced BERT with
disentangled attention) that improves the BERT and RoBERTa models using two novel techniques. The first is the
disentangled attention mechanism, where each word is represented using two vectors that encode its content and
position, respectively, and the attention weights among words are computed using disentangled matrices on their
contents and relative positions. Second, an enhanced mask decoder is used to replace the output softmax layer to
predict the masked tokens for model pretraining. We show that these two techniques significantly improve the efficiency
of model pretraining and performance of downstream tasks. Compared to RoBERTa-Large, a DeBERTa model trained on half of
the training data performs consistently better on a wide range of NLP tasks, achieving improvements on MNLI by +0.9%
(90.2% vs. 91.1%), on SQuAD v2.0 by +2.3% (88.4% vs. 90.7%) and RACE by +3.6% (83.2% vs. 86.8%). The DeBERTa code and
pre-trained models will be made publicly available at <a href="https://github.com/microsoft/DeBERTa" rel="nofollow">https://github.com/microsoft/DeBERTa</a>.</em>`,ge,ut,Ze=`This model was contributed by <a href="https://huggingface.co/DeBERTa" rel="nofollow">DeBERTa</a>. This model TF 2.0 implementation was
contributed by <a href="https://huggingface.co/kamalkraj" rel="nofollow">kamalkraj</a> . The original code can be found <a href="https://github.com/microsoft/DeBERTa" rel="nofollow">here</a>.`,Ct,K,rt,pe,it="A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with DeBERTa. If you’re interested in submitting a resource to be included here, please feel free to open a Pull Request and we’ll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.",ln,We,wt,ft,X='<li>A blog post on how to <a href="https://huggingface.co/blog/accelerate-deepspeed" rel="nofollow">Accelerate Large Model Training using DeepSpeed</a> with DeBERTa.</li> <li>A blog post on <a href="https://huggingface.co/blog/supercharge-customer-service-with-machine-learning" rel="nofollow">Supercharged Customer Service with Machine Learning</a> with DeBERTa.</li> <li><a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaForSequenceClassification">DebertaForSequenceClassification</a> is supported by this <a href="https://github.com/huggingface/transformers/tree/main/examples/pytorch/text-classification" rel="nofollow">example script</a> and <a href="https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/text_classification.ipynb" rel="nofollow">notebook</a>.</li> <li><a href="/docs/transformers/main/en/model_doc/deberta#transformers.TFDebertaForSequenceClassification">TFDebertaForSequenceClassification</a> is supported by this <a href="https://github.com/huggingface/transformers/tree/main/examples/tensorflow/text-classification" rel="nofollow">example script</a> and <a href="https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/text_classification-tf.ipynb" rel="nofollow">notebook</a>.</li> <li><a href="../tasks/sequence_classification">Text classification task guide</a></li>',je,kt,Xe,Nt,bn='<li><a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaForTokenClassification">DebertaForTokenClassification</a> is supported by this <a href="https://github.com/huggingface/transformers/tree/main/examples/pytorch/token-classification" rel="nofollow">example script</a> and <a href="https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/token_classification.ipynb" rel="nofollow">notebook</a>.</li> <li><a href="/docs/transformers/main/en/model_doc/deberta#transformers.TFDebertaForTokenClassification">TFDebertaForTokenClassification</a> is supported by this <a href="https://github.com/huggingface/transformers/tree/main/examples/tensorflow/token-classification" rel="nofollow">example script</a> and <a href="https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/token_classification-tf.ipynb" rel="nofollow">notebook</a>.</li> <li><a href="https://huggingface.co/course/chapter7/2?fw=pt" rel="nofollow">Token classification</a> chapter of the 🤗 Hugging Face Course.</li> <li><a href="https://huggingface.co/course/chapter6/5?fw=pt" rel="nofollow">Byte-Pair Encoding tokenization</a> chapter of the 🤗 Hugging Face Course.</li> <li><a href="../tasks/token_classification">Token classification task guide</a></li>',Se,Xt,Ot,P,St='<li><a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaForMaskedLM">DebertaForMaskedLM</a> is supported by this <a href="https://github.com/huggingface/transformers/tree/main/examples/pytorch/language-modeling#robertabertdistilbert-and-masked-language-modeling" rel="nofollow">example script</a> and <a href="https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/language_modeling.ipynb" rel="nofollow">notebook</a>.</li> <li><a href="/docs/transformers/main/en/model_doc/deberta#transformers.TFDebertaForMaskedLM">TFDebertaForMaskedLM</a> is supported by this <a href="https://github.com/huggingface/transformers/tree/main/examples/tensorflow/language-modeling#run_mlmpy" rel="nofollow">example script</a> and <a href="https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/language_modeling-tf.ipynb" rel="nofollow">notebook</a>.</li> <li><a href="https://huggingface.co/course/chapter7/3?fw=pt" rel="nofollow">Masked language modeling</a> chapter of the 🤗 Hugging Face Course.</li> <li><a href="../tasks/masked_language_modeling">Masked language modeling task guide</a></li>',Yt,be,Kt,A,Ee='<li><a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaForQuestionAnswering">DebertaForQuestionAnswering</a> is supported by this <a href="https://github.com/huggingface/transformers/tree/main/examples/pytorch/question-answering" rel="nofollow">example script</a> and <a href="https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/question_answering.ipynb" rel="nofollow">notebook</a>.</li> <li><a href="/docs/transformers/main/en/model_doc/deberta#transformers.TFDebertaForQuestionAnswering">TFDebertaForQuestionAnswering</a> is supported by this <a href="https://github.com/huggingface/transformers/tree/main/examples/tensorflow/question-answering" rel="nofollow">example script</a> and <a href="https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/question_answering-tf.ipynb" rel="nofollow">notebook</a>.</li> <li><a href="https://huggingface.co/course/chapter7/7?fw=pt" rel="nofollow">Question answering</a> chapter of the 🤗 Hugging Face Course.</li> <li><a href="../tasks/question_answering">Question answering task guide</a></li>',en,me,tn,ee,Le,_e,Ye,Mt=`This is the configuration class to store the configuration of a <a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaModel">DebertaModel</a> or a <a href="/docs/transformers/main/en/model_doc/deberta#transformers.TFDebertaModel">TFDebertaModel</a>. It is
used to instantiate a DeBERTa 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 DeBERTa
<a href="https://huggingface.co/microsoft/deberta-base" rel="nofollow">microsoft/deberta-base</a> architecture.`,Dt,ie,zt=`Configuration objects inherit from <a href="/docs/transformers/main/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/main/en/main_classes/configuration#transformers.PretrainedConfig">PretrainedConfig</a> for more information.`,Ut,Qe,dn,Pe,ht,B,H,$t,Ge,It="Construct a DeBERTa tokenizer. Based on byte-level Byte-Pair-Encoding.",qt,vt,S="This tokenizer has been trained to treat spaces like parts of the tokens (a bit like sentencepiece) so a word will",xt,Me,Rt,Ae,cn=`You can get around that behavior by passing <code>add_prefix_space=True</code> when instantiating this tokenizer or when you
call it on some text, but since the model was not pretrained this way, it might yield a decrease in performance.`,Oe,gt,ue,$e,kn=`This tokenizer inherits from <a href="/docs/transformers/main/en/main_classes/tokenizer#transformers.PreTrainedTokenizer">PreTrainedTokenizer</a> which contains most of the main methods. Users should refer to
this superclass for more information regarding those methods.`,Bt,Te,Jt,te,lt,_n=`Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
adding special tokens. A DeBERTa sequence has the following format:`,dt,nn,Tn="<li>single sequence: [CLS] X [SEP]</li> <li>pair of sequences: [CLS] A [SEP] B [SEP]</li>",ne,Fe,Ce,Zt,Ke,pn=`Retrieves sequence ids from a token list that has no special tokens added. This method is called when adding
special tokens using the tokenizer <code>prepare_for_model</code> or <code>encode_plus</code> methods.`,De,ve,W,bt,et,Wt="Create a mask from the two sequences passed to be used in a sequence-pair classification task. A DeBERTa",ye,ze,Et,_t,Mn="If <code>token_ids_1</code> is <code>None</code>, this method only returns the first portion of the mask (0s).",Lt,Tt,Gt,ct,tt,we,Z,le,nt,Qt,Pt=`Construct a “fast” DeBERTa tokenizer (backed by HuggingFace’s <em>tokenizers</em> library). Based on byte-level
Byte-Pair-Encoding.`,jt,Vt,Ue="This tokenizer has been trained to treat spaces like parts of the tokens (a bit like sentencepiece) so a word will",mn,O,At,f,j=`You can get around that behavior by passing <code>add_prefix_space=True</code> when instantiating this tokenizer, but since
the model was not pretrained this way, it might yield a decrease in performance.`,G,de,Ve,oe,He=`This tokenizer inherits from <a href="/docs/transformers/main/en/main_classes/tokenizer#transformers.PreTrainedTokenizerFast">PreTrainedTokenizerFast</a> which contains most of the main methods. Users should
refer to this superclass for more information regarding those methods.`,xe,N,ke,Je,se,xn=`Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
adding special tokens. A DeBERTa sequence has the following format:`,jn,Ft,Fn="<li>single sequence: [CLS] X [SEP]</li> <li>pair of sequences: [CLS] A [SEP] B [SEP]</li>",ot,Ie,un,yn,$n,Cn="Create a mask from the two sequences passed to be used in a sequence-pair classification task. A DeBERTa",on,sn,an,wn,fn="If <code>token_ids_1</code> is <code>None</code>, this method only returns the first portion of the mask (0s).",vn,c,J,Ne,Ht,pt,st;return _=new at({props:{title:"DeBERTa",local:"deberta",headingTag:"h1"}}),b=new at({props:{title:"Overview",local:"overview",headingTag:"h2"}}),K=new at({props:{title:"Resources",local:"resources",headingTag:"h2"}}),We=new Dn({props:{pipeline:"text-classification"}}),kt=new Dn({props:{pipeline:"token-classification"}}),Xt=new Dn({props:{pipeline:"fill-mask"}}),be=new Dn({props:{pipeline:"question-answering"}}),me=new at({props:{title:"DebertaConfig",local:"transformers.DebertaConfig",headingTag:"h2"}}),Le=new V({props:{name:"class transformers.DebertaConfig",anchor:"transformers.DebertaConfig",parameters:[{name:"vocab_size",val:" = 50265"},{name:"hidden_size",val:" = 768"},{name:"num_hidden_layers",val:" = 12"},{name:"num_attention_heads",val:" = 12"},{name:"intermediate_size",val:" = 3072"},{name:"hidden_act",val:" = 'gelu'"},{name:"hidden_dropout_prob",val:" = 0.1"},{name:"attention_probs_dropout_prob",val:" = 0.1"},{name:"max_position_embeddings",val:" = 512"},{name:"type_vocab_size",val:" = 0"},{name:"initializer_range",val:" = 0.02"},{name:"layer_norm_eps",val:" = 1e-07"},{name:"relative_attention",val:" = False"},{name:"max_relative_positions",val:" = -1"},{name:"pad_token_id",val:" = 0"},{name:"position_biased_input",val:" = True"},{name:"pos_att_type",val:" = None"},{name:"pooler_dropout",val:" = 0"},{name:"pooler_hidden_act",val:" = 'gelu'"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.DebertaConfig.vocab_size",description:`<strong>vocab_size</strong> (<code>int</code>, <em>optional</em>, defaults to 30522) &#x2014;
Vocabulary size of the DeBERTa model. Defines the number of different tokens that can be represented by the
<code>inputs_ids</code> passed when calling <a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaModel">DebertaModel</a> or <a href="/docs/transformers/main/en/model_doc/deberta#transformers.TFDebertaModel">TFDebertaModel</a>.`,name:"vocab_size"},{anchor:"transformers.DebertaConfig.hidden_size",description:`<strong>hidden_size</strong> (<code>int</code>, <em>optional</em>, defaults to 768) &#x2014;
Dimensionality of the encoder layers and the pooler layer.`,name:"hidden_size"},{anchor:"transformers.DebertaConfig.num_hidden_layers",description:`<strong>num_hidden_layers</strong> (<code>int</code>, <em>optional</em>, defaults to 12) &#x2014;
Number of hidden layers in the Transformer encoder.`,name:"num_hidden_layers"},{anchor:"transformers.DebertaConfig.num_attention_heads",description:`<strong>num_attention_heads</strong> (<code>int</code>, <em>optional</em>, defaults to 12) &#x2014;
Number of attention heads for each attention layer in the Transformer encoder.`,name:"num_attention_heads"},{anchor:"transformers.DebertaConfig.intermediate_size",description:`<strong>intermediate_size</strong> (<code>int</code>, <em>optional</em>, defaults to 3072) &#x2014;
Dimensionality of the &#x201C;intermediate&#x201D; (often named feed-forward) layer in the Transformer encoder.`,name:"intermediate_size"},{anchor:"transformers.DebertaConfig.hidden_act",description:`<strong>hidden_act</strong> (<code>str</code> or <code>Callable</code>, <em>optional</em>, defaults to <code>&quot;gelu&quot;</code>) &#x2014;
The non-linear activation function (function or string) in the encoder and pooler. If string, <code>&quot;gelu&quot;</code>,
<code>&quot;relu&quot;</code>, <code>&quot;silu&quot;</code>, <code>&quot;gelu&quot;</code>, <code>&quot;tanh&quot;</code>, <code>&quot;gelu_fast&quot;</code>, <code>&quot;mish&quot;</code>, <code>&quot;linear&quot;</code>, <code>&quot;sigmoid&quot;</code> and <code>&quot;gelu_new&quot;</code>
are supported.`,name:"hidden_act"},{anchor:"transformers.DebertaConfig.hidden_dropout_prob",description:`<strong>hidden_dropout_prob</strong> (<code>float</code>, <em>optional</em>, defaults to 0.1) &#x2014;
The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.`,name:"hidden_dropout_prob"},{anchor:"transformers.DebertaConfig.attention_probs_dropout_prob",description:`<strong>attention_probs_dropout_prob</strong> (<code>float</code>, <em>optional</em>, defaults to 0.1) &#x2014;
The dropout ratio for the attention probabilities.`,name:"attention_probs_dropout_prob"},{anchor:"transformers.DebertaConfig.max_position_embeddings",description:`<strong>max_position_embeddings</strong> (<code>int</code>, <em>optional</em>, defaults to 512) &#x2014;
The maximum sequence length that this model might ever be used with. Typically set this to something large
just in case (e.g., 512 or 1024 or 2048).`,name:"max_position_embeddings"},{anchor:"transformers.DebertaConfig.type_vocab_size",description:`<strong>type_vocab_size</strong> (<code>int</code>, <em>optional</em>, defaults to 2) &#x2014;
The vocabulary size of the <code>token_type_ids</code> passed when calling <a href="/docs/transformers/main/en/model_doc/deberta#transformers.DebertaModel">DebertaModel</a> or <a href="/docs/transformers/main/en/model_doc/deberta#transformers.TFDebertaModel">TFDebertaModel</a>.`,name:"type_vocab_size"},{anchor:"transformers.DebertaConfig.initializer_range",description:`<strong>initializer_range</strong> (<code>float</code>, <em>optional</em>, defaults to 0.02) &#x2014;
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.`,name:"initializer_range"},{anchor:"transformers.DebertaConfig.layer_norm_eps",description:`<strong>layer_norm_eps</strong> (<code>float</code>, <em>optional</em>, defaults to 1e-12) &#x2014;
The epsilon used by the layer normalization layers.`,name:"layer_norm_eps"},{anchor:"transformers.DebertaConfig.relative_attention",description:`<strong>relative_attention</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) &#x2014;
Whether use relative position encoding.`,name:"relative_attention"},{anchor:"transformers.DebertaConfig.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.DebertaConfig.pad_token_id",description:`<strong>pad_token_id</strong> (<code>int</code>, <em>optional</em>, defaults to 0) &#x2014;
The value used to pad input_ids.`,name:"pad_token_id"},{anchor:"transformers.DebertaConfig.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.DebertaConfig.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>.`,name:"pos_att_type"},{anchor:"transformers.DebertaConfig.layer_norm_eps",description:`<strong>layer_norm_eps</strong> (<code>float</code>, <em>optional</em>, defaults to 1e-12) &#x2014;
The epsilon used by the layer normalization layers.`,name:"layer_norm_eps"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/configuration_deberta.py#L32"}}),Qe=new qe({props:{anchor:"transformers.DebertaConfig.example",$$slots:{default:[Hn]},$$scope:{ctx:x}}}),Pe=new at({props:{title:"DebertaTokenizer",local:"transformers.DebertaTokenizer",headingTag:"h2"}}),H=new V({props:{name:"class transformers.DebertaTokenizer",anchor:"transformers.DebertaTokenizer",parameters:[{name:"vocab_file",val:""},{name:"merges_file",val:""},{name:"errors",val:" = 'replace'"},{name:"bos_token",val:" = '[CLS]'"},{name:"eos_token",val:" = '[SEP]'"},{name:"sep_token",val:" = '[SEP]'"},{name:"cls_token",val:" = '[CLS]'"},{name:"unk_token",val:" = '[UNK]'"},{name:"pad_token",val:" = '[PAD]'"},{name:"mask_token",val:" = '[MASK]'"},{name:"add_prefix_space",val:" = False"},{name:"add_bos_token",val:" = False"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.DebertaTokenizer.vocab_file",description:`<strong>vocab_file</strong> (<code>str</code>) &#x2014;
Path to the vocabulary file.`,name:"vocab_file"},{anchor:"transformers.DebertaTokenizer.merges_file",description:`<strong>merges_file</strong> (<code>str</code>) &#x2014;
Path to the merges file.`,name:"merges_file"},{anchor:"transformers.DebertaTokenizer.errors",description:`<strong>errors</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;replace&quot;</code>) &#x2014;
Paradigm to follow when decoding bytes to UTF-8. See
<a href="https://docs.python.org/3/library/stdtypes.html#bytes.decode" rel="nofollow">bytes.decode</a> for more information.`,name:"errors"},{anchor:"transformers.DebertaTokenizer.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.DebertaTokenizer.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.DebertaTokenizer.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.DebertaTokenizer.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.DebertaTokenizer.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.DebertaTokenizer.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.DebertaTokenizer.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.DebertaTokenizer.add_prefix_space",description:`<strong>add_prefix_space</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</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. (Deberta tokenizer detect beginning of words by the preceding space).`,name:"add_prefix_space"},{anchor:"transformers.DebertaTokenizer.add_bos_token",description:`<strong>add_bos_token</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) &#x2014;
Whether or not to add an initial &lt;|endoftext|&gt; to the input. This allows to treat the leading word just as
any other word.`,name:"add_bos_token"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/tokenization_deberta.py#L72"}}),Me=new qe({props:{anchor:"transformers.DebertaTokenizer.example",$$slots:{default:[Nn]},$$scope:{ctx:x}}}),gt=new mt({props:{$$slots:{default:[Xn]},$$scope:{ctx:x}}}),Jt=new V({props:{name:"build_inputs_with_special_tokens",anchor:"transformers.DebertaTokenizer.build_inputs_with_special_tokens",parameters:[{name:"token_ids_0",val:": List"},{name:"token_ids_1",val:": Optional = None"}],parametersDescription:[{anchor:"transformers.DebertaTokenizer.build_inputs_with_special_tokens.token_ids_0",description:`<strong>token_ids_0</strong> (<code>List[int]</code>) &#x2014;
List of IDs to which the special tokens will be added.`,name:"token_ids_0"},{anchor:"transformers.DebertaTokenizer.build_inputs_with_special_tokens.token_ids_1",description:`<strong>token_ids_1</strong> (<code>List[int]</code>, <em>optional</em>) &#x2014;
Optional second list of IDs for sequence pairs.`,name:"token_ids_1"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/tokenization_deberta.py#L249",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>List of <a href="../glossary#input-ids">input IDs</a> with the appropriate special tokens.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><code>List[int]</code></p>
`}}),Ce=new V({props:{name:"get_special_tokens_mask",anchor:"transformers.DebertaTokenizer.get_special_tokens_mask",parameters:[{name:"token_ids_0",val:": List"},{name:"token_ids_1",val:": Optional = None"},{name:"already_has_special_tokens",val:": bool = False"}],parametersDescription:[{anchor:"transformers.DebertaTokenizer.get_special_tokens_mask.token_ids_0",description:`<strong>token_ids_0</strong> (<code>List[int]</code>) &#x2014;
List of IDs.`,name:"token_ids_0"},{anchor:"transformers.DebertaTokenizer.get_special_tokens_mask.token_ids_1",description:`<strong>token_ids_1</strong> (<code>List[int]</code>, <em>optional</em>) &#x2014;
Optional second list of IDs for sequence pairs.`,name:"token_ids_1"},{anchor:"transformers.DebertaTokenizer.get_special_tokens_mask.already_has_special_tokens",description:`<strong>already_has_special_tokens</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) &#x2014;
Whether or not the token list is already formatted with special tokens for the model.`,name:"already_has_special_tokens"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/tokenization_deberta.py#L274",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><code>List[int]</code></p>
`}}),W=new V({props:{name:"create_token_type_ids_from_sequences",anchor:"transformers.DebertaTokenizer.create_token_type_ids_from_sequences",parameters:[{name:"token_ids_0",val:": List"},{name:"token_ids_1",val:": Optional = None"}],parametersDescription:[{anchor:"transformers.DebertaTokenizer.create_token_type_ids_from_sequences.token_ids_0",description:`<strong>token_ids_0</strong> (<code>List[int]</code>) &#x2014;
List of IDs.`,name:"token_ids_0"},{anchor:"transformers.DebertaTokenizer.create_token_type_ids_from_sequences.token_ids_1",description:`<strong>token_ids_1</strong> (<code>List[int]</code>, <em>optional</em>) &#x2014;
Optional second list of IDs for sequence pairs.`,name:"token_ids_1"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/tokenization_deberta.py#L301",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>List of <a href="../glossary#token-type-ids">token type IDs</a> according to the given sequence(s).</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><code>List[int]</code></p>
`}}),ze=new qe({props:{anchor:"transformers.DebertaTokenizer.create_token_type_ids_from_sequences.example",$$slots:{default:[Sn]},$$scope:{ctx:x}}}),Gt=new V({props:{name:"save_vocabulary",anchor:"transformers.DebertaTokenizer.save_vocabulary",parameters:[{name:"save_directory",val:": str"},{name:"filename_prefix",val:": Optional = None"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/tokenization_deberta.py#L360"}}),tt=new at({props:{title:"DebertaTokenizerFast",local:"transformers.DebertaTokenizerFast",headingTag:"h2"}}),le=new V({props:{name:"class transformers.DebertaTokenizerFast",anchor:"transformers.DebertaTokenizerFast",parameters:[{name:"vocab_file",val:" = None"},{name:"merges_file",val:" = None"},{name:"tokenizer_file",val:" = None"},{name:"errors",val:" = 'replace'"},{name:"bos_token",val:" = '[CLS]'"},{name:"eos_token",val:" = '[SEP]'"},{name:"sep_token",val:" = '[SEP]'"},{name:"cls_token",val:" = '[CLS]'"},{name:"unk_token",val:" = '[UNK]'"},{name:"pad_token",val:" = '[PAD]'"},{name:"mask_token",val:" = '[MASK]'"},{name:"add_prefix_space",val:" = False"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.DebertaTokenizerFast.vocab_file",description:`<strong>vocab_file</strong> (<code>str</code>, <em>optional</em>) &#x2014;
Path to the vocabulary file.`,name:"vocab_file"},{anchor:"transformers.DebertaTokenizerFast.merges_file",description:`<strong>merges_file</strong> (<code>str</code>, <em>optional</em>) &#x2014;
Path to the merges file.`,name:"merges_file"},{anchor:"transformers.DebertaTokenizerFast.tokenizer_file",description:`<strong>tokenizer_file</strong> (<code>str</code>, <em>optional</em>) &#x2014;
The path to a tokenizer file to use instead of the vocab file.`,name:"tokenizer_file"},{anchor:"transformers.DebertaTokenizerFast.errors",description:`<strong>errors</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;replace&quot;</code>) &#x2014;
Paradigm to follow when decoding bytes to UTF-8. See
<a href="https://docs.python.org/3/library/stdtypes.html#bytes.decode" rel="nofollow">bytes.decode</a> for more information.`,name:"errors"},{anchor:"transformers.DebertaTokenizerFast.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.DebertaTokenizerFast.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.DebertaTokenizerFast.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.DebertaTokenizerFast.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.DebertaTokenizerFast.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.DebertaTokenizerFast.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.DebertaTokenizerFast.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.DebertaTokenizerFast.add_prefix_space",description:`<strong>add_prefix_space</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</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. (Deberta tokenizer detect beginning of words by the preceding space).`,name:"add_prefix_space"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/tokenization_deberta_fast.py#L33"}}),O=new qe({props:{anchor:"transformers.DebertaTokenizerFast.example",$$slots:{default:[Yn]},$$scope:{ctx:x}}}),de=new mt({props:{$$slots:{default:[Qn]},$$scope:{ctx:x}}}),ke=new V({props:{name:"build_inputs_with_special_tokens",anchor:"transformers.DebertaTokenizerFast.build_inputs_with_special_tokens",parameters:[{name:"token_ids_0",val:": List"},{name:"token_ids_1",val:": Optional = None"}],parametersDescription:[{anchor:"transformers.DebertaTokenizerFast.build_inputs_with_special_tokens.token_ids_0",description:`<strong>token_ids_0</strong> (<code>List[int]</code>) &#x2014;
List of IDs to which the special tokens will be added.`,name:"token_ids_0"},{anchor:"transformers.DebertaTokenizerFast.build_inputs_with_special_tokens.token_ids_1",description:`<strong>token_ids_1</strong> (<code>List[int]</code>, <em>optional</em>) &#x2014;
Optional second list of IDs for sequence pairs.`,name:"token_ids_1"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/tokenization_deberta_fast.py#L168",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>List of <a href="../glossary#input-ids">input IDs</a> with the appropriate special tokens.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><code>List[int]</code></p>
`}}),un=new V({props:{name:"create_token_type_ids_from_sequences",anchor:"transformers.DebertaTokenizerFast.create_token_type_ids_from_sequences",parameters:[{name:"token_ids_0",val:": List"},{name:"token_ids_1",val:": Optional = None"}],parametersDescription:[{anchor:"transformers.DebertaTokenizerFast.create_token_type_ids_from_sequences.token_ids_0",description:`<strong>token_ids_0</strong> (<code>List[int]</code>) &#x2014;
List of IDs.`,name:"token_ids_0"},{anchor:"transformers.DebertaTokenizerFast.create_token_type_ids_from_sequences.token_ids_1",description:`<strong>token_ids_1</strong> (<code>List[int]</code>, <em>optional</em>) &#x2014;
Optional second list of IDs for sequence pairs.`,name:"token_ids_1"}],source:"https://github.com/huggingface/transformers/blob/main/src/transformers/models/deberta/tokenization_deberta_fast.py#L193",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>List of <a href="../glossary#token-type-ids">token type IDs</a> according to the given sequence(s).</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><code>List[int]</code></p>
`}}),sn=new qe({props:{anchor:"transformers.DebertaTokenizerFast.create_token_type_ids_from_sequences.example",$$slots:{default:[Pn]},$$scope:{ctx:x}}}),c=new Gn({props:{pytorch:!0,tensorflow:!0,jax:!1,$$slots:{tensorflow:[Do],pytorch:[co]},$$scope:{ctx:x}}}),Ne=new Vn({props:{source:"https://github.com/huggingface/transformers/blob/main/docs/source/en/model_doc/deberta.md"}}),{c(){e=p("meta"),u=r(),n=p("p"),o=r(),T(_.$$.fragment),t=r(),T(b.$$.fragment),Y=r(),U=p("p"),U.innerHTML=z,Q=r(),C=p("p"),C.textContent=D,R=r(),h=p("p"),h.textContent=F,yt=r(),Be=p("p"),Be.innerHTML=gn,ge=r(),ut=p("p"),ut.innerHTML=Ze,Ct=r(),T(K.$$.fragment),rt=r(),pe=p("p"),pe.textContent=it,ln=r(),T(We.$$.fragment),wt=r(),ft=p("ul"),ft.innerHTML=X,je=r(),T(kt.$$.fragment),Xe=r(),Nt=p("ul"),Nt.innerHTML=bn,Se=r(),T(Xt.$$.fragment),Ot=r(),P=p("ul"),P.innerHTML=St,Yt=r(),T(be.$$.fragment),Kt=r(),A=p("ul"),A.innerHTML=Ee,en=r(),T(me.$$.fragment),tn=r(),ee=p("div"),T(Le.$$.fragment),_e=r(),Ye=p("p"),Ye.innerHTML=Mt,Dt=r(),ie=p("p"),ie.innerHTML=zt,Ut=r(),T(Qe.$$.fragment),dn=r(),T(Pe.$$.fragment),ht=r(),B=p("div"),T(H.$$.fragment),$t=r(),Ge=p("p"),Ge.textContent=It,qt=r(),vt=p("p"),vt.textContent=S,xt=r(),T(Me.$$.fragment),Rt=r(),Ae=p("p"),Ae.innerHTML=cn,Oe=r(),T(gt.$$.fragment),ue=r(),$e=p("p"),$e.innerHTML=kn,Bt=r(),Te=p("div"),T(Jt.$$.fragment),te=r(),lt=p("p"),lt.textContent=_n,dt=r(),nn=p("ul"),nn.innerHTML=Tn,ne=r(),Fe=p("div"),T(Ce.$$.fragment),Zt=r(),Ke=p("p"),Ke.innerHTML=pn,De=r(),ve=p("div"),T(W.$$.fragment),bt=r(),et=p("p"),et.textContent=Wt,ye=r(),T(ze.$$.fragment),Et=r(),_t=p("p"),_t.innerHTML=Mn,Lt=r(),Tt=p("div"),T(Gt.$$.fragment),ct=r(),T(tt.$$.fragment),we=r(),Z=p("div"),T(le.$$.fragment),nt=r(),Qt=p("p"),Qt.innerHTML=Pt,jt=r(),Vt=p("p"),Vt.textContent=Ue,mn=r(),T(O.$$.fragment),At=r(),f=p("p"),f.innerHTML=j,G=r(),T(de.$$.fragment),Ve=r(),oe=p("p"),oe.innerHTML=He,xe=r(),N=p("div"),T(ke.$$.fragment),Je=r(),se=p("p"),se.textContent=xn,jn=r(),Ft=p("ul"),Ft.innerHTML=Fn,ot=r(),Ie=p("div"),T(un.$$.fragment),yn=r(),$n=p("p"),$n.textContent=Cn,on=r(),T(sn.$$.fragment),an=r(),wn=p("p"),wn.innerHTML=fn,vn=r(),T(c.$$.fragment),J=r(),T(Ne.$$.fragment),Ht=r(),pt=p("p"),this.h()},l(s){const v=Ln("svelte-u9bgzb",document.head);e=m(v,"META",{name:!0,content:!0}),v.forEach(a),u=i(s),n=m(s,"P",{}),E(n).forEach(a),o=i(s),y(_.$$.fragment,s),t=i(s),y(b.$$.fragment,s),Y=i(s),U=m(s,"P",{"data-svelte-h":!0}),g(U)!=="svelte-1xr8kny"&&(U.innerHTML=z),Q=i(s),C=m(s,"P",{"data-svelte-h":!0}),g(C)!=="svelte-1xixyy1"&&(C.textContent=D),R=i(s),h=m(s,"P",{"data-svelte-h":!0}),g(h)!=="svelte-vfdo9a"&&(h.textContent=F),yt=i(s),Be=m(s,"P",{"data-svelte-h":!0}),g(Be)!=="svelte-1ichrxp"&&(Be.innerHTML=gn),ge=i(s),ut=m(s,"P",{"data-svelte-h":!0}),g(ut)!=="svelte-yj8rl5"&&(ut.innerHTML=Ze),Ct=i(s),y(K.$$.fragment,s),rt=i(s),pe=m(s,"P",{"data-svelte-h":!0}),g(pe)!=="svelte-1vya9w4"&&(pe.textContent=it),ln=i(s),y(We.$$.fragment,s),wt=i(s),ft=m(s,"UL",{"data-svelte-h":!0}),g(ft)!=="svelte-15sdcyc"&&(ft.innerHTML=X),je=i(s),y(kt.$$.fragment,s),Xe=i(s),Nt=m(s,"UL",{"data-svelte-h":!0}),g(Nt)!=="svelte-1mbtbmv"&&(Nt.innerHTML=bn),Se=i(s),y(Xt.$$.fragment,s),Ot=i(s),P=m(s,"UL",{"data-svelte-h":!0}),g(P)!=="svelte-1az1jb0"&&(P.innerHTML=St),Yt=i(s),y(be.$$.fragment,s),Kt=i(s),A=m(s,"UL",{"data-svelte-h":!0}),g(A)!=="svelte-1m642vv"&&(A.innerHTML=Ee),en=i(s),y(me.$$.fragment,s),tn=i(s),ee=m(s,"DIV",{class:!0});var ae=E(ee);y(Le.$$.fragment,ae),_e=i(ae),Ye=m(ae,"P",{"data-svelte-h":!0}),g(Ye)!=="svelte-1floylm"&&(Ye.innerHTML=Mt),Dt=i(ae),ie=m(ae,"P",{"data-svelte-h":!0}),g(ie)!=="svelte-o55m63"&&(ie.innerHTML=zt),Ut=i(ae),y(Qe.$$.fragment,ae),ae.forEach(a),dn=i(s),y(Pe.$$.fragment,s),ht=i(s),B=m(s,"DIV",{class:!0});var I=E(B);y(H.$$.fragment,I),$t=i(I),Ge=m(I,"P",{"data-svelte-h":!0}),g(Ge)!=="svelte-d3pp9a"&&(Ge.textContent=It),qt=i(I),vt=m(I,"P",{"data-svelte-h":!0}),g(vt)!=="svelte-1s077p3"&&(vt.textContent=S),xt=i(I),y(Me.$$.fragment,I),Rt=i(I),Ae=m(I,"P",{"data-svelte-h":!0}),g(Ae)!=="svelte-1jfcabo"&&(Ae.innerHTML=cn),Oe=i(I),y(gt.$$.fragment,I),ue=i(I),$e=m(I,"P",{"data-svelte-h":!0}),g($e)!=="svelte-1c3t5ty"&&($e.innerHTML=kn),Bt=i(I),Te=m(I,"DIV",{class:!0});var fe=E(Te);y(Jt.$$.fragment,fe),te=i(fe),lt=m(fe,"P",{"data-svelte-h":!0}),g(lt)!=="svelte-vlm5xk"&&(lt.textContent=_n),dt=i(fe),nn=m(fe,"UL",{"data-svelte-h":!0}),g(nn)!=="svelte-1196rcj"&&(nn.innerHTML=Tn),fe.forEach(a),ne=i(I),Fe=m(I,"DIV",{class:!0});var he=E(Fe);y(Ce.$$.fragment,he),Zt=i(he),Ke=m(he,"P",{"data-svelte-h":!0}),g(Ke)!=="svelte-1wmjg8a"&&(Ke.innerHTML=pn),he.forEach(a),De=i(I),ve=m(I,"DIV",{class:!0});var re=E(ve);y(W.$$.fragment,re),bt=i(re),et=m(re,"P",{"data-svelte-h":!0}),g(et)!=="svelte-1oft5by"&&(et.textContent=Wt),ye=i(re),y(ze.$$.fragment,re),Et=i(re),_t=m(re,"P",{"data-svelte-h":!0}),g(_t)!=="svelte-owoxgn"&&(_t.innerHTML=Mn),re.forEach(a),Lt=i(I),Tt=m(I,"DIV",{class:!0});var Jn=E(Tt);y(Gt.$$.fragment,Jn),Jn.forEach(a),I.forEach(a),ct=i(s),y(tt.$$.fragment,s),we=i(s),Z=m(s,"DIV",{class:!0});var ce=E(Z);y(le.$$.fragment,ce),nt=i(ce),Qt=m(ce,"P",{"data-svelte-h":!0}),g(Qt)!=="svelte-1ikh8h5"&&(Qt.innerHTML=Pt),jt=i(ce),Vt=m(ce,"P",{"data-svelte-h":!0}),g(Vt)!=="svelte-1s077p3"&&(Vt.textContent=Ue),mn=i(ce),y(O.$$.fragment,ce),At=i(ce),f=m(ce,"P",{"data-svelte-h":!0}),g(f)!=="svelte-1afeqmz"&&(f.innerHTML=j),G=i(ce),y(de.$$.fragment,ce),Ve=i(ce),oe=m(ce,"P",{"data-svelte-h":!0}),g(oe)!=="svelte-fh0aq"&&(oe.innerHTML=He),xe=i(ce),N=m(ce,"DIV",{class:!0});var rn=E(N);y(ke.$$.fragment,rn),Je=i(rn),se=m(rn,"P",{"data-svelte-h":!0}),g(se)!=="svelte-vlm5xk"&&(se.textContent=xn),jn=i(rn),Ft=m(rn,"UL",{"data-svelte-h":!0}),g(Ft)!=="svelte-1196rcj"&&(Ft.innerHTML=Fn),rn.forEach(a),ot=i(ce),Ie=m(ce,"DIV",{class:!0});var hn=E(Ie);y(un.$$.fragment,hn),yn=i(hn),$n=m(hn,"P",{"data-svelte-h":!0}),g($n)!=="svelte-1oft5by"&&($n.textContent=Cn),on=i(hn),y(sn.$$.fragment,hn),an=i(hn),wn=m(hn,"P",{"data-svelte-h":!0}),g(wn)!=="svelte-owoxgn"&&(wn.innerHTML=fn),hn.forEach(a),ce.forEach(a),vn=i(s),y(c.$$.fragment,s),J=i(s),y(Ne.$$.fragment,s),Ht=i(s),pt=m(s,"P",{}),E(pt).forEach(a),this.h()},h(){L(e,"name","hf:doc:metadata"),L(e,"content",Uo),L(ee,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(Te,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(Fe,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(ve,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(Tt,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(B,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(N,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(Ie,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),L(Z,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8")},m(s,v){l(document.head,e),d(s,u,v),d(s,n,v),d(s,o,v),w(_,s,v),d(s,t,v),w(b,s,v),d(s,Y,v),d(s,U,v),d(s,Q,v),d(s,C,v),d(s,R,v),d(s,h,v),d(s,yt,v),d(s,Be,v),d(s,ge,v),d(s,ut,v),d(s,Ct,v),w(K,s,v),d(s,rt,v),d(s,pe,v),d(s,ln,v),w(We,s,v),d(s,wt,v),d(s,ft,v),d(s,je,v),w(kt,s,v),d(s,Xe,v),d(s,Nt,v),d(s,Se,v),w(Xt,s,v),d(s,Ot,v),d(s,P,v),d(s,Yt,v),w(be,s,v),d(s,Kt,v),d(s,A,v),d(s,en,v),w(me,s,v),d(s,tn,v),d(s,ee,v),w(Le,ee,null),l(ee,_e),l(ee,Ye),l(ee,Dt),l(ee,ie),l(ee,Ut),w(Qe,ee,null),d(s,dn,v),w(Pe,s,v),d(s,ht,v),d(s,B,v),w(H,B,null),l(B,$t),l(B,Ge),l(B,qt),l(B,vt),l(B,xt),w(Me,B,null),l(B,Rt),l(B,Ae),l(B,Oe),w(gt,B,null),l(B,ue),l(B,$e),l(B,Bt),l(B,Te),w(Jt,Te,null),l(Te,te),l(Te,lt),l(Te,dt),l(Te,nn),l(B,ne),l(B,Fe),w(Ce,Fe,null),l(Fe,Zt),l(Fe,Ke),l(B,De),l(B,ve),w(W,ve,null),l(ve,bt),l(ve,et),l(ve,ye),w(ze,ve,null),l(ve,Et),l(ve,_t),l(B,Lt),l(B,Tt),w(Gt,Tt,null),d(s,ct,v),w(tt,s,v),d(s,we,v),d(s,Z,v),w(le,Z,null),l(Z,nt),l(Z,Qt),l(Z,jt),l(Z,Vt),l(Z,mn),w(O,Z,null),l(Z,At),l(Z,f),l(Z,G),w(de,Z,null),l(Z,Ve),l(Z,oe),l(Z,xe),l(Z,N),w(ke,N,null),l(N,Je),l(N,se),l(N,jn),l(N,Ft),l(Z,ot),l(Z,Ie),w(un,Ie,null),l(Ie,yn),l(Ie,$n),l(Ie,on),w(sn,Ie,null),l(Ie,an),l(Ie,wn),d(s,vn,v),w(c,s,v),d(s,J,v),w(Ne,s,v),d(s,Ht,v),d(s,pt,v),st=!0},p(s,[v]){const ae={};v&2&&(ae.$$scope={dirty:v,ctx:s}),Qe.$set(ae);const I={};v&2&&(I.$$scope={dirty:v,ctx:s}),Me.$set(I);const fe={};v&2&&(fe.$$scope={dirty:v,ctx:s}),gt.$set(fe);const he={};v&2&&(he.$$scope={dirty:v,ctx:s}),ze.$set(he);const re={};v&2&&(re.$$scope={dirty:v,ctx:s}),O.$set(re);const Jn={};v&2&&(Jn.$$scope={dirty:v,ctx:s}),de.$set(Jn);const ce={};v&2&&(ce.$$scope={dirty:v,ctx:s}),sn.$set(ce);const rn={};v&2&&(rn.$$scope={dirty:v,ctx:s}),c.$set(rn)},i(s){st||(k(_.$$.fragment,s),k(b.$$.fragment,s),k(K.$$.fragment,s),k(We.$$.fragment,s),k(kt.$$.fragment,s),k(Xt.$$.fragment,s),k(be.$$.fragment,s),k(me.$$.fragment,s),k(Le.$$.fragment,s),k(Qe.$$.fragment,s),k(Pe.$$.fragment,s),k(H.$$.fragment,s),k(Me.$$.fragment,s),k(gt.$$.fragment,s),k(Jt.$$.fragment,s),k(Ce.$$.fragment,s),k(W.$$.fragment,s),k(ze.$$.fragment,s),k(Gt.$$.fragment,s),k(tt.$$.fragment,s),k(le.$$.fragment,s),k(O.$$.fragment,s),k(de.$$.fragment,s),k(ke.$$.fragment,s),k(un.$$.fragment,s),k(sn.$$.fragment,s),k(c.$$.fragment,s),k(Ne.$$.fragment,s),st=!0)},o(s){M(_.$$.fragment,s),M(b.$$.fragment,s),M(K.$$.fragment,s),M(We.$$.fragment,s),M(kt.$$.fragment,s),M(Xt.$$.fragment,s),M(be.$$.fragment,s),M(me.$$.fragment,s),M(Le.$$.fragment,s),M(Qe.$$.fragment,s),M(Pe.$$.fragment,s),M(H.$$.fragment,s),M(Me.$$.fragment,s),M(gt.$$.fragment,s),M(Jt.$$.fragment,s),M(Ce.$$.fragment,s),M(W.$$.fragment,s),M(ze.$$.fragment,s),M(Gt.$$.fragment,s),M(tt.$$.fragment,s),M(le.$$.fragment,s),M(O.$$.fragment,s),M(de.$$.fragment,s),M(ke.$$.fragment,s),M(un.$$.fragment,s),M(sn.$$.fragment,s),M(c.$$.fragment,s),M(Ne.$$.fragment,s),st=!1},d(s){s&&(a(u),a(n),a(o),a(t),a(Y),a(U),a(Q),a(C),a(R),a(h),a(yt),a(Be),a(ge),a(ut),a(Ct),a(rt),a(pe),a(ln),a(wt),a(ft),a(je),a(Xe),a(Nt),a(Se),a(Ot),a(P),a(Yt),a(Kt),a(A),a(en),a(tn),a(ee),a(dn),a(ht),a(B),a(ct),a(we),a(Z),a(vn),a(J),a(Ht),a(pt)),a(e),$(_,s),$(b,s),$(K,s),$(We,s),$(kt,s),$(Xt,s),$(be,s),$(me,s),$(Le),$(Qe),$(Pe,s),$(H),$(Me),$(gt),$(Jt),$(Ce),$(W),$(ze),$(Gt),$(tt,s),$(le),$(O),$(de),$(ke),$(un),$(sn),$(c,s),$(Ne,s)}}}const Uo='{"title":"DeBERTa","local":"deberta","sections":[{"title":"Overview","local":"overview","sections":[],"depth":2},{"title":"Resources","local":"resources","sections":[],"depth":2},{"title":"DebertaConfig","local":"transformers.DebertaConfig","sections":[],"depth":2},{"title":"DebertaTokenizer","local":"transformers.DebertaTokenizer","sections":[],"depth":2},{"title":"DebertaTokenizerFast","local":"transformers.DebertaTokenizerFast","sections":[],"depth":2},{"title":"DebertaModel","local":"transformers.DebertaModel","sections":[],"depth":2},{"title":"DebertaPreTrainedModel","local":"transformers.DebertaPreTrainedModel","sections":[],"depth":2},{"title":"DebertaForMaskedLM","local":"transformers.DebertaForMaskedLM","sections":[],"depth":2},{"title":"DebertaForSequenceClassification","local":"transformers.DebertaForSequenceClassification","sections":[],"depth":2},{"title":"DebertaForTokenClassification","local":"transformers.DebertaForTokenClassification","sections":[],"depth":2},{"title":"DebertaForQuestionAnswering","local":"transformers.DebertaForQuestionAnswering","sections":[],"depth":2},{"title":"TFDebertaModel","local":"transformers.TFDebertaModel","sections":[],"depth":2},{"title":"TFDebertaPreTrainedModel","local":"transformers.TFDebertaPreTrainedModel","sections":[],"depth":2},{"title":"TFDebertaForMaskedLM","local":"transformers.TFDebertaForMaskedLM","sections":[],"depth":2},{"title":"TFDebertaForSequenceClassification","local":"transformers.TFDebertaForSequenceClassification","sections":[],"depth":2},{"title":"TFDebertaForTokenClassification","local":"transformers.TFDebertaForTokenClassification","sections":[],"depth":2},{"title":"TFDebertaForQuestionAnswering","local":"transformers.TFDebertaForQuestionAnswering","sections":[],"depth":2}],"depth":1}';function Io(x){return Zn(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class Ho extends Wn{constructor(e){super(),En(this,e,Io,zo,Bn,{})}}export{Ho as component};