transformers/pr_37350/en/_app/immutable/nodes/106.00a52645.js

import{s as Un,o as Fn,n as U}from"../chunks/scheduler.01eeda35.js";import{S as Wn,i as Zn,g as c,s as a,r as h,A as Vn,h as p,f as l,c as i,j as k,x as M,u as g,k as j,l as Bn,y as s,a as u,v as _,d as b,t as T,w as y}from"../chunks/index.6dd51b66.js";import{T as ct}from"../chunks/Tip.de9bae2b.js";import{D as J}from"../chunks/Docstring.cb556860.js";import{C as je}from"../chunks/CodeBlock.19ec9b8c.js";import{F as Gn,M as $n}from"../chunks/Markdown.3138439e.js";import{E as Re}from"../chunks/ExampleCodeBlock.69db56ad.js";import{H as ke,E as Nn}from"../chunks/index.58fe8f9d.js";import{H as qn,a as zn}from"../chunks/HfOption.f7f04550.js";function En(w){let e,d="Click on the CLIP models in the right sidebar for more examples of how to apply CLIP to different image and language tasks.";return{c(){e=c("p"),e.textContent=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-r81a8f"&&(e.textContent=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function Rn(w){let e,d;return e=new je({props:{code:"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",highlighted:`<span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> pipeline

clip = pipeline(
   task=<span class="hljs-string">&quot;zero-shot-image-classification&quot;</span>,
   model=<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>,
   torch_dtype=torch.bfloat16,
   device=<span class="hljs-number">0</span>
)
labels = [<span class="hljs-string">&quot;a photo of a cat&quot;</span>, <span class="hljs-string">&quot;a photo of a dog&quot;</span>, <span class="hljs-string">&quot;a photo of a car&quot;</span>]
clip(<span class="hljs-string">&quot;http://images.cocodataset.org/val2017/000000039769.jpg&quot;</span>, candidate_labels=labels)`,wrap:!1}}),{c(){h(e.$$.fragment)},l(o){g(e.$$.fragment,o)},m(o,n){_(e,o,n),d=!0},p:U,i(o){d||(b(e.$$.fragment,o),d=!0)},o(o){T(e.$$.fragment,o),d=!1},d(o){y(e,o)}}}function Xn(w){let e,d;return e=new je({props:{code:"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",highlighted:`<span class="hljs-keyword">import</span> requests
<span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image
<span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoProcessor, AutoModel

model = AutoModel.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>, torch_dtype=torch.bfloat16, attn_implementation=<span class="hljs-string">&quot;sdpa&quot;</span>)
processor = AutoProcessor.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

url = <span class="hljs-string">&quot;http://images.cocodataset.org/val2017/000000039769.jpg&quot;</span>
image = Image.<span class="hljs-built_in">open</span>(requests.get(url, stream=<span class="hljs-literal">True</span>).raw)
labels = [<span class="hljs-string">&quot;a photo of a cat&quot;</span>, <span class="hljs-string">&quot;a photo of a dog&quot;</span>, <span class="hljs-string">&quot;a photo of a car&quot;</span>]

inputs = processor(text=labels, images=image, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>, padding=<span class="hljs-literal">True</span>)

outputs = model(**inputs)
logits_per_image = outputs.logits_per_image
probs = logits_per_image.softmax(dim=<span class="hljs-number">1</span>)
most_likely_idx = probs.argmax(dim=<span class="hljs-number">1</span>).item()
most_likely_label = labels[most_likely_idx]
<span class="hljs-built_in">print</span>(<span class="hljs-string">f&quot;Most likely label: <span class="hljs-subst">{most_likely_label}</span> with probability: <span class="hljs-subst">{probs[<span class="hljs-number">0</span>][most_likely_idx].item():<span class="hljs-number">.3</span>f}</span>&quot;</span>)`,wrap:!1}}),{c(){h(e.$$.fragment)},l(o){g(e.$$.fragment,o)},m(o,n){_(e,o,n),d=!0},p:U,i(o){d||(b(e.$$.fragment,o),d=!0)},o(o){T(e.$$.fragment,o),d=!1},d(o){y(e,o)}}}function Hn(w){let e,d,o,n;return e=new zn({props:{id:"usage",option:"Pipeline",$$slots:{default:[Rn]},$$scope:{ctx:w}}}),o=new zn({props:{id:"usage",option:"AutoModel",$$slots:{default:[Xn]},$$scope:{ctx:w}}}),{c(){h(e.$$.fragment),d=a(),h(o.$$.fragment)},l(m){g(e.$$.fragment,m),d=i(m),g(o.$$.fragment,m)},m(m,t){_(e,m,t),u(m,d,t),_(o,m,t),n=!0},p(m,t){const v={};t&2&&(v.$$scope={dirty:t,ctx:m}),e.$set(v);const Me={};t&2&&(Me.$$scope={dirty:t,ctx:m}),o.$set(Me)},i(m){n||(b(e.$$.fragment,m),b(o.$$.fragment,m),n=!0)},o(m){T(e.$$.fragment,m),T(o.$$.fragment,m),n=!1},d(m){m&&l(d),y(e,m),y(o,m)}}}function Yn(w){let e,d="Example:",o,n,m;return n=new je({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> CLIPConfig, CLIPModel

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># Initializing a CLIPConfig with openai/clip-vit-base-patch32 style configuration</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>configuration = CLIPConfig()

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># Initializing a CLIPModel (with random weights) from the openai/clip-vit-base-patch32 style configuration</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>model = CLIPModel(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

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># We can also initialize a CLIPConfig from a CLIPTextConfig and a CLIPVisionConfig</span>
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> CLIPTextConfig, CLIPVisionConfig

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># Initializing a CLIPText and CLIPVision configuration</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>config_text = CLIPTextConfig()
<span class="hljs-meta">&gt;&gt;&gt; </span>config_vision = CLIPVisionConfig()

<span class="hljs-meta">&gt;&gt;&gt; </span>config = CLIPConfig.from_text_vision_configs(config_text, config_vision)`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-11lpom8"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function Sn(w){let e,d="Example:",o,n,m;return n=new je({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> CLIPTextConfig, CLIPTextModel

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># Initializing a CLIPTextConfig with openai/clip-vit-base-patch32 style configuration</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>configuration = CLIPTextConfig()

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># Initializing a CLIPTextModel (with random weights) from the openai/clip-vit-base-patch32 style configuration</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>model = CLIPTextModel(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=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-11lpom8"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function Dn(w){let e,d="Example:",o,n,m;return n=new je({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> CLIPVisionConfig, CLIPVisionModel

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># Initializing a CLIPVisionConfig with openai/clip-vit-base-patch32 style configuration</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>configuration = CLIPVisionConfig()

<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-comment"># Initializing a CLIPVisionModel (with random weights) from the openai/clip-vit-base-patch32 style configuration</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>model = CLIPVisionModel(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=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-11lpom8"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function Qn(w){let e,d=`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=c("p"),e.innerHTML=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-fincs2"&&(e.innerHTML=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function An(w){let e,d="Examples:",o,n,m;return n=new je({props:{code:"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",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> requests
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoProcessor, CLIPModel

<span class="hljs-meta">&gt;&gt;&gt; </span>model = CLIPModel.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>processor = AutoProcessor.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>url = <span class="hljs-string">&quot;http://images.cocodataset.org/val2017/000000039769.jpg&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>image = Image.<span class="hljs-built_in">open</span>(requests.get(url, stream=<span class="hljs-literal">True</span>).raw)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = processor(
<span class="hljs-meta">... </span>    text=[<span class="hljs-string">&quot;a photo of a cat&quot;</span>, <span class="hljs-string">&quot;a photo of a dog&quot;</span>], images=image, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>, padding=<span class="hljs-literal">True</span>
<span class="hljs-meta">... </span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>outputs = model(**inputs)
<span class="hljs-meta">&gt;&gt;&gt; </span>logits_per_image = outputs.logits_per_image  <span class="hljs-comment"># this is the image-text similarity score</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>probs = logits_per_image.softmax(dim=<span class="hljs-number">1</span>)  <span class="hljs-comment"># we can take the softmax to get the label probabilities</span>`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-kvfsh7"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function On(w){let e,d=`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=c("p"),e.innerHTML=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-fincs2"&&(e.innerHTML=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function Kn(w){let e,d="Examples:",o,n,m;return n=new je({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, CLIPModel

<span class="hljs-meta">&gt;&gt;&gt; </span>model = CLIPModel.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = tokenizer([<span class="hljs-string">&quot;a photo of a cat&quot;</span>, <span class="hljs-string">&quot;a photo of a dog&quot;</span>], padding=<span class="hljs-literal">True</span>, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>text_features = model.get_text_features(**inputs)`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-kvfsh7"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function es(w){let e,d=`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=c("p"),e.innerHTML=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-fincs2"&&(e.innerHTML=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function ts(w){let e,d="Examples:",o,n,m;return n=new je({props:{code:"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",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> requests
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoProcessor, CLIPModel

<span class="hljs-meta">&gt;&gt;&gt; </span>model = CLIPModel.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>processor = AutoProcessor.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>url = <span class="hljs-string">&quot;http://images.cocodataset.org/val2017/000000039769.jpg&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>image = Image.<span class="hljs-built_in">open</span>(requests.get(url, stream=<span class="hljs-literal">True</span>).raw)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = processor(images=image, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>image_features = model.get_image_features(**inputs)`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-kvfsh7"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function os(w){let e,d=`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=c("p"),e.innerHTML=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-fincs2"&&(e.innerHTML=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function ns(w){let e,d="Examples:",o,n,m;return n=new je({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, CLIPTextModel

<span class="hljs-meta">&gt;&gt;&gt; </span>model = CLIPTextModel.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = tokenizer([<span class="hljs-string">&quot;a photo of a cat&quot;</span>, <span class="hljs-string">&quot;a photo of a dog&quot;</span>], padding=<span class="hljs-literal">True</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_state = outputs.last_hidden_state
<span class="hljs-meta">&gt;&gt;&gt; </span>pooled_output = outputs.pooler_output  <span class="hljs-comment"># pooled (EOS token) states</span>`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-kvfsh7"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function ss(w){let e,d=`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=c("p"),e.innerHTML=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-fincs2"&&(e.innerHTML=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function rs(w){let e,d="Examples:",o,n,m;return n=new je({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, CLIPTextModelWithProjection

<span class="hljs-meta">&gt;&gt;&gt; </span>model = CLIPTextModelWithProjection.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = tokenizer([<span class="hljs-string">&quot;a photo of a cat&quot;</span>, <span class="hljs-string">&quot;a photo of a dog&quot;</span>], padding=<span class="hljs-literal">True</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>text_embeds = outputs.text_embeds`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-kvfsh7"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function as(w){let e,d=`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=c("p"),e.innerHTML=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-fincs2"&&(e.innerHTML=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function is(w){let e,d="Examples:",o,n,m;return n=new je({props:{code:"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",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> requests
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoProcessor, CLIPVisionModelWithProjection

<span class="hljs-meta">&gt;&gt;&gt; </span>model = CLIPVisionModelWithProjection.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>processor = AutoProcessor.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>url = <span class="hljs-string">&quot;http://images.cocodataset.org/val2017/000000039769.jpg&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>image = Image.<span class="hljs-built_in">open</span>(requests.get(url, stream=<span class="hljs-literal">True</span>).raw)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = processor(images=image, 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>image_embeds = outputs.image_embeds`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-kvfsh7"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function ls(w){let e,d=`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=c("p"),e.innerHTML=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-fincs2"&&(e.innerHTML=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function ds(w){let e,d="Examples:",o,n,m;return n=new je({props:{code:"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",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> requests
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoProcessor, CLIPVisionModel

<span class="hljs-meta">&gt;&gt;&gt; </span>model = CLIPVisionModel.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>processor = AutoProcessor.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>url = <span class="hljs-string">&quot;http://images.cocodataset.org/val2017/000000039769.jpg&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>image = Image.<span class="hljs-built_in">open</span>(requests.get(url, stream=<span class="hljs-literal">True</span>).raw)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = processor(images=image, 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_state = outputs.last_hidden_state
<span class="hljs-meta">&gt;&gt;&gt; </span>pooled_output = outputs.pooler_output  <span class="hljs-comment"># pooled CLS states</span>`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-kvfsh7"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function cs(w){let e,d=`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=c("p"),e.innerHTML=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-fincs2"&&(e.innerHTML=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function ps(w){let e,d="Example:",o,n,m;return n=new je({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> AutoImageProcessor, CLIPForImageClassification
<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> datasets <span class="hljs-keyword">import</span> load_dataset

<span class="hljs-meta">&gt;&gt;&gt; </span>dataset = load_dataset(<span class="hljs-string">&quot;huggingface/cats-image&quot;</span>, trust_remote_code=<span class="hljs-literal">True</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>image = dataset[<span class="hljs-string">&quot;test&quot;</span>][<span class="hljs-string">&quot;image&quot;</span>][<span class="hljs-number">0</span>]

<span class="hljs-meta">&gt;&gt;&gt; </span>image_processor = AutoImageProcessor.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = CLIPForImageClassification.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = image_processor(image, 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"># model predicts one of the 1000 ImageNet classes</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>predicted_label = logits.argmax(-<span class="hljs-number">1</span>).item()
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-built_in">print</span>(model.config.id2label[predicted_label])
LABEL_0`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-11lpom8"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function ms(w){let e,d,o,n,m,t,v=`This model inherits from <a href="/docs/transformers/pr_37350/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
etc.)`,Me,X,$e=`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.`,bt,W,H,me,z,ae='The <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPModel">CLIPModel</a> forward method, overrides the <code>__call__</code> special method.',tt,G,pt,fe,Je,K,ze,Y,ue,mt='The <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPModel">CLIPModel</a> forward method, overrides the <code>__call__</code> special method.',ot,ie,De,Z,xe,ee,de,ce,Be,Ge='The <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPModel">CLIPModel</a> forward method, overrides the <code>__call__</code> special method.',ft,Ue,Xe,N,Ie,le,he,V,nt,ve,Ce,wt=`The text model from CLIP without any head or projection on top.
This model inherits from <a href="/docs/transformers/pr_37350/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
etc.)`,F,ge,st=`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.`,Ne,_e,Qe,te,He,rt='The <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPTextModel">CLIPTextModel</a> forward method, overrides the <code>__call__</code> special method.',S,oe,Ae,D,ne,Fe,Q,B,We,at,Pe,Vt="CLIP Text Model with a projection layer on top (a linear layer on top of the pooled output).",be,pe,Oe=`This model inherits from <a href="/docs/transformers/pr_37350/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
etc.)`,se,C,P=`This model is also a PyTorch <a href="https://pytorch.org/docs/stable/nn.html#torch.nn.Module" rel="nofollow">torch.nn.Module</a> subclass.
Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
and behavior.`,A,q,O,Ze,Te,ye='The <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPTextModelWithProjection">CLIPTextModelWithProjection</a> forward method, overrides the <code>__call__</code> special method.',Ve,I,L,E,it,qe,$t,R,Ee,ut,ht,Ye="CLIP Vision Model with a projection layer on top (a linear layer on top of the pooled output).",To,io,Ot=`This model inherits from <a href="/docs/transformers/pr_37350/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
etc.)`,Kt,lo,yo=`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.`,Xo,yt,Mt,eo,co,Mo='The <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPVisionModelWithProjection">CLIPVisionModelWithProjection</a> forward method, overrides the <code>__call__</code> special method.',Ho,Dt,po,It,Io,xt,vo,re,Pt,Jo,Rt,on=`The vision model from CLIP without any head or projection on top.
This model inherits from <a href="/docs/transformers/pr_37350/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
etc.)`,zo,Xt,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.`,Uo,we,Lt,Fo,Ht,sn='The <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPVisionModel">CLIPVisionModel</a> forward method, overrides the <code>__call__</code> special method.',Wo,kt,Yo,Qt,to,mo,wo,Le,jt,Zo,Yt,rn=`CLIP vision encoder with an image classification head on top (a linear layer on top of the pooled final hidden states of
the patch tokens) e.g. for ImageNet.`,$o,Bt,Vo=`This model inherits from <a href="/docs/transformers/pr_37350/en/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a>. Check the superclass documentation for the generic methods the
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
etc.)`,Jt,Gt,So=`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.`,fo,zt,oo,At,Nt,Do='The <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPForImageClassification">CLIPForImageClassification</a> forward method, overrides the <code>__call__</code> special method.',uo,no,xo,vt,Co;return e=new ke({props:{title:"CLIPModel",local:"transformers.CLIPModel",headingTag:"h2"}}),n=new J({props:{name:"class transformers.CLIPModel",anchor:"transformers.CLIPModel",parameters:[{name:"config",val:": CLIPConfig"}],parametersDescription:[{anchor:"transformers.CLIPModel.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPConfig">CLIPConfig</a>) &#x2014; Model configuration class with all the parameters of the model.
Initializing with a config file does not load the weights associated with the model, only the
configuration. Check out the <a href="/docs/transformers/pr_37350/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/vr_37350/src/transformers/models/clip/modeling_clip.py#L1171"}}),H=new J({props:{name:"forward",anchor:"transformers.CLIPModel.forward",parameters:[{name:"input_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"pixel_values",val:": typing.Optional[torch.FloatTensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"position_ids",val:": typing.Optional[torch.LongTensor] = None"},{name:"return_loss",val:": typing.Optional[bool] = None"},{name:"output_attentions",val:": typing.Optional[bool] = None"},{name:"output_hidden_states",val:": typing.Optional[bool] = None"},{name:"interpolate_pos_encoding",val:": bool = False"}],parametersDescription:[{anchor:"transformers.CLIPModel.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. Padding will be ignored by default should you provide
it.</p>
<p>Indices can be obtained using <a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_37350/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/pr_37350/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.CLIPModel.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p>
<ul>
<li>1 for tokens that are <strong>not masked</strong>,</li>
<li>0 for tokens that are <strong>masked</strong>.</li>
</ul>
<p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.CLIPModel.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.CLIPModel.forward.pixel_values",description:`<strong>pixel_values</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, num_channels, height, width)</code>) &#x2014;
Pixel values. Padding will be ignored by default should you provide it. Pixel values can be obtained using
<a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoImageProcessor">AutoImageProcessor</a>. See <a href="/docs/transformers/pr_37350/en/model_doc/vilt#transformers.ViltFeatureExtractor.__call__">CLIPImageProcessor.<strong>call</strong>()</a> for details.`,name:"pixel_values"},{anchor:"transformers.CLIPModel.forward.return_loss",description:`<strong>return_loss</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return the contrastive loss.`,name:"return_loss"},{anchor:"transformers.CLIPModel.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.CLIPModel.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.CLIPModel.forward.interpolate_pos_encoding",description:`<strong>interpolate_pos_encoding</strong> (<code>bool</code>, <em>optional</em>, defaults <code>False</code>) &#x2014;
Whether to interpolate the pre-trained position encodings.`,name:"interpolate_pos_encoding"},{anchor:"transformers.CLIPModel.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_37350/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_clip.py#L1303",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <code>transformers.models.clip.modeling_clip.CLIPOutput</code> 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 (<code>&lt;class 'transformers.models.clip.configuration_clip.CLIPConfig'&gt;</code>) and inputs.</p>
<ul>
<li><strong>loss</strong> (<code>torch.FloatTensor</code> of shape <code>(1,)</code>, <em>optional</em>, returned when <code>return_loss</code> is <code>True</code>) — Contrastive loss for image-text similarity.</li>
<li><strong>logits_per_image</strong> (<code>torch.FloatTensor</code> of shape <code>(image_batch_size, text_batch_size)</code>) — The scaled dot product scores between <code>image_embeds</code> and <code>text_embeds</code>. This represents the image-text
similarity scores.</li>
<li><strong>logits_per_text</strong> (<code>torch.FloatTensor</code> of shape <code>(text_batch_size, image_batch_size)</code>) — The scaled dot product scores between <code>text_embeds</code> and <code>image_embeds</code>. This represents the text-image
similarity scores.</li>
<li><strong>text_embeds</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, output_dim</code>) — The text embeddings obtained by applying the projection layer to the pooled output of <a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPTextModel"
>CLIPTextModel</a>.</li>
<li><strong>image_embeds</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, output_dim</code>) — The image embeddings obtained by applying the projection layer to the pooled output of <a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPVisionModel"
>CLIPVisionModel</a>.</li>
<li><strong>text_model_output</strong> (<code>BaseModelOutputWithPooling</code>) — The output of the <a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPTextModel"
>CLIPTextModel</a>.</li>
<li><strong>vision_model_output</strong> (<code>BaseModelOutputWithPooling</code>) — The output of the <a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPVisionModel"
>CLIPVisionModel</a>.</li>
</ul>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><code>transformers.models.clip.modeling_clip.CLIPOutput</code> or <code>tuple(torch.FloatTensor)</code></p>
`}}),G=new ct({props:{$$slots:{default:[Qn]},$$scope:{ctx:w}}}),fe=new Re({props:{anchor:"transformers.CLIPModel.forward.example",$$slots:{default:[An]},$$scope:{ctx:w}}}),ze=new J({props:{name:"get_text_features",anchor:"transformers.CLIPModel.get_text_features",parameters:[{name:"input_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"position_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": typing.Optional[bool] = None"},{name:"output_hidden_states",val:": typing.Optional[bool] = None"}],parametersDescription:[{anchor:"transformers.CLIPModel.get_text_features.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. Padding will be ignored by default should you provide
it.</p>
<p>Indices can be obtained using <a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_37350/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/pr_37350/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.CLIPModel.get_text_features.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p>
<ul>
<li>1 for tokens that are <strong>not masked</strong>,</li>
<li>0 for tokens that are <strong>masked</strong>.</li>
</ul>
<p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.CLIPModel.get_text_features.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.CLIPModel.get_text_features.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.CLIPModel.get_text_features.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.CLIPModel.get_text_features.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_37350/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_clip.py#L1211",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>The text embeddings obtained by
applying the projection layer to the pooled output of <a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPTextModel"
>CLIPTextModel</a>.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p>text_features (<code>torch.FloatTensor</code> of shape <code>(batch_size, output_dim</code>)</p>
`}}),ie=new ct({props:{$$slots:{default:[On]},$$scope:{ctx:w}}}),Z=new Re({props:{anchor:"transformers.CLIPModel.get_text_features.example",$$slots:{default:[Kn]},$$scope:{ctx:w}}}),de=new J({props:{name:"get_image_features",anchor:"transformers.CLIPModel.get_image_features",parameters:[{name:"pixel_values",val:": typing.Optional[torch.FloatTensor] = None"},{name:"output_attentions",val:": typing.Optional[bool] = None"},{name:"output_hidden_states",val:": typing.Optional[bool] = None"},{name:"interpolate_pos_encoding",val:": bool = False"}],parametersDescription:[{anchor:"transformers.CLIPModel.get_image_features.pixel_values",description:`<strong>pixel_values</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, num_channels, height, width)</code>) &#x2014;
Pixel values. Padding will be ignored by default should you provide it. Pixel values can be obtained using
<a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoImageProcessor">AutoImageProcessor</a>. See <a href="/docs/transformers/pr_37350/en/model_doc/vilt#transformers.ViltFeatureExtractor.__call__">CLIPImageProcessor.<strong>call</strong>()</a> for details.`,name:"pixel_values"},{anchor:"transformers.CLIPModel.get_image_features.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.CLIPModel.get_image_features.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.CLIPModel.get_image_features.interpolate_pos_encoding",description:`<strong>interpolate_pos_encoding</strong> (<code>bool</code>, <em>optional</em>, defaults <code>False</code>) &#x2014;
Whether to interpolate the pre-trained position encodings.`,name:"interpolate_pos_encoding"},{anchor:"transformers.CLIPModel.get_image_features.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_37350/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_clip.py#L1255",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>The image embeddings obtained by
applying the projection layer to the pooled output of <a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPVisionModel"
>CLIPVisionModel</a>.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p>image_features (<code>torch.FloatTensor</code> of shape <code>(batch_size, output_dim</code>)</p>
`}}),Ue=new ct({props:{$$slots:{default:[es]},$$scope:{ctx:w}}}),N=new Re({props:{anchor:"transformers.CLIPModel.get_image_features.example",$$slots:{default:[ts]},$$scope:{ctx:w}}}),le=new ke({props:{title:"CLIPTextModel",local:"transformers.CLIPTextModel",headingTag:"h2"}}),nt=new J({props:{name:"class transformers.CLIPTextModel",anchor:"transformers.CLIPTextModel",parameters:[{name:"config",val:": CLIPTextConfig"}],parametersDescription:[{anchor:"transformers.CLIPTextModel.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPConfig">CLIPConfig</a>) &#x2014; Model configuration class with all the parameters of the model.
Initializing with a config file does not load the weights associated with the model, only the
configuration. Check out the <a href="/docs/transformers/pr_37350/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/vr_37350/src/transformers/models/clip/modeling_clip.py#L999"}}),Qe=new J({props:{name:"forward",anchor:"transformers.CLIPTextModel.forward",parameters:[{name:"input_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"position_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": typing.Optional[bool] = None"},{name:"output_hidden_states",val:": typing.Optional[bool] = None"}],parametersDescription:[{anchor:"transformers.CLIPTextModel.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. Padding will be ignored by default should you provide
it.</p>
<p>Indices can be obtained using <a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_37350/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/pr_37350/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.CLIPTextModel.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p>
<ul>
<li>1 for tokens that are <strong>not masked</strong>,</li>
<li>0 for tokens that are <strong>masked</strong>.</li>
</ul>
<p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.CLIPTextModel.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.CLIPTextModel.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.CLIPTextModel.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.CLIPTextModel.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_37350/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_clip.py#L1020",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/pr_37350/en/main_classes/output#transformers.modeling_outputs.BaseModelOutputWithPooling"
>transformers.modeling_outputs.BaseModelOutputWithPooling</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 (<code>&lt;class 'transformers.models.clip.configuration_clip.CLIPTextConfig'&gt;</code>) 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>pooler_output</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, hidden_size)</code>) — Last layer hidden-state of the first token of the sequence (classification token) after further processing
through the layers used for the auxiliary pretraining task. E.g. for BERT-family of models, this returns
the classification token after processing through a linear layer and a tanh activation function. The linear
layer weights are trained from the next sentence prediction (classification) objective during pretraining.</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/pr_37350/en/main_classes/output#transformers.modeling_outputs.BaseModelOutputWithPooling"
>transformers.modeling_outputs.BaseModelOutputWithPooling</a> or <code>tuple(torch.FloatTensor)</code></p>
`}}),oe=new ct({props:{$$slots:{default:[os]},$$scope:{ctx:w}}}),D=new Re({props:{anchor:"transformers.CLIPTextModel.forward.example",$$slots:{default:[ns]},$$scope:{ctx:w}}}),Fe=new ke({props:{title:"CLIPTextModelWithProjection",local:"transformers.CLIPTextModelWithProjection",headingTag:"h2"}}),We=new J({props:{name:"class transformers.CLIPTextModelWithProjection",anchor:"transformers.CLIPTextModelWithProjection",parameters:[{name:"config",val:": CLIPTextConfig"}],parametersDescription:[{anchor:"transformers.CLIPTextModelWithProjection.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPConfig">CLIPConfig</a>) &#x2014; Model configuration class with all the parameters of the model.
Initializing with a config file does not load the weights associated with the model, only the
configuration. Check out the <a href="/docs/transformers/pr_37350/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/vr_37350/src/transformers/models/clip/modeling_clip.py#L1393"}}),O=new J({props:{name:"forward",anchor:"transformers.CLIPTextModelWithProjection.forward",parameters:[{name:"input_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"position_ids",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": typing.Optional[bool] = None"},{name:"output_hidden_states",val:": typing.Optional[bool] = None"}],parametersDescription:[{anchor:"transformers.CLIPTextModelWithProjection.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. Padding will be ignored by default should you provide
it.</p>
<p>Indices can be obtained using <a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_37350/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/pr_37350/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.CLIPTextModelWithProjection.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p>
<ul>
<li>1 for tokens that are <strong>not masked</strong>,</li>
<li>0 for tokens that are <strong>masked</strong>.</li>
</ul>
<p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.CLIPTextModelWithProjection.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.CLIPTextModelWithProjection.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.CLIPTextModelWithProjection.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.CLIPTextModelWithProjection.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_37350/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_clip.py#L1421",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <code>transformers.models.clip.modeling_clip.CLIPTextModelOutput</code> 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 (<code>&lt;class 'transformers.models.clip.configuration_clip.CLIPTextConfig'&gt;</code>) and inputs.</p>
<ul>
<li>
<p><strong>text_embeds</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, output_dim)</code> <em>optional</em> returned when model is initialized with <code>with_projection=True</code>) — The text embeddings obtained by applying the projection layer to the pooler_output.</p>
</li>
<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><code>transformers.models.clip.modeling_clip.CLIPTextModelOutput</code> or <code>tuple(torch.FloatTensor)</code></p>
`}}),I=new ct({props:{$$slots:{default:[ss]},$$scope:{ctx:w}}}),E=new Re({props:{anchor:"transformers.CLIPTextModelWithProjection.forward.example",$$slots:{default:[rs]},$$scope:{ctx:w}}}),qe=new ke({props:{title:"CLIPVisionModelWithProjection",local:"transformers.CLIPVisionModelWithProjection",headingTag:"h2"}}),Ee=new J({props:{name:"class transformers.CLIPVisionModelWithProjection",anchor:"transformers.CLIPVisionModelWithProjection",parameters:[{name:"config",val:": CLIPVisionConfig"}],parametersDescription:[{anchor:"transformers.CLIPVisionModelWithProjection.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPConfig">CLIPConfig</a>) &#x2014; Model configuration class with all the parameters of the model.
Initializing with a config file does not load the weights associated with the model, only the
configuration. Check out the <a href="/docs/transformers/pr_37350/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/vr_37350/src/transformers/models/clip/modeling_clip.py#L1467"}}),Mt=new J({props:{name:"forward",anchor:"transformers.CLIPVisionModelWithProjection.forward",parameters:[{name:"pixel_values",val:": typing.Optional[torch.FloatTensor] = None"},{name:"output_attentions",val:": typing.Optional[bool] = None"},{name:"output_hidden_states",val:": typing.Optional[bool] = None"},{name:"interpolate_pos_encoding",val:": bool = False"}],parametersDescription:[{anchor:"transformers.CLIPVisionModelWithProjection.forward.pixel_values",description:`<strong>pixel_values</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, num_channels, height, width)</code>) &#x2014;
Pixel values. Padding will be ignored by default should you provide it. Pixel values can be obtained using
<a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoImageProcessor">AutoImageProcessor</a>. See <a href="/docs/transformers/pr_37350/en/model_doc/vilt#transformers.ViltFeatureExtractor.__call__">CLIPImageProcessor.<strong>call</strong>()</a> for details.`,name:"pixel_values"},{anchor:"transformers.CLIPVisionModelWithProjection.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.CLIPVisionModelWithProjection.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.CLIPVisionModelWithProjection.forward.interpolate_pos_encoding",description:`<strong>interpolate_pos_encoding</strong> (<code>bool</code>, <em>optional</em>, defaults <code>False</code>) &#x2014;
Whether to interpolate the pre-trained position encodings.`,name:"interpolate_pos_encoding"},{anchor:"transformers.CLIPVisionModelWithProjection.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_37350/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_clip.py#L1491",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <code>transformers.models.clip.modeling_clip.CLIPVisionModelOutput</code> 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 (<code>&lt;class 'transformers.models.clip.configuration_clip.CLIPVisionConfig'&gt;</code>) and inputs.</p>
<ul>
<li>
<p><strong>image_embeds</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, output_dim)</code> <em>optional</em> returned when model is initialized with <code>with_projection=True</code>) — The image embeddings obtained by applying the projection layer to the pooler_output.</p>
</li>
<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><code>transformers.models.clip.modeling_clip.CLIPVisionModelOutput</code> or <code>tuple(torch.FloatTensor)</code></p>
`}}),Dt=new ct({props:{$$slots:{default:[as]},$$scope:{ctx:w}}}),It=new Re({props:{anchor:"transformers.CLIPVisionModelWithProjection.forward.example",$$slots:{default:[is]},$$scope:{ctx:w}}}),xt=new ke({props:{title:"CLIPVisionModel",local:"transformers.CLIPVisionModel",headingTag:"h2"}}),Pt=new J({props:{name:"class transformers.CLIPVisionModel",anchor:"transformers.CLIPVisionModel",parameters:[{name:"config",val:": CLIPVisionConfig"}],parametersDescription:[{anchor:"transformers.CLIPVisionModel.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPConfig">CLIPConfig</a>) &#x2014; Model configuration class with all the parameters of the model.
Initializing with a config file does not load the weights associated with the model, only the
configuration. Check out the <a href="/docs/transformers/pr_37350/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/vr_37350/src/transformers/models/clip/modeling_clip.py#L1112"}}),Lt=new J({props:{name:"forward",anchor:"transformers.CLIPVisionModel.forward",parameters:[{name:"pixel_values",val:": typing.Optional[torch.FloatTensor] = None"},{name:"output_attentions",val:": typing.Optional[bool] = None"},{name:"output_hidden_states",val:": typing.Optional[bool] = None"},{name:"interpolate_pos_encoding",val:": bool = False"}],parametersDescription:[{anchor:"transformers.CLIPVisionModel.forward.pixel_values",description:`<strong>pixel_values</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, num_channels, height, width)</code>) &#x2014;
Pixel values. Padding will be ignored by default should you provide it. Pixel values can be obtained using
<a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoImageProcessor">AutoImageProcessor</a>. See <a href="/docs/transformers/pr_37350/en/model_doc/vilt#transformers.ViltFeatureExtractor.__call__">CLIPImageProcessor.<strong>call</strong>()</a> for details.`,name:"pixel_values"},{anchor:"transformers.CLIPVisionModel.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.CLIPVisionModel.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.CLIPVisionModel.forward.interpolate_pos_encoding",description:`<strong>interpolate_pos_encoding</strong> (<code>bool</code>, <em>optional</em>, defaults <code>False</code>) &#x2014;
Whether to interpolate the pre-trained position encodings.`,name:"interpolate_pos_encoding"},{anchor:"transformers.CLIPVisionModel.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_37350/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_clip.py#L1130",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/pr_37350/en/main_classes/output#transformers.modeling_outputs.BaseModelOutputWithPooling"
>transformers.modeling_outputs.BaseModelOutputWithPooling</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 (<code>&lt;class 'transformers.models.clip.configuration_clip.CLIPVisionConfig'&gt;</code>) 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>pooler_output</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, hidden_size)</code>) — Last layer hidden-state of the first token of the sequence (classification token) after further processing
through the layers used for the auxiliary pretraining task. E.g. for BERT-family of models, this returns
the classification token after processing through a linear layer and a tanh activation function. The linear
layer weights are trained from the next sentence prediction (classification) objective during pretraining.</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/pr_37350/en/main_classes/output#transformers.modeling_outputs.BaseModelOutputWithPooling"
>transformers.modeling_outputs.BaseModelOutputWithPooling</a> or <code>tuple(torch.FloatTensor)</code></p>
`}}),kt=new ct({props:{$$slots:{default:[ls]},$$scope:{ctx:w}}}),Qt=new Re({props:{anchor:"transformers.CLIPVisionModel.forward.example",$$slots:{default:[ds]},$$scope:{ctx:w}}}),mo=new ke({props:{title:"CLIPForImageClassification",local:"transformers.CLIPForImageClassification",headingTag:"h2"}}),jt=new J({props:{name:"class transformers.CLIPForImageClassification",anchor:"transformers.CLIPForImageClassification",parameters:[{name:"config",val:": CLIPConfig"}],parametersDescription:[{anchor:"transformers.CLIPForImageClassification.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPConfig">CLIPConfig</a>) &#x2014; Model configuration class with all the parameters of the model.
Initializing with a config file does not load the weights associated with the model, only the
configuration. Check out the <a href="/docs/transformers/pr_37350/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/vr_37350/src/transformers/models/clip/modeling_clip.py#L1540"}}),oo=new J({props:{name:"forward",anchor:"transformers.CLIPForImageClassification.forward",parameters:[{name:"pixel_values",val:": typing.Optional[torch.Tensor] = None"},{name:"labels",val:": typing.Optional[torch.Tensor] = None"},{name:"output_attentions",val:": typing.Optional[bool] = None"},{name:"output_hidden_states",val:": typing.Optional[bool] = None"}],parametersDescription:[{anchor:"transformers.CLIPForImageClassification.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. Padding will be ignored by default should you provide
it.</p>
<p>Indices can be obtained using <a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_37350/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/pr_37350/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.CLIPForImageClassification.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) &#x2014;
Mask to avoid performing attention on padding token indices. Mask values selected in <code>[0, 1]</code>:</p>
<ul>
<li>1 for tokens that are <strong>not masked</strong>,</li>
<li>0 for tokens that are <strong>masked</strong>.</li>
</ul>
<p><a href="../glossary#attention-mask">What are attention masks?</a>`,name:"attention_mask"},{anchor:"transformers.CLIPForImageClassification.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.CLIPForImageClassification.forward.pixel_values",description:`<strong>pixel_values</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, num_channels, height, width)</code>) &#x2014;
Pixel values. Padding will be ignored by default should you provide it. Pixel values can be obtained using
<a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoImageProcessor">AutoImageProcessor</a>. See <a href="/docs/transformers/pr_37350/en/model_doc/vilt#transformers.ViltFeatureExtractor.__call__">CLIPImageProcessor.<strong>call</strong>()</a> for details.`,name:"pixel_values"},{anchor:"transformers.CLIPForImageClassification.forward.return_loss",description:`<strong>return_loss</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return the contrastive loss.`,name:"return_loss"},{anchor:"transformers.CLIPForImageClassification.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.CLIPForImageClassification.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.CLIPForImageClassification.forward.interpolate_pos_encoding",description:`<strong>interpolate_pos_encoding</strong> (<code>bool</code>, <em>optional</em>, defaults <code>False</code>) &#x2014;
Whether to interpolate the pre-trained position encodings.`,name:"interpolate_pos_encoding"},{anchor:"transformers.CLIPForImageClassification.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_37350/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"},{anchor:"transformers.CLIPForImageClassification.forward.labels",description:`<strong>labels</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size,)</code>, <em>optional</em>) &#x2014;
Labels for computing the image 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/vr_37350/src/transformers/models/clip/modeling_clip.py#L1565",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/pr_37350/en/main_classes/output#transformers.modeling_outputs.ImageClassifierOutput"
>transformers.modeling_outputs.ImageClassifierOutput</a> or a tuple of
<code>torch.FloatTensor</code> (if <code>return_dict=False</code> is passed or when <code>config.return_dict=False</code>) comprising various
elements depending on the configuration (<a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPConfig"
>CLIPConfig</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 stage) of shape <code>(batch_size, sequence_length, hidden_size)</code>. Hidden-states
(also called feature maps) of the model at the output of each stage.</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, patch_size, 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/pr_37350/en/main_classes/output#transformers.modeling_outputs.ImageClassifierOutput"
>transformers.modeling_outputs.ImageClassifierOutput</a> or <code>tuple(torch.FloatTensor)</code></p>
`}}),no=new ct({props:{$$slots:{default:[cs]},$$scope:{ctx:w}}}),vt=new Re({props:{anchor:"transformers.CLIPForImageClassification.forward.example",$$slots:{default:[ps]},$$scope:{ctx:w}}}),{c(){h(e.$$.fragment),d=a(),o=c("div"),h(n.$$.fragment),m=a(),t=c("p"),t.innerHTML=v,Me=a(),X=c("p"),X.innerHTML=$e,bt=a(),W=c("div"),h(H.$$.fragment),me=a(),z=c("p"),z.innerHTML=ae,tt=a(),h(G.$$.fragment),pt=a(),h(fe.$$.fragment),Je=a(),K=c("div"),h(ze.$$.fragment),Y=a(),ue=c("p"),ue.innerHTML=mt,ot=a(),h(ie.$$.fragment),De=a(),h(Z.$$.fragment),xe=a(),ee=c("div"),h(de.$$.fragment),ce=a(),Be=c("p"),Be.innerHTML=Ge,ft=a(),h(Ue.$$.fragment),Xe=a(),h(N.$$.fragment),Ie=a(),h(le.$$.fragment),he=a(),V=c("div"),h(nt.$$.fragment),ve=a(),Ce=c("p"),Ce.innerHTML=wt,F=a(),ge=c("p"),ge.innerHTML=st,Ne=a(),_e=c("div"),h(Qe.$$.fragment),te=a(),He=c("p"),He.innerHTML=rt,S=a(),h(oe.$$.fragment),Ae=a(),h(D.$$.fragment),ne=a(),h(Fe.$$.fragment),Q=a(),B=c("div"),h(We.$$.fragment),at=a(),Pe=c("p"),Pe.textContent=Vt,be=a(),pe=c("p"),pe.innerHTML=Oe,se=a(),C=c("p"),C.innerHTML=P,A=a(),q=c("div"),h(O.$$.fragment),Ze=a(),Te=c("p"),Te.innerHTML=ye,Ve=a(),h(I.$$.fragment),L=a(),h(E.$$.fragment),it=a(),h(qe.$$.fragment),$t=a(),R=c("div"),h(Ee.$$.fragment),ut=a(),ht=c("p"),ht.textContent=Ye,To=a(),io=c("p"),io.innerHTML=Ot,Kt=a(),lo=c("p"),lo.innerHTML=yo,Xo=a(),yt=c("div"),h(Mt.$$.fragment),eo=a(),co=c("p"),co.innerHTML=Mo,Ho=a(),h(Dt.$$.fragment),po=a(),h(It.$$.fragment),Io=a(),h(xt.$$.fragment),vo=a(),re=c("div"),h(Pt.$$.fragment),Jo=a(),Rt=c("p"),Rt.innerHTML=on,zo=a(),Xt=c("p"),Xt.innerHTML=nn,Uo=a(),we=c("div"),h(Lt.$$.fragment),Fo=a(),Ht=c("p"),Ht.innerHTML=sn,Wo=a(),h(kt.$$.fragment),Yo=a(),h(Qt.$$.fragment),to=a(),h(mo.$$.fragment),wo=a(),Le=c("div"),h(jt.$$.fragment),Zo=a(),Yt=c("p"),Yt.textContent=rn,$o=a(),Bt=c("p"),Bt.innerHTML=Vo,Jt=a(),Gt=c("p"),Gt.innerHTML=So,fo=a(),zt=c("div"),h(oo.$$.fragment),At=a(),Nt=c("p"),Nt.innerHTML=Do,uo=a(),h(no.$$.fragment),xo=a(),h(vt.$$.fragment),this.h()},l(f){g(e.$$.fragment,f),d=i(f),o=p(f,"DIV",{class:!0});var x=k(o);g(n.$$.fragment,x),m=i(x),t=p(x,"P",{"data-svelte-h":!0}),M(t)!=="svelte-1nknwqm"&&(t.innerHTML=v),Me=i(x),X=p(x,"P",{"data-svelte-h":!0}),M(X)!=="svelte-hswkmf"&&(X.innerHTML=$e),bt=i(x),W=p(x,"DIV",{class:!0});var Ut=k(W);g(H.$$.fragment,Ut),me=i(Ut),z=p(Ut,"P",{"data-svelte-h":!0}),M(z)!=="svelte-1hyqtiv"&&(z.innerHTML=ae),tt=i(Ut),g(G.$$.fragment,Ut),pt=i(Ut),g(fe.$$.fragment,Ut),Ut.forEach(l),Je=i(x),K=p(x,"DIV",{class:!0});var gt=k(K);g(ze.$$.fragment,gt),Y=i(gt),ue=p(gt,"P",{"data-svelte-h":!0}),M(ue)!=="svelte-1hyqtiv"&&(ue.innerHTML=mt),ot=i(gt),g(ie.$$.fragment,gt),De=i(gt),g(Z.$$.fragment,gt),gt.forEach(l),xe=i(x),ee=p(x,"DIV",{class:!0});var qt=k(ee);g(de.$$.fragment,qt),ce=i(qt),Be=p(qt,"P",{"data-svelte-h":!0}),M(Be)!=="svelte-1hyqtiv"&&(Be.innerHTML=Ge),ft=i(qt),g(Ue.$$.fragment,qt),Xe=i(qt),g(N.$$.fragment,qt),qt.forEach(l),x.forEach(l),Ie=i(f),g(le.$$.fragment,f),he=i(f),V=p(f,"DIV",{class:!0});var Ft=k(V);g(nt.$$.fragment,Ft),ve=i(Ft),Ce=p(Ft,"P",{"data-svelte-h":!0}),M(Ce)!=="svelte-285f9x"&&(Ce.innerHTML=wt),F=i(Ft),ge=p(Ft,"P",{"data-svelte-h":!0}),M(ge)!=="svelte-hswkmf"&&(ge.innerHTML=st),Ne=i(Ft),_e=p(Ft,"DIV",{class:!0});var Ke=k(_e);g(Qe.$$.fragment,Ke),te=i(Ke),He=p(Ke,"P",{"data-svelte-h":!0}),M(He)!=="svelte-1mbu317"&&(He.innerHTML=rt),S=i(Ke),g(oe.$$.fragment,Ke),Ae=i(Ke),g(D.$$.fragment,Ke),Ke.forEach(l),Ft.forEach(l),ne=i(f),g(Fe.$$.fragment,f),Q=i(f),B=p(f,"DIV",{class:!0});var Se=k(B);g(We.$$.fragment,Se),at=i(Se),Pe=p(Se,"P",{"data-svelte-h":!0}),M(Pe)!=="svelte-1gij27p"&&(Pe.textContent=Vt),be=i(Se),pe=p(Se,"P",{"data-svelte-h":!0}),M(pe)!=="svelte-1nknwqm"&&(pe.innerHTML=Oe),se=i(Se),C=p(Se,"P",{"data-svelte-h":!0}),M(C)!=="svelte-hswkmf"&&(C.innerHTML=P),A=i(Se),q=p(Se,"DIV",{class:!0});var Wt=k(q);g(O.$$.fragment,Wt),Ze=i(Wt),Te=p(Wt,"P",{"data-svelte-h":!0}),M(Te)!=="svelte-1dlmxnn"&&(Te.innerHTML=ye),Ve=i(Wt),g(I.$$.fragment,Wt),L=i(Wt),g(E.$$.fragment,Wt),Wt.forEach(l),Se.forEach(l),it=i(f),g(qe.$$.fragment,f),$t=i(f),R=p(f,"DIV",{class:!0});var et=k(R);g(Ee.$$.fragment,et),ut=i(et),ht=p(et,"P",{"data-svelte-h":!0}),M(ht)!=="svelte-sjom46"&&(ht.textContent=Ye),To=i(et),io=p(et,"P",{"data-svelte-h":!0}),M(io)!=="svelte-1nknwqm"&&(io.innerHTML=Ot),Kt=i(et),lo=p(et,"P",{"data-svelte-h":!0}),M(lo)!=="svelte-hswkmf"&&(lo.innerHTML=yo),Xo=i(et),yt=p(et,"DIV",{class:!0});var Et=k(yt);g(Mt.$$.fragment,Et),eo=i(Et),co=p(Et,"P",{"data-svelte-h":!0}),M(co)!=="svelte-1nxv7i3"&&(co.innerHTML=Mo),Ho=i(Et),g(Dt.$$.fragment,Et),po=i(Et),g(It.$$.fragment,Et),Et.forEach(l),et.forEach(l),Io=i(f),g(xt.$$.fragment,f),vo=i(f),re=p(f,"DIV",{class:!0});var Ct=k(re);g(Pt.$$.fragment,Ct),Jo=i(Ct),Rt=p(Ct,"P",{"data-svelte-h":!0}),M(Rt)!=="svelte-175p3yg"&&(Rt.innerHTML=on),zo=i(Ct),Xt=p(Ct,"P",{"data-svelte-h":!0}),M(Xt)!=="svelte-hswkmf"&&(Xt.innerHTML=nn),Uo=i(Ct),we=p(Ct,"DIV",{class:!0});var lt=k(we);g(Lt.$$.fragment,lt),Fo=i(lt),Ht=p(lt,"P",{"data-svelte-h":!0}),M(Ht)!=="svelte-a9oaxn"&&(Ht.innerHTML=sn),Wo=i(lt),g(kt.$$.fragment,lt),Yo=i(lt),g(Qt.$$.fragment,lt),lt.forEach(l),Ct.forEach(l),to=i(f),g(mo.$$.fragment,f),wo=i(f),Le=p(f,"DIV",{class:!0});var _t=k(Le);g(jt.$$.fragment,_t),Zo=i(_t),Yt=p(_t,"P",{"data-svelte-h":!0}),M(Yt)!=="svelte-fhevdc"&&(Yt.textContent=rn),$o=i(_t),Bt=p(_t,"P",{"data-svelte-h":!0}),M(Bt)!=="svelte-1nknwqm"&&(Bt.innerHTML=Vo),Jt=i(_t),Gt=p(_t,"P",{"data-svelte-h":!0}),M(Gt)!=="svelte-hswkmf"&&(Gt.innerHTML=So),fo=i(_t),zt=p(_t,"DIV",{class:!0});var dt=k(zt);g(oo.$$.fragment,dt),At=i(dt),Nt=p(dt,"P",{"data-svelte-h":!0}),M(Nt)!=="svelte-nzj1ef"&&(Nt.innerHTML=Do),uo=i(dt),g(no.$$.fragment,dt),xo=i(dt),g(vt.$$.fragment,dt),dt.forEach(l),_t.forEach(l),this.h()},h(){j(W,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(K,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(ee,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(o,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(_e,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(V,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(q,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(B,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(yt,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(R,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(we,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(re,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(zt,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(Le,"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,x){_(e,f,x),u(f,d,x),u(f,o,x),_(n,o,null),s(o,m),s(o,t),s(o,Me),s(o,X),s(o,bt),s(o,W),_(H,W,null),s(W,me),s(W,z),s(W,tt),_(G,W,null),s(W,pt),_(fe,W,null),s(o,Je),s(o,K),_(ze,K,null),s(K,Y),s(K,ue),s(K,ot),_(ie,K,null),s(K,De),_(Z,K,null),s(o,xe),s(o,ee),_(de,ee,null),s(ee,ce),s(ee,Be),s(ee,ft),_(Ue,ee,null),s(ee,Xe),_(N,ee,null),u(f,Ie,x),_(le,f,x),u(f,he,x),u(f,V,x),_(nt,V,null),s(V,ve),s(V,Ce),s(V,F),s(V,ge),s(V,Ne),s(V,_e),_(Qe,_e,null),s(_e,te),s(_e,He),s(_e,S),_(oe,_e,null),s(_e,Ae),_(D,_e,null),u(f,ne,x),_(Fe,f,x),u(f,Q,x),u(f,B,x),_(We,B,null),s(B,at),s(B,Pe),s(B,be),s(B,pe),s(B,se),s(B,C),s(B,A),s(B,q),_(O,q,null),s(q,Ze),s(q,Te),s(q,Ve),_(I,q,null),s(q,L),_(E,q,null),u(f,it,x),_(qe,f,x),u(f,$t,x),u(f,R,x),_(Ee,R,null),s(R,ut),s(R,ht),s(R,To),s(R,io),s(R,Kt),s(R,lo),s(R,Xo),s(R,yt),_(Mt,yt,null),s(yt,eo),s(yt,co),s(yt,Ho),_(Dt,yt,null),s(yt,po),_(It,yt,null),u(f,Io,x),_(xt,f,x),u(f,vo,x),u(f,re,x),_(Pt,re,null),s(re,Jo),s(re,Rt),s(re,zo),s(re,Xt),s(re,Uo),s(re,we),_(Lt,we,null),s(we,Fo),s(we,Ht),s(we,Wo),_(kt,we,null),s(we,Yo),_(Qt,we,null),u(f,to,x),_(mo,f,x),u(f,wo,x),u(f,Le,x),_(jt,Le,null),s(Le,Zo),s(Le,Yt),s(Le,$o),s(Le,Bt),s(Le,Jt),s(Le,Gt),s(Le,fo),s(Le,zt),_(oo,zt,null),s(zt,At),s(zt,Nt),s(zt,uo),_(no,zt,null),s(zt,xo),_(vt,zt,null),Co=!0},p(f,x){const Ut={};x&2&&(Ut.$$scope={dirty:x,ctx:f}),G.$set(Ut);const gt={};x&2&&(gt.$$scope={dirty:x,ctx:f}),fe.$set(gt);const qt={};x&2&&(qt.$$scope={dirty:x,ctx:f}),ie.$set(qt);const Ft={};x&2&&(Ft.$$scope={dirty:x,ctx:f}),Z.$set(Ft);const Ke={};x&2&&(Ke.$$scope={dirty:x,ctx:f}),Ue.$set(Ke);const Se={};x&2&&(Se.$$scope={dirty:x,ctx:f}),N.$set(Se);const Wt={};x&2&&(Wt.$$scope={dirty:x,ctx:f}),oe.$set(Wt);const et={};x&2&&(et.$$scope={dirty:x,ctx:f}),D.$set(et);const Et={};x&2&&(Et.$$scope={dirty:x,ctx:f}),I.$set(Et);const Ct={};x&2&&(Ct.$$scope={dirty:x,ctx:f}),E.$set(Ct);const lt={};x&2&&(lt.$$scope={dirty:x,ctx:f}),Dt.$set(lt);const _t={};x&2&&(_t.$$scope={dirty:x,ctx:f}),It.$set(_t);const dt={};x&2&&(dt.$$scope={dirty:x,ctx:f}),kt.$set(dt);const ho={};x&2&&(ho.$$scope={dirty:x,ctx:f}),Qt.$set(ho);const Bo={};x&2&&(Bo.$$scope={dirty:x,ctx:f}),no.$set(Bo);const go={};x&2&&(go.$$scope={dirty:x,ctx:f}),vt.$set(go)},i(f){Co||(b(e.$$.fragment,f),b(n.$$.fragment,f),b(H.$$.fragment,f),b(G.$$.fragment,f),b(fe.$$.fragment,f),b(ze.$$.fragment,f),b(ie.$$.fragment,f),b(Z.$$.fragment,f),b(de.$$.fragment,f),b(Ue.$$.fragment,f),b(N.$$.fragment,f),b(le.$$.fragment,f),b(nt.$$.fragment,f),b(Qe.$$.fragment,f),b(oe.$$.fragment,f),b(D.$$.fragment,f),b(Fe.$$.fragment,f),b(We.$$.fragment,f),b(O.$$.fragment,f),b(I.$$.fragment,f),b(E.$$.fragment,f),b(qe.$$.fragment,f),b(Ee.$$.fragment,f),b(Mt.$$.fragment,f),b(Dt.$$.fragment,f),b(It.$$.fragment,f),b(xt.$$.fragment,f),b(Pt.$$.fragment,f),b(Lt.$$.fragment,f),b(kt.$$.fragment,f),b(Qt.$$.fragment,f),b(mo.$$.fragment,f),b(jt.$$.fragment,f),b(oo.$$.fragment,f),b(no.$$.fragment,f),b(vt.$$.fragment,f),Co=!0)},o(f){T(e.$$.fragment,f),T(n.$$.fragment,f),T(H.$$.fragment,f),T(G.$$.fragment,f),T(fe.$$.fragment,f),T(ze.$$.fragment,f),T(ie.$$.fragment,f),T(Z.$$.fragment,f),T(de.$$.fragment,f),T(Ue.$$.fragment,f),T(N.$$.fragment,f),T(le.$$.fragment,f),T(nt.$$.fragment,f),T(Qe.$$.fragment,f),T(oe.$$.fragment,f),T(D.$$.fragment,f),T(Fe.$$.fragment,f),T(We.$$.fragment,f),T(O.$$.fragment,f),T(I.$$.fragment,f),T(E.$$.fragment,f),T(qe.$$.fragment,f),T(Ee.$$.fragment,f),T(Mt.$$.fragment,f),T(Dt.$$.fragment,f),T(It.$$.fragment,f),T(xt.$$.fragment,f),T(Pt.$$.fragment,f),T(Lt.$$.fragment,f),T(kt.$$.fragment,f),T(Qt.$$.fragment,f),T(mo.$$.fragment,f),T(jt.$$.fragment,f),T(oo.$$.fragment,f),T(no.$$.fragment,f),T(vt.$$.fragment,f),Co=!1},d(f){f&&(l(d),l(o),l(Ie),l(he),l(V),l(ne),l(Q),l(B),l(it),l($t),l(R),l(Io),l(vo),l(re),l(to),l(wo),l(Le)),y(e,f),y(n),y(H),y(G),y(fe),y(ze),y(ie),y(Z),y(de),y(Ue),y(N),y(le,f),y(nt),y(Qe),y(oe),y(D),y(Fe,f),y(We),y(O),y(I),y(E),y(qe,f),y(Ee),y(Mt),y(Dt),y(It),y(xt,f),y(Pt),y(Lt),y(kt),y(Qt),y(mo,f),y(jt),y(oo),y(no),y(vt)}}}function fs(w){let e,d;return e=new $n({props:{$$slots:{default:[ms]},$$scope:{ctx:w}}}),{c(){h(e.$$.fragment)},l(o){g(e.$$.fragment,o)},m(o,n){_(e,o,n),d=!0},p(o,n){const m={};n&2&&(m.$$scope={dirty:n,ctx:o}),e.$set(m)},i(o){d||(b(e.$$.fragment,o),d=!0)},o(o){T(e.$$.fragment,o),d=!1},d(o){y(e,o)}}}function us(w){let e,d="TensorFlow models and layers in <code>transformers</code> accept two formats as input:",o,n,m="<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,v,Me=`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:`,X,$e,bt=`<li>a single Tensor with <code>input_ids</code> only and nothing else: <code>model(input_ids)</code></li> <li>a list of varying length with one or several input Tensors IN THE ORDER given in the docstring:
<code>model([input_ids, attention_mask])</code> or <code>model([input_ids, attention_mask, token_type_ids])</code></li> <li>a dictionary with one or several input Tensors associated to the input names given in the docstring:
<code>model({&quot;input_ids&quot;: input_ids, &quot;token_type_ids&quot;: token_type_ids})</code></li>`,W,H,me=`Note that when creating models and layers with
<a href="https://keras.io/guides/making_new_layers_and_models_via_subclassing/" rel="nofollow">subclassing</a> then you don’t need to worry
about any of this, as you can just pass inputs like you would to any other Python function!`;return{c(){e=c("p"),e.innerHTML=d,o=a(),n=c("ul"),n.innerHTML=m,t=a(),v=c("p"),v.innerHTML=Me,X=a(),$e=c("ul"),$e.innerHTML=bt,W=a(),H=c("p"),H.innerHTML=me},l(z){e=p(z,"P",{"data-svelte-h":!0}),M(e)!=="svelte-1ajbfxg"&&(e.innerHTML=d),o=i(z),n=p(z,"UL",{"data-svelte-h":!0}),M(n)!=="svelte-qm1t26"&&(n.innerHTML=m),t=i(z),v=p(z,"P",{"data-svelte-h":!0}),M(v)!=="svelte-1v9qsc5"&&(v.innerHTML=Me),X=i(z),$e=p(z,"UL",{"data-svelte-h":!0}),M($e)!=="svelte-15scerc"&&($e.innerHTML=bt),W=i(z),H=p(z,"P",{"data-svelte-h":!0}),M(H)!=="svelte-1an3odd"&&(H.innerHTML=me)},m(z,ae){u(z,e,ae),u(z,o,ae),u(z,n,ae),u(z,t,ae),u(z,v,ae),u(z,X,ae),u(z,$e,ae),u(z,W,ae),u(z,H,ae)},p:U,d(z){z&&(l(e),l(o),l(n),l(t),l(v),l(X),l($e),l(W),l(H))}}}function hs(w){let e,d=`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=c("p"),e.innerHTML=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-fincs2"&&(e.innerHTML=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function gs(w){let e,d="Examples:",o,n,m;return n=new je({props:{code:"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",highlighted:`<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><span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> requests
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoProcessor, TFCLIPModel

<span class="hljs-meta">&gt;&gt;&gt; </span>model = TFCLIPModel.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>processor = AutoProcessor.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>url = <span class="hljs-string">&quot;http://images.cocodataset.org/val2017/000000039769.jpg&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>image = Image.<span class="hljs-built_in">open</span>(requests.get(url, stream=<span class="hljs-literal">True</span>).raw)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = processor(
<span class="hljs-meta">... </span>    text=[<span class="hljs-string">&quot;a photo of a cat&quot;</span>, <span class="hljs-string">&quot;a photo of a dog&quot;</span>], images=image, return_tensors=<span class="hljs-string">&quot;tf&quot;</span>, padding=<span class="hljs-literal">True</span>
<span class="hljs-meta">... </span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>outputs = model(**inputs)
<span class="hljs-meta">&gt;&gt;&gt; </span>logits_per_image = outputs.logits_per_image  <span class="hljs-comment"># this is the image-text similarity score</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>probs = tf.nn.softmax(logits_per_image, axis=<span class="hljs-number">1</span>)  <span class="hljs-comment"># we can take the softmax to get the label probabilities</span>`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-kvfsh7"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function _s(w){let e,d=`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=c("p"),e.innerHTML=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-fincs2"&&(e.innerHTML=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function bs(w){let e,d="Examples:",o,n,m;return n=new je({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, TFCLIPModel

<span class="hljs-meta">&gt;&gt;&gt; </span>model = TFCLIPModel.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = tokenizer([<span class="hljs-string">&quot;a photo of a cat&quot;</span>, <span class="hljs-string">&quot;a photo of a dog&quot;</span>], padding=<span class="hljs-literal">True</span>, return_tensors=<span class="hljs-string">&quot;tf&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>text_features = model.get_text_features(**inputs)`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-kvfsh7"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function Ts(w){let e,d=`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=c("p"),e.innerHTML=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-fincs2"&&(e.innerHTML=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function ys(w){let e,d="Examples:",o,n,m;return n=new je({props:{code:"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",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> requests
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoProcessor, TFCLIPModel

<span class="hljs-meta">&gt;&gt;&gt; </span>model = TFCLIPModel.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>processor = AutoProcessor.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>url = <span class="hljs-string">&quot;http://images.cocodataset.org/val2017/000000039769.jpg&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>image = Image.<span class="hljs-built_in">open</span>(requests.get(url, stream=<span class="hljs-literal">True</span>).raw)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = processor(images=image, return_tensors=<span class="hljs-string">&quot;tf&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>image_features = model.get_image_features(**inputs)`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-kvfsh7"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function Ms(w){let e,d=`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=c("p"),e.innerHTML=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-fincs2"&&(e.innerHTML=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function Is(w){let e,d="Examples:",o,n,m;return n=new je({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, TFCLIPTextModel

<span class="hljs-meta">&gt;&gt;&gt; </span>model = TFCLIPTextModel.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = tokenizer([<span class="hljs-string">&quot;a photo of a cat&quot;</span>, <span class="hljs-string">&quot;a photo of a dog&quot;</span>], padding=<span class="hljs-literal">True</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_state = outputs.last_hidden_state
<span class="hljs-meta">&gt;&gt;&gt; </span>pooled_output = outputs.pooler_output  <span class="hljs-comment"># pooled (EOS token) states</span>`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-kvfsh7"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function vs(w){let e,d=`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=c("p"),e.innerHTML=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-fincs2"&&(e.innerHTML=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function Cs(w){let e,d="Examples:",o,n,m;return n=new je({props:{code:"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",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> requests
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoProcessor, TFCLIPVisionModel

<span class="hljs-meta">&gt;&gt;&gt; </span>model = TFCLIPVisionModel.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>processor = AutoProcessor.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>url = <span class="hljs-string">&quot;http://images.cocodataset.org/val2017/000000039769.jpg&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>image = Image.<span class="hljs-built_in">open</span>(requests.get(url, stream=<span class="hljs-literal">True</span>).raw)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = processor(images=image, 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_state = outputs.last_hidden_state
<span class="hljs-meta">&gt;&gt;&gt; </span>pooled_output = outputs.pooler_output  <span class="hljs-comment"># pooled CLS states</span>`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-kvfsh7"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function ws(w){let e,d,o,n,m,t,v=`This model inherits from <a href="/docs/transformers/pr_37350/en/main_classes/model#transformers.TFPreTrainedModel">TFPreTrainedModel</a>. Check the superclass documentation for the generic methods the
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
etc.)`,Me,X,$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.`,bt,W,H,me,z,ae,tt,G='The <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.TFCLIPModel">TFCLIPModel</a> forward method, overrides the <code>__call__</code> special method.',pt,fe,Je,K,ze,Y,ue,mt,ot,ie='The <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.TFCLIPModel">TFCLIPModel</a> forward method, overrides the <code>__call__</code> special method.',De,Z,xe,ee,de,ce,Be,Ge,ft,Ue='The <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.TFCLIPModel">TFCLIPModel</a> forward method, overrides the <code>__call__</code> special method.',Xe,N,Ie,le,he,V,nt,ve,Ce,wt,F,ge,st,Ne,_e='The <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.TFCLIPTextModel">TFCLIPTextModel</a> forward method, overrides the <code>__call__</code> special method.',Qe,te,He,rt,S,oe,Ae,D,ne,Fe,Q,B,We,at,Pe='The <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.TFCLIPVisionModel">TFCLIPVisionModel</a> forward method, overrides the <code>__call__</code> special method.',Vt,be,pe,Oe,se;return e=new ke({props:{title:"TFCLIPModel",local:"transformers.TFCLIPModel",headingTag:"h2"}}),n=new J({props:{name:"class transformers.TFCLIPModel",anchor:"transformers.TFCLIPModel",parameters:[{name:"config",val:": CLIPConfig"},{name:"*inputs",val:""},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.TFCLIPModel.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPConfig">CLIPConfig</a>) &#x2014; Model configuration class with all the parameters of the model.
Initializing with a config file does not load the weights associated with the model, only the
configuration. Check out the <a href="/docs/transformers/pr_37350/en/main_classes/model#transformers.TFPreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_tf_clip.py#L1300"}}),W=new ct({props:{$$slots:{default:[us]},$$scope:{ctx:w}}}),z=new J({props:{name:"call",anchor:"transformers.TFCLIPModel.call",parameters:[{name:"input_ids",val:": TFModelInputType | None = None"},{name:"pixel_values",val:": TFModelInputType | None = None"},{name:"attention_mask",val:": np.ndarray | tf.Tensor | None = None"},{name:"position_ids",val:": np.ndarray | tf.Tensor | None = None"},{name:"return_loss",val:": Optional[bool] = 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:": bool = False"}],parametersDescription:[{anchor:"transformers.TFCLIPModel.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/pr_37350/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_37350/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> and
<a href="/docs/transformers/pr_37350/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> for details.</p>
<p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.TFCLIPModel.call.pixel_values",description:`<strong>pixel_values</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, num_channels, height, width)</code>) &#x2014;
Pixel values. Pixel values can be obtained using <a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoImageProcessor">AutoImageProcessor</a>. See
<a href="/docs/transformers/pr_37350/en/model_doc/vilt#transformers.ViltFeatureExtractor.__call__">CLIPImageProcessor.<strong>call</strong>()</a> for details.`,name:"pixel_values"},{anchor:"transformers.TFCLIPModel.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.TFCLIPModel.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.TFCLIPModel.call.return_loss",description:`<strong>return_loss</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return the contrastive loss.`,name:"return_loss"},{anchor:"transformers.TFCLIPModel.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. This argument can be used only in eager mode, in graph mode the value in the
config will be used instead.`,name:"output_attentions"},{anchor:"transformers.TFCLIPModel.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. This argument can be used only in eager mode, in graph mode the value in the config will be
used instead.`,name:"output_hidden_states"},{anchor:"transformers.TFCLIPModel.call.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_37350/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple. This argument can be used in
eager mode, in graph mode the value will always be set to True.`,name:"return_dict"},{anchor:"transformers.TFCLIPModel.call.training",description:`<strong>training</strong> (<code>bool</code>, <em>optional</em>, defaults to \`False&#x201C;) &#x2014;
Whether or not to use the model in training mode (some modules like dropout modules have different
behaviors between training and evaluation).`,name:"training"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_tf_clip.py#L1391",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <code>transformers.models.clip.modeling_tf_clip.TFCLIPOutput</code> 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 (<code>&lt;class 'transformers.models.clip.configuration_clip.CLIPConfig'&gt;</code>) and inputs.</p>
<ul>
<li><strong>loss</strong> (<code>tf.Tensor</code> of shape <code>(1,)</code>, <em>optional</em>, returned when <code>return_loss</code> is <code>True</code>) — Contrastive loss for image-text similarity.</li>
<li><strong>logits_per_image:(<code>tf.Tensor</code></strong> of shape <code>(image_batch_size, text_batch_size)</code>) — The scaled dot product scores between <code>image_embeds</code> and <code>text_embeds</code>. This represents the image-text
similarity scores.</li>
<li><strong>logits_per_text:(<code>tf.Tensor</code></strong> of shape <code>(text_batch_size, image_batch_size)</code>) — The scaled dot product scores between <code>text_embeds</code> and <code>image_embeds</code>. This represents the text-image
similarity scores.</li>
<li><strong>text_embeds(<code>tf.Tensor</code></strong> of shape <code>(batch_size, output_dim</code>) — The text embeddings obtained by applying the projection layer to the pooled output of <a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.TFCLIPTextModel"
>TFCLIPTextModel</a>.</li>
<li><strong>image_embeds(<code>tf.Tensor</code></strong> of shape <code>(batch_size, output_dim</code>) — The image embeddings obtained by applying the projection layer to the pooled output of
<a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.TFCLIPVisionModel"
>TFCLIPVisionModel</a>.</li>
<li><strong>text_model_output(<code>~modeling_tf_utils.TFBaseModelOutputWithPooling</code>):</strong>
The output of the <a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.TFCLIPTextModel"
>TFCLIPTextModel</a>.</li>
<li><strong>vision_model_output(<code>~modeling_tf_utils.TFBaseModelOutputWithPooling</code>):</strong>
The output of the <a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.TFCLIPVisionModel"
>TFCLIPVisionModel</a>.</li>
</ul>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><code>transformers.models.clip.modeling_tf_clip.TFCLIPOutput</code> or <code>tuple(tf.Tensor)</code></p>
`}}),fe=new ct({props:{$$slots:{default:[hs]},$$scope:{ctx:w}}}),K=new Re({props:{anchor:"transformers.TFCLIPModel.call.example",$$slots:{default:[gs]},$$scope:{ctx:w}}}),ue=new J({props:{name:"get_text_features",anchor:"transformers.TFCLIPModel.get_text_features",parameters:[{name:"input_ids",val:": TFModelInputType | None = None"},{name:"attention_mask",val:": np.ndarray | tf.Tensor | None = None"},{name:"position_ids",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:": bool = False"}],parametersDescription:[{anchor:"transformers.TFCLIPModel.get_text_features.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/pr_37350/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_37350/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> and
<a href="/docs/transformers/pr_37350/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> for details.</p>
<p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.TFCLIPModel.get_text_features.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.TFCLIPModel.get_text_features.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.TFCLIPModel.get_text_features.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. This argument can be used only in eager mode, in graph mode the value in the
config will be used instead.`,name:"output_attentions"},{anchor:"transformers.TFCLIPModel.get_text_features.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. This argument can be used only in eager mode, in graph mode the value in the config will be
used instead.`,name:"output_hidden_states"},{anchor:"transformers.TFCLIPModel.get_text_features.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_37350/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple. This argument can be used in
eager mode, in graph mode the value will always be set to True.`,name:"return_dict"},{anchor:"transformers.TFCLIPModel.get_text_features.training",description:`<strong>training</strong> (<code>bool</code>, <em>optional</em>, defaults to \`False&#x201C;) &#x2014;
Whether or not to use the model in training mode (some modules like dropout modules have different
behaviors between training and evaluation).`,name:"training"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_tf_clip.py#L1309",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>The text embeddings obtained by applying
the projection layer to the pooled output of <a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.TFCLIPTextModel"
>TFCLIPTextModel</a>.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p>text_features (<code>tf.Tensor</code> of shape <code>(batch_size, output_dim</code>)</p>
`}}),Z=new ct({props:{$$slots:{default:[_s]},$$scope:{ctx:w}}}),ee=new Re({props:{anchor:"transformers.TFCLIPModel.get_text_features.example",$$slots:{default:[bs]},$$scope:{ctx:w}}}),Be=new J({props:{name:"get_image_features",anchor:"transformers.TFCLIPModel.get_image_features",parameters:[{name:"pixel_values",val:": TFModelInputType | 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:": bool = False"}],parametersDescription:[{anchor:"transformers.TFCLIPModel.get_image_features.pixel_values",description:`<strong>pixel_values</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, num_channels, height, width)</code>) &#x2014;
Pixel values. Pixel values can be obtained using <a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoImageProcessor">AutoImageProcessor</a>. See
<a href="/docs/transformers/pr_37350/en/model_doc/vilt#transformers.ViltFeatureExtractor.__call__">CLIPImageProcessor.<strong>call</strong>()</a> for details. output_attentions (<code>bool</code>, <em>optional</em>): Whether or not to
return the attentions tensors of all attention layers. See <code>attentions</code> under returned tensors for more
detail. This argument can be used only in eager mode, in graph mode the value in the config will be used
instead.`,name:"pixel_values"},{anchor:"transformers.TFCLIPModel.get_image_features.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. This argument can be used only in eager mode, in graph mode the value in the config will be
used instead.`,name:"output_hidden_states"},{anchor:"transformers.TFCLIPModel.get_image_features.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_37350/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple. This argument can be used in
eager mode, in graph mode the value will always be set to True.`,name:"return_dict"},{anchor:"transformers.TFCLIPModel.get_image_features.training",description:`<strong>training</strong> (<code>bool</code>, <em>optional</em>, defaults to \`False&#x201C;) &#x2014;
Whether or not to use the model in training mode (some modules like dropout modules have different
behaviors between training and evaluation).`,name:"training"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_tf_clip.py#L1349",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>The image embeddings obtained by applying
the projection layer to the pooled output of <a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.TFCLIPVisionModel"
>TFCLIPVisionModel</a>.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p>image_features (<code>tf.Tensor</code> of shape <code>(batch_size, output_dim</code>)</p>
`}}),N=new ct({props:{$$slots:{default:[Ts]},$$scope:{ctx:w}}}),le=new Re({props:{anchor:"transformers.TFCLIPModel.get_image_features.example",$$slots:{default:[ys]},$$scope:{ctx:w}}}),V=new ke({props:{title:"TFCLIPTextModel",local:"transformers.TFCLIPTextModel",headingTag:"h2"}}),Ce=new J({props:{name:"class transformers.TFCLIPTextModel",anchor:"transformers.TFCLIPTextModel",parameters:[{name:"config",val:": CLIPTextConfig"},{name:"*inputs",val:""},{name:"**kwargs",val:""}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_tf_clip.py#L1178"}}),ge=new J({props:{name:"call",anchor:"transformers.TFCLIPTextModel.call",parameters:[{name:"input_ids",val:": TFModelInputType | None = None"},{name:"attention_mask",val:": np.ndarray | tf.Tensor | None = None"},{name:"position_ids",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.TFCLIPTextModel.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/pr_37350/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_37350/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> and
<a href="/docs/transformers/pr_37350/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> for details.</p>
<p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.TFCLIPTextModel.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.TFCLIPTextModel.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.TFCLIPTextModel.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. This argument can be used only in eager mode, in graph mode the value in the
config will be used instead.`,name:"output_attentions"},{anchor:"transformers.TFCLIPTextModel.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. This argument can be used only in eager mode, in graph mode the value in the config will be
used instead.`,name:"output_hidden_states"},{anchor:"transformers.TFCLIPTextModel.call.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_37350/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple. This argument can be used in
eager mode, in graph mode the value will always be set to True.`,name:"return_dict"},{anchor:"transformers.TFCLIPTextModel.call.training",description:`<strong>training</strong> (<code>bool</code>, <em>optional</em>, defaults to \`False&#x201C;) &#x2014;
Whether or not to use the model in training mode (some modules like dropout modules have different
behaviors between training and evaluation).`,name:"training"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_tf_clip.py#L1186",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/pr_37350/en/main_classes/output#transformers.modeling_tf_outputs.TFBaseModelOutputWithPooling"
>transformers.modeling_tf_outputs.TFBaseModelOutputWithPooling</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 (<code>&lt;class 'transformers.models.clip.configuration_clip.CLIPTextConfig'&gt;</code>) 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>pooler_output</strong> (<code>tf.Tensor</code> of shape <code>(batch_size, hidden_size)</code>) — Last layer hidden-state of the first token of the sequence (classification token) further processed by a
Linear layer and a Tanh activation function. The Linear layer weights are trained from the next sentence
prediction (classification) objective during pretraining.</p>
<p>This output is usually <em>not</em> a good summary of the semantic content of the input, you’re often better with
averaging or pooling the sequence of hidden-states for the whole input sequence.</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/pr_37350/en/main_classes/output#transformers.modeling_tf_outputs.TFBaseModelOutputWithPooling"
>transformers.modeling_tf_outputs.TFBaseModelOutputWithPooling</a> or <code>tuple(tf.Tensor)</code></p>
`}}),te=new ct({props:{$$slots:{default:[Ms]},$$scope:{ctx:w}}}),rt=new Re({props:{anchor:"transformers.TFCLIPTextModel.call.example",$$slots:{default:[Is]},$$scope:{ctx:w}}}),oe=new ke({props:{title:"TFCLIPVisionModel",local:"transformers.TFCLIPVisionModel",headingTag:"h2"}}),ne=new J({props:{name:"class transformers.TFCLIPVisionModel",anchor:"transformers.TFCLIPVisionModel",parameters:[{name:"config",val:": CLIPVisionConfig"},{name:"*inputs",val:""},{name:"**kwargs",val:""}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_tf_clip.py#L1238"}}),B=new J({props:{name:"call",anchor:"transformers.TFCLIPVisionModel.call",parameters:[{name:"pixel_values",val:": TFModelInputType | 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.TFCLIPVisionModel.call.pixel_values",description:`<strong>pixel_values</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, num_channels, height, width)</code>) &#x2014;
Pixel values. Pixel values can be obtained using <a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoImageProcessor">AutoImageProcessor</a>. See
<a href="/docs/transformers/pr_37350/en/model_doc/vilt#transformers.ViltFeatureExtractor.__call__">CLIPImageProcessor.<strong>call</strong>()</a> for details. output_attentions (<code>bool</code>, <em>optional</em>): Whether or not to
return the attentions tensors of all attention layers. See <code>attentions</code> under returned tensors for more
detail. This argument can be used only in eager mode, in graph mode the value in the config will be used
instead.`,name:"pixel_values"},{anchor:"transformers.TFCLIPVisionModel.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. This argument can be used only in eager mode, in graph mode the value in the config will be
used instead.`,name:"output_hidden_states"},{anchor:"transformers.TFCLIPVisionModel.call.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_37350/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple. This argument can be used in
eager mode, in graph mode the value will always be set to True.`,name:"return_dict"},{anchor:"transformers.TFCLIPVisionModel.call.training",description:`<strong>training</strong> (<code>bool</code>, <em>optional</em>, defaults to \`False&#x201C;) &#x2014;
Whether or not to use the model in training mode (some modules like dropout modules have different
behaviors between training and evaluation).`,name:"training"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_tf_clip.py#L1247",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/pr_37350/en/main_classes/output#transformers.modeling_tf_outputs.TFBaseModelOutputWithPooling"
>transformers.modeling_tf_outputs.TFBaseModelOutputWithPooling</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 (<code>&lt;class 'transformers.models.clip.configuration_clip.CLIPVisionConfig'&gt;</code>) 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>pooler_output</strong> (<code>tf.Tensor</code> of shape <code>(batch_size, hidden_size)</code>) — Last layer hidden-state of the first token of the sequence (classification token) further processed by a
Linear layer and a Tanh activation function. The Linear layer weights are trained from the next sentence
prediction (classification) objective during pretraining.</p>
<p>This output is usually <em>not</em> a good summary of the semantic content of the input, you’re often better with
averaging or pooling the sequence of hidden-states for the whole input sequence.</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/pr_37350/en/main_classes/output#transformers.modeling_tf_outputs.TFBaseModelOutputWithPooling"
>transformers.modeling_tf_outputs.TFBaseModelOutputWithPooling</a> or <code>tuple(tf.Tensor)</code></p>
`}}),be=new ct({props:{$$slots:{default:[vs]},$$scope:{ctx:w}}}),Oe=new Re({props:{anchor:"transformers.TFCLIPVisionModel.call.example",$$slots:{default:[Cs]},$$scope:{ctx:w}}}),{c(){h(e.$$.fragment),d=a(),o=c("div"),h(n.$$.fragment),m=a(),t=c("p"),t.innerHTML=v,Me=a(),X=c("p"),X.innerHTML=$e,bt=a(),h(W.$$.fragment),H=a(),me=c("div"),h(z.$$.fragment),ae=a(),tt=c("p"),tt.innerHTML=G,pt=a(),h(fe.$$.fragment),Je=a(),h(K.$$.fragment),ze=a(),Y=c("div"),h(ue.$$.fragment),mt=a(),ot=c("p"),ot.innerHTML=ie,De=a(),h(Z.$$.fragment),xe=a(),h(ee.$$.fragment),de=a(),ce=c("div"),h(Be.$$.fragment),Ge=a(),ft=c("p"),ft.innerHTML=Ue,Xe=a(),h(N.$$.fragment),Ie=a(),h(le.$$.fragment),he=a(),h(V.$$.fragment),nt=a(),ve=c("div"),h(Ce.$$.fragment),wt=a(),F=c("div"),h(ge.$$.fragment),st=a(),Ne=c("p"),Ne.innerHTML=_e,Qe=a(),h(te.$$.fragment),He=a(),h(rt.$$.fragment),S=a(),h(oe.$$.fragment),Ae=a(),D=c("div"),h(ne.$$.fragment),Fe=a(),Q=c("div"),h(B.$$.fragment),We=a(),at=c("p"),at.innerHTML=Pe,Vt=a(),h(be.$$.fragment),pe=a(),h(Oe.$$.fragment),this.h()},l(C){g(e.$$.fragment,C),d=i(C),o=p(C,"DIV",{class:!0});var P=k(o);g(n.$$.fragment,P),m=i(P),t=p(P,"P",{"data-svelte-h":!0}),M(t)!=="svelte-1jtltr2"&&(t.innerHTML=v),Me=i(P),X=p(P,"P",{"data-svelte-h":!0}),M(X)!=="svelte-1be7e3c"&&(X.innerHTML=$e),bt=i(P),g(W.$$.fragment,P),H=i(P),me=p(P,"DIV",{class:!0});var A=k(me);g(z.$$.fragment,A),ae=i(A),tt=p(A,"P",{"data-svelte-h":!0}),M(tt)!=="svelte-16xp4cr"&&(tt.innerHTML=G),pt=i(A),g(fe.$$.fragment,A),Je=i(A),g(K.$$.fragment,A),A.forEach(l),ze=i(P),Y=p(P,"DIV",{class:!0});var q=k(Y);g(ue.$$.fragment,q),mt=i(q),ot=p(q,"P",{"data-svelte-h":!0}),M(ot)!=="svelte-16xp4cr"&&(ot.innerHTML=ie),De=i(q),g(Z.$$.fragment,q),xe=i(q),g(ee.$$.fragment,q),q.forEach(l),de=i(P),ce=p(P,"DIV",{class:!0});var O=k(ce);g(Be.$$.fragment,O),Ge=i(O),ft=p(O,"P",{"data-svelte-h":!0}),M(ft)!=="svelte-16xp4cr"&&(ft.innerHTML=Ue),Xe=i(O),g(N.$$.fragment,O),Ie=i(O),g(le.$$.fragment,O),O.forEach(l),P.forEach(l),he=i(C),g(V.$$.fragment,C),nt=i(C),ve=p(C,"DIV",{class:!0});var Ze=k(ve);g(Ce.$$.fragment,Ze),wt=i(Ze),F=p(Ze,"DIV",{class:!0});var Te=k(F);g(ge.$$.fragment,Te),st=i(Te),Ne=p(Te,"P",{"data-svelte-h":!0}),M(Ne)!=="svelte-18k7duv"&&(Ne.innerHTML=_e),Qe=i(Te),g(te.$$.fragment,Te),He=i(Te),g(rt.$$.fragment,Te),Te.forEach(l),Ze.forEach(l),S=i(C),g(oe.$$.fragment,C),Ae=i(C),D=p(C,"DIV",{class:!0});var ye=k(D);g(ne.$$.fragment,ye),Fe=i(ye),Q=p(ye,"DIV",{class:!0});var Ve=k(Q);g(B.$$.fragment,Ve),We=i(Ve),at=p(Ve,"P",{"data-svelte-h":!0}),M(at)!=="svelte-9e3hdb"&&(at.innerHTML=Pe),Vt=i(Ve),g(be.$$.fragment,Ve),pe=i(Ve),g(Oe.$$.fragment,Ve),Ve.forEach(l),ye.forEach(l),this.h()},h(){j(me,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(Y,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(ce,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(o,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(F,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(ve,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(Q,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(D,"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,P){_(e,C,P),u(C,d,P),u(C,o,P),_(n,o,null),s(o,m),s(o,t),s(o,Me),s(o,X),s(o,bt),_(W,o,null),s(o,H),s(o,me),_(z,me,null),s(me,ae),s(me,tt),s(me,pt),_(fe,me,null),s(me,Je),_(K,me,null),s(o,ze),s(o,Y),_(ue,Y,null),s(Y,mt),s(Y,ot),s(Y,De),_(Z,Y,null),s(Y,xe),_(ee,Y,null),s(o,de),s(o,ce),_(Be,ce,null),s(ce,Ge),s(ce,ft),s(ce,Xe),_(N,ce,null),s(ce,Ie),_(le,ce,null),u(C,he,P),_(V,C,P),u(C,nt,P),u(C,ve,P),_(Ce,ve,null),s(ve,wt),s(ve,F),_(ge,F,null),s(F,st),s(F,Ne),s(F,Qe),_(te,F,null),s(F,He),_(rt,F,null),u(C,S,P),_(oe,C,P),u(C,Ae,P),u(C,D,P),_(ne,D,null),s(D,Fe),s(D,Q),_(B,Q,null),s(Q,We),s(Q,at),s(Q,Vt),_(be,Q,null),s(Q,pe),_(Oe,Q,null),se=!0},p(C,P){const A={};P&2&&(A.$$scope={dirty:P,ctx:C}),W.$set(A);const q={};P&2&&(q.$$scope={dirty:P,ctx:C}),fe.$set(q);const O={};P&2&&(O.$$scope={dirty:P,ctx:C}),K.$set(O);const Ze={};P&2&&(Ze.$$scope={dirty:P,ctx:C}),Z.$set(Ze);const Te={};P&2&&(Te.$$scope={dirty:P,ctx:C}),ee.$set(Te);const ye={};P&2&&(ye.$$scope={dirty:P,ctx:C}),N.$set(ye);const Ve={};P&2&&(Ve.$$scope={dirty:P,ctx:C}),le.$set(Ve);const I={};P&2&&(I.$$scope={dirty:P,ctx:C}),te.$set(I);const L={};P&2&&(L.$$scope={dirty:P,ctx:C}),rt.$set(L);const E={};P&2&&(E.$$scope={dirty:P,ctx:C}),be.$set(E);const it={};P&2&&(it.$$scope={dirty:P,ctx:C}),Oe.$set(it)},i(C){se||(b(e.$$.fragment,C),b(n.$$.fragment,C),b(W.$$.fragment,C),b(z.$$.fragment,C),b(fe.$$.fragment,C),b(K.$$.fragment,C),b(ue.$$.fragment,C),b(Z.$$.fragment,C),b(ee.$$.fragment,C),b(Be.$$.fragment,C),b(N.$$.fragment,C),b(le.$$.fragment,C),b(V.$$.fragment,C),b(Ce.$$.fragment,C),b(ge.$$.fragment,C),b(te.$$.fragment,C),b(rt.$$.fragment,C),b(oe.$$.fragment,C),b(ne.$$.fragment,C),b(B.$$.fragment,C),b(be.$$.fragment,C),b(Oe.$$.fragment,C),se=!0)},o(C){T(e.$$.fragment,C),T(n.$$.fragment,C),T(W.$$.fragment,C),T(z.$$.fragment,C),T(fe.$$.fragment,C),T(K.$$.fragment,C),T(ue.$$.fragment,C),T(Z.$$.fragment,C),T(ee.$$.fragment,C),T(Be.$$.fragment,C),T(N.$$.fragment,C),T(le.$$.fragment,C),T(V.$$.fragment,C),T(Ce.$$.fragment,C),T(ge.$$.fragment,C),T(te.$$.fragment,C),T(rt.$$.fragment,C),T(oe.$$.fragment,C),T(ne.$$.fragment,C),T(B.$$.fragment,C),T(be.$$.fragment,C),T(Oe.$$.fragment,C),se=!1},d(C){C&&(l(d),l(o),l(he),l(nt),l(ve),l(S),l(Ae),l(D)),y(e,C),y(n),y(W),y(z),y(fe),y(K),y(ue),y(Z),y(ee),y(Be),y(N),y(le),y(V,C),y(Ce),y(ge),y(te),y(rt),y(oe,C),y(ne),y(B),y(be),y(Oe)}}}function $s(w){let e,d;return e=new $n({props:{$$slots:{default:[ws]},$$scope:{ctx:w}}}),{c(){h(e.$$.fragment)},l(o){g(e.$$.fragment,o)},m(o,n){_(e,o,n),d=!0},p(o,n){const m={};n&2&&(m.$$scope={dirty:n,ctx:o}),e.$set(m)},i(o){d||(b(e.$$.fragment,o),d=!0)},o(o){T(e.$$.fragment,o),d=!1},d(o){y(e,o)}}}function xs(w){let e,d=`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=c("p"),e.innerHTML=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-fincs2"&&(e.innerHTML=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function Ps(w){let e,d="Example:",o,n,m;return n=new je({props:{code:"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",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> jax
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> requests
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoProcessor, FlaxCLIPModel

<span class="hljs-meta">&gt;&gt;&gt; </span>model = FlaxCLIPModel.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>processor = AutoProcessor.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>url = <span class="hljs-string">&quot;http://images.cocodataset.org/val2017/000000039769.jpg&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>image = Image.<span class="hljs-built_in">open</span>(requests.get(url, stream=<span class="hljs-literal">True</span>).raw)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = processor(
<span class="hljs-meta">... </span>    text=[<span class="hljs-string">&quot;a photo of a cat&quot;</span>, <span class="hljs-string">&quot;a photo of a dog&quot;</span>], images=image, return_tensors=<span class="hljs-string">&quot;np&quot;</span>, padding=<span class="hljs-literal">True</span>
<span class="hljs-meta">... </span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>outputs = model(**inputs)
<span class="hljs-meta">&gt;&gt;&gt; </span>logits_per_image = outputs.logits_per_image  <span class="hljs-comment"># this is the image-text similarity score</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>probs = jax.nn.softmax(logits_per_image, axis=<span class="hljs-number">1</span>)  <span class="hljs-comment"># we can take the softmax to get the label probabilities</span>`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-11lpom8"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function Ls(w){let e,d="Examples:",o,n,m;return n=new je({props:{code:"ZnJvbSUyMHRyYW5zZm9ybWVycyUyMGltcG9ydCUyMEF1dG9Ub2tlbml6ZXIlMkMlMjBGbGF4Q0xJUE1vZGVsJTBBJTBBbW9kZWwlMjAlM0QlMjBGbGF4Q0xJUE1vZGVsLmZyb21fcHJldHJhaW5lZCglMjJvcGVuYWklMkZjbGlwLXZpdC1iYXNlLXBhdGNoMzIlMjIpJTBBdG9rZW5pemVyJTIwJTNEJTIwQXV0b1Rva2VuaXplci5mcm9tX3ByZXRyYWluZWQoJTIyb3BlbmFpJTJGY2xpcC12aXQtYmFzZS1wYXRjaDMyJTIyKSUwQSUwQWlucHV0cyUyMCUzRCUyMHRva2VuaXplciglNUIlMjJhJTIwcGhvdG8lMjBvZiUyMGElMjBjYXQlMjIlMkMlMjAlMjJhJTIwcGhvdG8lMjBvZiUyMGElMjBkb2clMjIlNUQlMkMlMjBwYWRkaW5nJTNEVHJ1ZSUyQyUyMHJldHVybl90ZW5zb3JzJTNEJTIybnAlMjIpJTBBdGV4dF9mZWF0dXJlcyUyMCUzRCUyMG1vZGVsLmdldF90ZXh0X2ZlYXR1cmVzKCoqaW5wdXRzKQ==",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, FlaxCLIPModel

<span class="hljs-meta">&gt;&gt;&gt; </span>model = FlaxCLIPModel.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = tokenizer([<span class="hljs-string">&quot;a photo of a cat&quot;</span>, <span class="hljs-string">&quot;a photo of a dog&quot;</span>], padding=<span class="hljs-literal">True</span>, return_tensors=<span class="hljs-string">&quot;np&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>text_features = model.get_text_features(**inputs)`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-kvfsh7"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function ks(w){let e,d="Examples:",o,n,m;return n=new je({props:{code:"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",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> requests
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoProcessor, FlaxCLIPModel

<span class="hljs-meta">&gt;&gt;&gt; </span>model = FlaxCLIPModel.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>processor = AutoProcessor.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>url = <span class="hljs-string">&quot;http://images.cocodataset.org/val2017/000000039769.jpg&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>image = Image.<span class="hljs-built_in">open</span>(requests.get(url, stream=<span class="hljs-literal">True</span>).raw)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = processor(images=image, return_tensors=<span class="hljs-string">&quot;np&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>image_features = model.get_image_features(**inputs)`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-kvfsh7"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function js(w){let e,d=`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=c("p"),e.innerHTML=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-fincs2"&&(e.innerHTML=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function Js(w){let e,d="Example:",o,n,m;return n=new je({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, FlaxCLIPTextModel

<span class="hljs-meta">&gt;&gt;&gt; </span>model = FlaxCLIPTextModel.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = tokenizer([<span class="hljs-string">&quot;a photo of a cat&quot;</span>, <span class="hljs-string">&quot;a photo of a dog&quot;</span>], padding=<span class="hljs-literal">True</span>, return_tensors=<span class="hljs-string">&quot;np&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>outputs = model(**inputs)
<span class="hljs-meta">&gt;&gt;&gt; </span>last_hidden_state = outputs.last_hidden_state
<span class="hljs-meta">&gt;&gt;&gt; </span>pooler_output = outputs.pooler_output  <span class="hljs-comment"># pooled (EOS token) states</span>`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-11lpom8"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function zs(w){let e,d=`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=c("p"),e.innerHTML=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-fincs2"&&(e.innerHTML=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function Us(w){let e,d="Example:",o,n,m;return n=new je({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, FlaxCLIPTextModelWithProjection

<span class="hljs-meta">&gt;&gt;&gt; </span>model = FlaxCLIPTextModelWithProjection.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = tokenizer([<span class="hljs-string">&quot;a photo of a cat&quot;</span>, <span class="hljs-string">&quot;a photo of a dog&quot;</span>], padding=<span class="hljs-literal">True</span>, return_tensors=<span class="hljs-string">&quot;np&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>outputs = model(**inputs)
<span class="hljs-meta">&gt;&gt;&gt; </span>text_embeds = outputs.text_embeds`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-11lpom8"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function Fs(w){let e,d=`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=c("p"),e.innerHTML=d},l(o){e=p(o,"P",{"data-svelte-h":!0}),M(e)!=="svelte-fincs2"&&(e.innerHTML=d)},m(o,n){u(o,e,n)},p:U,d(o){o&&l(e)}}}function Ws(w){let e,d="Example:",o,n,m;return n=new je({props:{code:"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",highlighted:`<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> requests
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoProcessor, FlaxCLIPVisionModel

<span class="hljs-meta">&gt;&gt;&gt; </span>model = FlaxCLIPVisionModel.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>processor = AutoProcessor.from_pretrained(<span class="hljs-string">&quot;openai/clip-vit-base-patch32&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>url = <span class="hljs-string">&quot;http://images.cocodataset.org/val2017/000000039769.jpg&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>image = Image.<span class="hljs-built_in">open</span>(requests.get(url, stream=<span class="hljs-literal">True</span>).raw)

<span class="hljs-meta">&gt;&gt;&gt; </span>inputs = processor(images=image, return_tensors=<span class="hljs-string">&quot;np&quot;</span>)

<span class="hljs-meta">&gt;&gt;&gt; </span>outputs = model(**inputs)
<span class="hljs-meta">&gt;&gt;&gt; </span>last_hidden_state = outputs.last_hidden_state
<span class="hljs-meta">&gt;&gt;&gt; </span>pooler_output = outputs.pooler_output  <span class="hljs-comment"># pooled CLS states</span>`,wrap:!1}}),{c(){e=c("p"),e.textContent=d,o=a(),h(n.$$.fragment)},l(t){e=p(t,"P",{"data-svelte-h":!0}),M(e)!=="svelte-11lpom8"&&(e.textContent=d),o=i(t),g(n.$$.fragment,t)},m(t,v){u(t,e,v),u(t,o,v),_(n,t,v),m=!0},p:U,i(t){m||(b(n.$$.fragment,t),m=!0)},o(t){T(n.$$.fragment,t),m=!1},d(t){t&&(l(e),l(o)),y(n,t)}}}function Zs(w){let e,d,o,n,m,t,v=`This model inherits from <a href="/docs/transformers/pr_37350/en/main_classes/model#transformers.FlaxPreTrainedModel">FlaxPreTrainedModel</a>. Check the superclass documentation for the generic methods the
library implements for all its model (such as downloading, saving and converting weights from PyTorch models)`,Me,X,$e=`This model is also a
<a href="https://flax.readthedocs.io/en/latest/api_reference/flax.linen/module.html" rel="nofollow">flax.linen.Module</a> subclass. Use it as
a regular Flax linen Module and refer to the Flax documentation for all matter related to general usage and
behavior.`,bt,W,H="Finally, this model supports inherent JAX features such as:",me,z,ae='<li><a href="https://jax.readthedocs.io/en/latest/jax.html#just-in-time-compilation-jit" rel="nofollow">Just-In-Time (JIT) compilation</a></li> <li><a href="https://jax.readthedocs.io/en/latest/jax.html#automatic-differentiation" rel="nofollow">Automatic Differentiation</a></li> <li><a href="https://jax.readthedocs.io/en/latest/jax.html#vectorization-vmap" rel="nofollow">Vectorization</a></li> <li><a href="https://jax.readthedocs.io/en/latest/jax.html#parallelization-pmap" rel="nofollow">Parallelization</a></li>',tt,G,pt,fe,Je,K="The <code>FlaxCLIPPreTrainedModel</code> forward method, overrides the <code>__call__</code> special method.",ze,Y,ue,mt,ot,ie,De,Z,xe,ee,de,ce,Be,Ge,ft,Ue,Xe,N,Ie,le,he,V,nt,ve,Ce="The <code>FlaxCLIPTextPreTrainedModel</code> forward method, overrides the <code>__call__</code> special method.",wt,F,ge,st,Ne,_e,Qe,te,He,rt,S,oe,Ae,D,ne="The <code>FlaxCLIPTextPreTrainedModel</code> forward method, overrides the <code>__call__</code> special method.",Fe,Q,B,We,at,Pe,Vt,be,pe,Oe,se,C,P,A,q="The <code>FlaxCLIPVisionPreTrainedModel</code> forward method, overrides the <code>__call__</code> special method.",O,Ze,Te,ye,Ve;return e=new ke({props:{title:"FlaxCLIPModel",local:"transformers.FlaxCLIPModel",headingTag:"h2"}}),n=new J({props:{name:"class transformers.FlaxCLIPModel",anchor:"transformers.FlaxCLIPModel",parameters:[{name:"config",val:": CLIPConfig"},{name:"input_shape",val:": typing.Optional[typing.Tuple] = None"},{name:"seed",val:": int = 0"},{name:"dtype",val:": dtype = <class 'jax.numpy.float32'>"},{name:"_do_init",val:": bool = True"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.FlaxCLIPModel.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPConfig">CLIPConfig</a>) &#x2014; Model configuration class with all the parameters of the model.
Initializing with a config file does not load the weights associated with the model, only the
configuration. Check out the <a href="/docs/transformers/pr_37350/en/main_classes/model#transformers.FlaxPreTrainedModel.from_pretrained">from_pretrained()</a> method to load the model weights.`,name:"config"},{anchor:"transformers.FlaxCLIPModel.dtype",description:`<strong>dtype</strong> (<code>jax.numpy.dtype</code>, <em>optional</em>, defaults to <code>jax.numpy.float32</code>) &#x2014;
The data type of the computation. Can be one of <code>jax.numpy.float32</code>, <code>jax.numpy.float16</code> (on GPUs) and
<code>jax.numpy.bfloat16</code> (on TPUs).</p>
<p>This can be used to enable mixed-precision training or half-precision inference on GPUs or TPUs. If
specified all the computation will be performed with the given <code>dtype</code>.</p>
<p><strong>Note that this only specifies the dtype of the computation and does not influence the dtype of model
parameters.</strong></p>
<p>If you wish to change the dtype of the model parameters, see <a href="/docs/transformers/pr_37350/en/main_classes/model#transformers.FlaxPreTrainedModel.to_fp16">to_fp16()</a> and
<a href="/docs/transformers/pr_37350/en/main_classes/model#transformers.FlaxPreTrainedModel.to_bf16">to_bf16()</a>.`,name:"dtype"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_flax_clip.py#L1262"}}),pt=new J({props:{name:"__call__",anchor:"transformers.FlaxCLIPModel.__call__",parameters:[{name:"input_ids",val:""},{name:"pixel_values",val:""},{name:"attention_mask",val:" = None"},{name:"position_ids",val:" = None"},{name:"params",val:": dict = None"},{name:"dropout_rng",val:": <function PRNGKey at 0x7f86b48be440> = None"},{name:"train",val:": bool = False"},{name:"output_attentions",val:": typing.Optional[bool] = None"},{name:"output_hidden_states",val:": typing.Optional[bool] = None"},{name:"return_dict",val:": typing.Optional[bool] = None"}],parametersDescription:[{anchor:"transformers.FlaxCLIPModel.__call__.input_ids",description:`<strong>input_ids</strong> (<code>numpy.ndarray</code> of shape <code>(batch_size, sequence_length)</code>) &#x2014;
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
it.</p>
<p>Indices can be obtained using <a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_37350/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/pr_37350/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.FlaxCLIPModel.__call__.attention_mask",description:`<strong>attention_mask</strong> (<code>numpy.ndarray</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.FlaxCLIPModel.__call__.position_ids",description:`<strong>position_ids</strong> (<code>numpy.ndarray</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.FlaxCLIPModel.__call__.pixel_values",description:`<strong>pixel_values</strong> (<code>numpy.ndarray</code> of shape <code>(batch_size, num_channels, height, width)</code>) &#x2014;
Pixel values. Padding will be ignored by default should you provide it. Pixel values can be obtained using
<a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoImageProcessor">AutoImageProcessor</a>. See <a href="/docs/transformers/pr_37350/en/model_doc/vilt#transformers.ViltFeatureExtractor.__call__">CLIPImageProcessor.<strong>call</strong>()</a> for details.`,name:"pixel_values"},{anchor:"transformers.FlaxCLIPModel.__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.FlaxCLIPModel.__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.FlaxCLIPModel.__call__.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_37350/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_flax_clip.py#L820",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <code>transformers.models.clip.modeling_flax_clip.FlaxCLIPOutput</code> 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 (<code>&lt;class 'transformers.models.clip.configuration_clip.CLIPConfig'&gt;</code>) and inputs.</p>
<ul>
<li><strong>logits_per_image:(<code>jnp.ndarray</code></strong> of shape <code>(image_batch_size, text_batch_size)</code>) — The scaled dot product scores between <code>image_embeds</code> and <code>text_embeds</code>. This represents the image-text
similarity scores.</li>
<li><strong>logits_per_text:(<code>jnp.ndarray</code></strong> of shape <code>(text_batch_size, image_batch_size)</code>) — The scaled dot product scores between <code>text_embeds</code> and <code>image_embeds</code>. This represents the text-image
similarity scores.</li>
<li><strong>text_embeds(<code>jnp.ndarray</code></strong> of shape <code>(batch_size, output_dim</code>) — The text embeddings obtained by applying the projection layer to the pooled output of
<a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.FlaxCLIPTextModel"
>FlaxCLIPTextModel</a>.</li>
<li><strong>image_embeds(<code>jnp.ndarray</code></strong> of shape <code>(batch_size, output_dim</code>) — The image embeddings obtained by applying the projection layer to the pooled output of
<a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.FlaxCLIPVisionModel"
>FlaxCLIPVisionModel</a>.</li>
<li><strong>text_model_output(<code>FlaxBaseModelOutputWithPooling</code>):</strong>
The output of the <a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.FlaxCLIPTextModel"
>FlaxCLIPTextModel</a>.</li>
<li><strong>vision_model_output(<code>FlaxBaseModelOutputWithPooling</code>):</strong>
The output of the <a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.FlaxCLIPVisionModel"
>FlaxCLIPVisionModel</a>.</li>
</ul>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><code>transformers.models.clip.modeling_flax_clip.FlaxCLIPOutput</code> or <code>tuple(torch.FloatTensor)</code></p>
`}}),Y=new ct({props:{$$slots:{default:[xs]},$$scope:{ctx:w}}}),mt=new Re({props:{anchor:"transformers.FlaxCLIPModel.__call__.example",$$slots:{default:[Ps]},$$scope:{ctx:w}}}),De=new J({props:{name:"get_text_features",anchor:"transformers.FlaxCLIPModel.get_text_features",parameters:[{name:"input_ids",val:""},{name:"attention_mask",val:" = None"},{name:"position_ids",val:" = None"},{name:"params",val:": dict = None"},{name:"dropout_rng",val:": <function PRNGKey at 0x7f86b48be440> = None"},{name:"train",val:" = False"}],parametersDescription:[{anchor:"transformers.FlaxCLIPModel.get_text_features.input_ids",description:`<strong>input_ids</strong> (<code>numpy.ndarray</code> of shape <code>(batch_size, sequence_length)</code>) &#x2014;
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you
provide it.</p>
<p>Indices can be obtained using <a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_37350/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/pr_37350/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"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_flax_clip.py#L865",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>The text embeddings obtained by applying
the projection layer to the pooled output of <a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.FlaxCLIPTextModel"
>FlaxCLIPTextModel</a>.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p>text_features (<code>jnp.ndarray</code> of shape <code>(batch_size, output_dim</code>)</p>
`}}),xe=new Re({props:{anchor:"transformers.FlaxCLIPModel.get_text_features.example",$$slots:{default:[Ls]},$$scope:{ctx:w}}}),ce=new J({props:{name:"get_image_features",anchor:"transformers.FlaxCLIPModel.get_image_features",parameters:[{name:"pixel_values",val:""},{name:"params",val:": dict = None"},{name:"dropout_rng",val:": <function PRNGKey at 0x7f86b48be440> = None"},{name:"train",val:" = False"}],parametersDescription:[{anchor:"transformers.FlaxCLIPModel.get_image_features.pixel_values",description:`<strong>pixel_values</strong> (<code>numpy.ndarray</code> of shape <code>(batch_size, num_channels, height, width)</code>) &#x2014;
Pixel values. Padding will be ignored by default should you provide it. Pixel values can be obtained
using <a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoImageProcessor">AutoImageProcessor</a>. See <a href="/docs/transformers/pr_37350/en/model_doc/vilt#transformers.ViltFeatureExtractor.__call__">CLIPImageProcessor.<strong>call</strong>()</a> for details.`,name:"pixel_values"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_flax_clip.py#L932",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>The image embeddings obtained by
applying the projection layer to the pooled output of <a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.FlaxCLIPVisionModel"
>FlaxCLIPVisionModel</a></p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p>image_features (<code>jnp.ndarray</code> of shape <code>(batch_size, output_dim</code>)</p>
`}}),Ge=new Re({props:{anchor:"transformers.FlaxCLIPModel.get_image_features.example",$$slots:{default:[ks]},$$scope:{ctx:w}}}),Ue=new ke({props:{title:"FlaxCLIPTextModel",local:"transformers.FlaxCLIPTextModel",headingTag:"h2"}}),Ie=new J({props:{name:"class transformers.FlaxCLIPTextModel",anchor:"transformers.FlaxCLIPTextModel",parameters:[{name:"config",val:": CLIPTextConfig"},{name:"input_shape",val:" = (1, 1)"},{name:"seed",val:": int = 0"},{name:"dtype",val:": dtype = <class 'jax.numpy.float32'>"},{name:"_do_init",val:": bool = True"},{name:"**kwargs",val:""}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_flax_clip.py#L1012"}}),V=new J({props:{name:"__call__",anchor:"transformers.FlaxCLIPTextModel.__call__",parameters:[{name:"input_ids",val:""},{name:"attention_mask",val:" = None"},{name:"position_ids",val:" = None"},{name:"params",val:": dict = None"},{name:"dropout_rng",val:": <function PRNGKey at 0x7f86b48be440> = None"},{name:"train",val:": bool = False"},{name:"output_attentions",val:": typing.Optional[bool] = None"},{name:"output_hidden_states",val:": typing.Optional[bool] = None"},{name:"return_dict",val:": typing.Optional[bool] = None"}],parametersDescription:[{anchor:"transformers.FlaxCLIPTextModel.__call__.input_ids",description:`<strong>input_ids</strong> (<code>numpy.ndarray</code> of shape <code>(batch_size, sequence_length)</code>) &#x2014;
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
it.</p>
<p>Indices can be obtained using <a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_37350/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/pr_37350/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.FlaxCLIPTextModel.__call__.attention_mask",description:`<strong>attention_mask</strong> (<code>numpy.ndarray</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.FlaxCLIPTextModel.__call__.position_ids",description:`<strong>position_ids</strong> (<code>numpy.ndarray</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.FlaxCLIPTextModel.__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.FlaxCLIPTextModel.__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.FlaxCLIPTextModel.__call__.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_37350/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_flax_clip.py#L665",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/pr_37350/en/main_classes/output#transformers.modeling_flax_outputs.FlaxBaseModelOutputWithPooling"
>transformers.modeling_flax_outputs.FlaxBaseModelOutputWithPooling</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 (<code>&lt;class 'transformers.models.clip.configuration_clip.CLIPTextConfig'&gt;</code>) and inputs.</p>
<ul>
<li>
<p><strong>last_hidden_state</strong> (<code>jnp.ndarray</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>pooler_output</strong> (<code>jnp.ndarray</code> of shape <code>(batch_size, hidden_size)</code>) — Last layer hidden-state of the first token of the sequence (classification token) further processed by a
Linear layer and a Tanh activation function. The Linear layer weights are trained from the next sentence
prediction (classification) objective during pretraining.</p>
</li>
<li>
<p><strong>hidden_states</strong> (<code>tuple(jnp.ndarray)</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>jnp.ndarray</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(jnp.ndarray)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>jnp.ndarray</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/pr_37350/en/main_classes/output#transformers.modeling_flax_outputs.FlaxBaseModelOutputWithPooling"
>transformers.modeling_flax_outputs.FlaxBaseModelOutputWithPooling</a> or <code>tuple(torch.FloatTensor)</code></p>
`}}),F=new ct({props:{$$slots:{default:[js]},$$scope:{ctx:w}}}),st=new Re({props:{anchor:"transformers.FlaxCLIPTextModel.__call__.example",$$slots:{default:[Js]},$$scope:{ctx:w}}}),_e=new ke({props:{title:"FlaxCLIPTextModelWithProjection",local:"transformers.FlaxCLIPTextModelWithProjection",headingTag:"h2"}}),He=new J({props:{name:"class transformers.FlaxCLIPTextModelWithProjection",anchor:"transformers.FlaxCLIPTextModelWithProjection",parameters:[{name:"config",val:": CLIPTextConfig"},{name:"input_shape",val:" = (1, 1)"},{name:"seed",val:": int = 0"},{name:"dtype",val:": dtype = <class 'jax.numpy.float32'>"},{name:"_do_init",val:": bool = True"},{name:"**kwargs",val:""}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_flax_clip.py#L1083"}}),oe=new J({props:{name:"__call__",anchor:"transformers.FlaxCLIPTextModelWithProjection.__call__",parameters:[{name:"input_ids",val:""},{name:"attention_mask",val:" = None"},{name:"position_ids",val:" = None"},{name:"params",val:": dict = None"},{name:"dropout_rng",val:": <function PRNGKey at 0x7f86b48be440> = None"},{name:"train",val:": bool = False"},{name:"output_attentions",val:": typing.Optional[bool] = None"},{name:"output_hidden_states",val:": typing.Optional[bool] = None"},{name:"return_dict",val:": typing.Optional[bool] = None"}],parametersDescription:[{anchor:"transformers.FlaxCLIPTextModelWithProjection.__call__.input_ids",description:`<strong>input_ids</strong> (<code>numpy.ndarray</code> of shape <code>(batch_size, sequence_length)</code>) &#x2014;
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
it.</p>
<p>Indices can be obtained using <a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_37350/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
<a href="/docs/transformers/pr_37350/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.FlaxCLIPTextModelWithProjection.__call__.attention_mask",description:`<strong>attention_mask</strong> (<code>numpy.ndarray</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.FlaxCLIPTextModelWithProjection.__call__.position_ids",description:`<strong>position_ids</strong> (<code>numpy.ndarray</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.FlaxCLIPTextModelWithProjection.__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.FlaxCLIPTextModelWithProjection.__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.FlaxCLIPTextModelWithProjection.__call__.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_37350/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_flax_clip.py#L665",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <code>transformers.models.clip.modeling_flax_clip.FlaxCLIPTextModelOutput</code> 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 (<code>&lt;class 'transformers.models.clip.configuration_clip.CLIPTextConfig'&gt;</code>) and inputs.</p>
<ul>
<li>
<p><strong>text_embeds</strong> (<code>jnp.ndarray</code> of shape <code>(batch_size, output_dim</code>) — The text embeddings obtained by applying the projection layer to the pooled output of
<a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.FlaxCLIPTextModel"
>FlaxCLIPTextModel</a>.</p>
</li>
<li>
<p><strong>last_hidden_state</strong> (<code>jnp.ndarray</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(jnp.ndarray)</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>jnp.ndarray</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(jnp.ndarray)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>jnp.ndarray</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><code>transformers.models.clip.modeling_flax_clip.FlaxCLIPTextModelOutput</code> or <code>tuple(torch.FloatTensor)</code></p>
`}}),Q=new ct({props:{$$slots:{default:[zs]},$$scope:{ctx:w}}}),We=new Re({props:{anchor:"transformers.FlaxCLIPTextModelWithProjection.__call__.example",$$slots:{default:[Us]},$$scope:{ctx:w}}}),Pe=new ke({props:{title:"FlaxCLIPVisionModel",local:"transformers.FlaxCLIPVisionModel",headingTag:"h2"}}),pe=new J({props:{name:"class transformers.FlaxCLIPVisionModel",anchor:"transformers.FlaxCLIPVisionModel",parameters:[{name:"config",val:": CLIPVisionConfig"},{name:"input_shape",val:": typing.Optional[typing.Tuple] = None"},{name:"seed",val:": int = 0"},{name:"dtype",val:": dtype = <class 'jax.numpy.float32'>"},{name:"_do_init",val:": bool = True"},{name:"**kwargs",val:""}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_flax_clip.py#L1137"}}),C=new J({props:{name:"__call__",anchor:"transformers.FlaxCLIPVisionModel.__call__",parameters:[{name:"pixel_values",val:""},{name:"params",val:": dict = None"},{name:"dropout_rng",val:": <function PRNGKey at 0x7f86b48be440> = None"},{name:"train",val:": bool = False"},{name:"output_attentions",val:": typing.Optional[bool] = None"},{name:"output_hidden_states",val:": typing.Optional[bool] = None"},{name:"return_dict",val:": typing.Optional[bool] = None"}],parametersDescription:[{anchor:"transformers.FlaxCLIPVisionModel.__call__.pixel_values",description:`<strong>pixel_values</strong> (<code>numpy.ndarray</code> of shape <code>(batch_size, num_channels, height, width)</code>) &#x2014;
Pixel values. Padding will be ignored by default should you provide it. Pixel values can be obtained using
<a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoImageProcessor">AutoImageProcessor</a>. See <a href="/docs/transformers/pr_37350/en/model_doc/vilt#transformers.ViltFeatureExtractor.__call__">CLIPImageProcessor.<strong>call</strong>()</a> for details.`,name:"pixel_values"},{anchor:"transformers.FlaxCLIPVisionModel.__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.FlaxCLIPVisionModel.__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.FlaxCLIPVisionModel.__call__.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) &#x2014;
Whether or not to return a <a href="/docs/transformers/pr_37350/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/modeling_flax_clip.py#L745",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>A <a
  href="/docs/transformers/pr_37350/en/main_classes/output#transformers.modeling_flax_outputs.FlaxBaseModelOutputWithPooling"
>transformers.modeling_flax_outputs.FlaxBaseModelOutputWithPooling</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 (<code>&lt;class 'transformers.models.clip.configuration_clip.CLIPVisionConfig'&gt;</code>) and inputs.</p>
<ul>
<li>
<p><strong>last_hidden_state</strong> (<code>jnp.ndarray</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>pooler_output</strong> (<code>jnp.ndarray</code> of shape <code>(batch_size, hidden_size)</code>) — Last layer hidden-state of the first token of the sequence (classification token) further processed by a
Linear layer and a Tanh activation function. The Linear layer weights are trained from the next sentence
prediction (classification) objective during pretraining.</p>
</li>
<li>
<p><strong>hidden_states</strong> (<code>tuple(jnp.ndarray)</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>jnp.ndarray</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(jnp.ndarray)</code>, <em>optional</em>, returned when <code>output_attentions=True</code> is passed or when <code>config.output_attentions=True</code>) — Tuple of <code>jnp.ndarray</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/pr_37350/en/main_classes/output#transformers.modeling_flax_outputs.FlaxBaseModelOutputWithPooling"
>transformers.modeling_flax_outputs.FlaxBaseModelOutputWithPooling</a> or <code>tuple(torch.FloatTensor)</code></p>
`}}),Ze=new ct({props:{$$slots:{default:[Fs]},$$scope:{ctx:w}}}),ye=new Re({props:{anchor:"transformers.FlaxCLIPVisionModel.__call__.example",$$slots:{default:[Ws]},$$scope:{ctx:w}}}),{c(){h(e.$$.fragment),d=a(),o=c("div"),h(n.$$.fragment),m=a(),t=c("p"),t.innerHTML=v,Me=a(),X=c("p"),X.innerHTML=$e,bt=a(),W=c("p"),W.textContent=H,me=a(),z=c("ul"),z.innerHTML=ae,tt=a(),G=c("div"),h(pt.$$.fragment),fe=a(),Je=c("p"),Je.innerHTML=K,ze=a(),h(Y.$$.fragment),ue=a(),h(mt.$$.fragment),ot=a(),ie=c("div"),h(De.$$.fragment),Z=a(),h(xe.$$.fragment),ee=a(),de=c("div"),h(ce.$$.fragment),Be=a(),h(Ge.$$.fragment),ft=a(),h(Ue.$$.fragment),Xe=a(),N=c("div"),h(Ie.$$.fragment),le=a(),he=c("div"),h(V.$$.fragment),nt=a(),ve=c("p"),ve.innerHTML=Ce,wt=a(),h(F.$$.fragment),ge=a(),h(st.$$.fragment),Ne=a(),h(_e.$$.fragment),Qe=a(),te=c("div"),h(He.$$.fragment),rt=a(),S=c("div"),h(oe.$$.fragment),Ae=a(),D=c("p"),D.innerHTML=ne,Fe=a(),h(Q.$$.fragment),B=a(),h(We.$$.fragment),at=a(),h(Pe.$$.fragment),Vt=a(),be=c("div"),h(pe.$$.fragment),Oe=a(),se=c("div"),h(C.$$.fragment),P=a(),A=c("p"),A.innerHTML=q,O=a(),h(Ze.$$.fragment),Te=a(),h(ye.$$.fragment),this.h()},l(I){g(e.$$.fragment,I),d=i(I),o=p(I,"DIV",{class:!0});var L=k(o);g(n.$$.fragment,L),m=i(L),t=p(L,"P",{"data-svelte-h":!0}),M(t)!=="svelte-9jczwz"&&(t.innerHTML=v),Me=i(L),X=p(L,"P",{"data-svelte-h":!0}),M(X)!=="svelte-10nfsf3"&&(X.innerHTML=$e),bt=i(L),W=p(L,"P",{"data-svelte-h":!0}),M(W)!=="svelte-1pplc4a"&&(W.textContent=H),me=i(L),z=p(L,"UL",{"data-svelte-h":!0}),M(z)!=="svelte-1w7z84m"&&(z.innerHTML=ae),tt=i(L),G=p(L,"DIV",{class:!0});var E=k(G);g(pt.$$.fragment,E),fe=i(E),Je=p(E,"P",{"data-svelte-h":!0}),M(Je)!=="svelte-qhvesh"&&(Je.innerHTML=K),ze=i(E),g(Y.$$.fragment,E),ue=i(E),g(mt.$$.fragment,E),E.forEach(l),ot=i(L),ie=p(L,"DIV",{class:!0});var it=k(ie);g(De.$$.fragment,it),Z=i(it),g(xe.$$.fragment,it),it.forEach(l),ee=i(L),de=p(L,"DIV",{class:!0});var qe=k(de);g(ce.$$.fragment,qe),Be=i(qe),g(Ge.$$.fragment,qe),qe.forEach(l),L.forEach(l),ft=i(I),g(Ue.$$.fragment,I),Xe=i(I),N=p(I,"DIV",{class:!0});var $t=k(N);g(Ie.$$.fragment,$t),le=i($t),he=p($t,"DIV",{class:!0});var R=k(he);g(V.$$.fragment,R),nt=i(R),ve=p(R,"P",{"data-svelte-h":!0}),M(ve)!=="svelte-1v5d1sw"&&(ve.innerHTML=Ce),wt=i(R),g(F.$$.fragment,R),ge=i(R),g(st.$$.fragment,R),R.forEach(l),$t.forEach(l),Ne=i(I),g(_e.$$.fragment,I),Qe=i(I),te=p(I,"DIV",{class:!0});var Ee=k(te);g(He.$$.fragment,Ee),rt=i(Ee),S=p(Ee,"DIV",{class:!0});var ut=k(S);g(oe.$$.fragment,ut),Ae=i(ut),D=p(ut,"P",{"data-svelte-h":!0}),M(D)!=="svelte-1v5d1sw"&&(D.innerHTML=ne),Fe=i(ut),g(Q.$$.fragment,ut),B=i(ut),g(We.$$.fragment,ut),ut.forEach(l),Ee.forEach(l),at=i(I),g(Pe.$$.fragment,I),Vt=i(I),be=p(I,"DIV",{class:!0});var ht=k(be);g(pe.$$.fragment,ht),Oe=i(ht),se=p(ht,"DIV",{class:!0});var Ye=k(se);g(C.$$.fragment,Ye),P=i(Ye),A=p(Ye,"P",{"data-svelte-h":!0}),M(A)!=="svelte-tkfbrd"&&(A.innerHTML=q),O=i(Ye),g(Ze.$$.fragment,Ye),Te=i(Ye),g(ye.$$.fragment,Ye),Ye.forEach(l),ht.forEach(l),this.h()},h(){j(G,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(ie,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(de,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(o,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(he,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(N,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(S,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(te,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(se,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(be,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8")},m(I,L){_(e,I,L),u(I,d,L),u(I,o,L),_(n,o,null),s(o,m),s(o,t),s(o,Me),s(o,X),s(o,bt),s(o,W),s(o,me),s(o,z),s(o,tt),s(o,G),_(pt,G,null),s(G,fe),s(G,Je),s(G,ze),_(Y,G,null),s(G,ue),_(mt,G,null),s(o,ot),s(o,ie),_(De,ie,null),s(ie,Z),_(xe,ie,null),s(o,ee),s(o,de),_(ce,de,null),s(de,Be),_(Ge,de,null),u(I,ft,L),_(Ue,I,L),u(I,Xe,L),u(I,N,L),_(Ie,N,null),s(N,le),s(N,he),_(V,he,null),s(he,nt),s(he,ve),s(he,wt),_(F,he,null),s(he,ge),_(st,he,null),u(I,Ne,L),_(_e,I,L),u(I,Qe,L),u(I,te,L),_(He,te,null),s(te,rt),s(te,S),_(oe,S,null),s(S,Ae),s(S,D),s(S,Fe),_(Q,S,null),s(S,B),_(We,S,null),u(I,at,L),_(Pe,I,L),u(I,Vt,L),u(I,be,L),_(pe,be,null),s(be,Oe),s(be,se),_(C,se,null),s(se,P),s(se,A),s(se,O),_(Ze,se,null),s(se,Te),_(ye,se,null),Ve=!0},p(I,L){const E={};L&2&&(E.$$scope={dirty:L,ctx:I}),Y.$set(E);const it={};L&2&&(it.$$scope={dirty:L,ctx:I}),mt.$set(it);const qe={};L&2&&(qe.$$scope={dirty:L,ctx:I}),xe.$set(qe);const $t={};L&2&&($t.$$scope={dirty:L,ctx:I}),Ge.$set($t);const R={};L&2&&(R.$$scope={dirty:L,ctx:I}),F.$set(R);const Ee={};L&2&&(Ee.$$scope={dirty:L,ctx:I}),st.$set(Ee);const ut={};L&2&&(ut.$$scope={dirty:L,ctx:I}),Q.$set(ut);const ht={};L&2&&(ht.$$scope={dirty:L,ctx:I}),We.$set(ht);const Ye={};L&2&&(Ye.$$scope={dirty:L,ctx:I}),Ze.$set(Ye);const To={};L&2&&(To.$$scope={dirty:L,ctx:I}),ye.$set(To)},i(I){Ve||(b(e.$$.fragment,I),b(n.$$.fragment,I),b(pt.$$.fragment,I),b(Y.$$.fragment,I),b(mt.$$.fragment,I),b(De.$$.fragment,I),b(xe.$$.fragment,I),b(ce.$$.fragment,I),b(Ge.$$.fragment,I),b(Ue.$$.fragment,I),b(Ie.$$.fragment,I),b(V.$$.fragment,I),b(F.$$.fragment,I),b(st.$$.fragment,I),b(_e.$$.fragment,I),b(He.$$.fragment,I),b(oe.$$.fragment,I),b(Q.$$.fragment,I),b(We.$$.fragment,I),b(Pe.$$.fragment,I),b(pe.$$.fragment,I),b(C.$$.fragment,I),b(Ze.$$.fragment,I),b(ye.$$.fragment,I),Ve=!0)},o(I){T(e.$$.fragment,I),T(n.$$.fragment,I),T(pt.$$.fragment,I),T(Y.$$.fragment,I),T(mt.$$.fragment,I),T(De.$$.fragment,I),T(xe.$$.fragment,I),T(ce.$$.fragment,I),T(Ge.$$.fragment,I),T(Ue.$$.fragment,I),T(Ie.$$.fragment,I),T(V.$$.fragment,I),T(F.$$.fragment,I),T(st.$$.fragment,I),T(_e.$$.fragment,I),T(He.$$.fragment,I),T(oe.$$.fragment,I),T(Q.$$.fragment,I),T(We.$$.fragment,I),T(Pe.$$.fragment,I),T(pe.$$.fragment,I),T(C.$$.fragment,I),T(Ze.$$.fragment,I),T(ye.$$.fragment,I),Ve=!1},d(I){I&&(l(d),l(o),l(ft),l(Xe),l(N),l(Ne),l(Qe),l(te),l(at),l(Vt),l(be)),y(e,I),y(n),y(pt),y(Y),y(mt),y(De),y(xe),y(ce),y(Ge),y(Ue,I),y(Ie),y(V),y(F),y(st),y(_e,I),y(He),y(oe),y(Q),y(We),y(Pe,I),y(pe),y(C),y(Ze),y(ye)}}}function Vs(w){let e,d;return e=new $n({props:{$$slots:{default:[Zs]},$$scope:{ctx:w}}}),{c(){h(e.$$.fragment)},l(o){g(e.$$.fragment,o)},m(o,n){_(e,o,n),d=!0},p(o,n){const m={};n&2&&(m.$$scope={dirty:n,ctx:o}),e.$set(m)},i(o){d||(b(e.$$.fragment,o),d=!0)},o(o){T(e.$$.fragment,o),d=!1},d(o){y(e,o)}}}function Bs(w){let e,d,o,n,m,t=`<div class="flex flex-wrap space-x-1"><img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&amp;logo=pytorch&amp;logoColor=white"/> <img alt="TensorFlow" src="https://img.shields.io/badge/TensorFlow-FF6F00?style=flat&amp;logo=tensorflow&amp;logoColor=white"/> <img alt="Flax" src="https://img.shields.io/badge/Flax-29a79b.svg?style=flat&amp;logo=data:image/png;base64,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
        "/> <img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat"/> <img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&amp;logo=pytorch&amp;logoColor=white"/></div>`,v,Me,X,$e,bt='<a href="https://huggingface.co/papers/2103.00020" rel="nofollow">CLIP</a> is a is a multimodal vision and language model motivated by overcoming the fixed number of object categories when training a computer vision model. CLIP learns about images directly from raw text by jointly training on 400M (image, text) pairs. Pretraining on this scale enables zero-shot transfer to downstream tasks. CLIP uses an image encoder and text encoder to get visual features and text features. Both features are projected to a latent space with the same number of dimensions and their dot product gives a similarity score.',W,H,me='You can find all the original CLIP checkpoints under the <a href="https://huggingface.co/openai?search_models=clip" rel="nofollow">OpenAI</a> organization.',z,ae,tt,G,pt='The example below demonstrates how to calculate similarity scores between multiple text descriptions and an image with <a href="/docs/transformers/pr_37350/en/main_classes/pipelines#transformers.Pipeline">Pipeline</a> or the <a href="/docs/transformers/pr_37350/en/model_doc/auto#transformers.AutoModel">AutoModel</a> class.',fe,Je,K,ze,Y,ue,mt='<li>Use <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPImageProcessor">CLIPImageProcessor</a> to resize (or rescale) and normalizes images for the model.</li>',ot,ie,De,Z,xe,ee,de,ce=`<a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPConfig">CLIPConfig</a> is the configuration class to store the configuration of a <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPModel">CLIPModel</a>. It is used to instantiate
a CLIP model according to the specified arguments, defining the text model and vision model configs. Instantiating
a configuration with the defaults will yield a similar configuration to that of the CLIP
<a href="https://huggingface.co/openai/clip-vit-base-patch32" rel="nofollow">openai/clip-vit-base-patch32</a> architecture.`,Be,Ge,ft=`Configuration objects inherit from <a href="/docs/transformers/pr_37350/en/main_classes/configuration#transformers.PretrainedConfig">PretrainedConfig</a> and can be used to control the model outputs. Read the
documentation from <a href="/docs/transformers/pr_37350/en/main_classes/configuration#transformers.PretrainedConfig">PretrainedConfig</a> for more information.`,Ue,Xe,N,Ie,le,he,V,nt=`Instantiate a <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPConfig">CLIPConfig</a> (or a derived class) from clip text model configuration and clip vision model
configuration.`,ve,Ce,wt,F,ge,st,Ne,_e=`This is the configuration class to store the configuration of a <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPTextModel">CLIPTextModel</a>. It is used to instantiate a CLIP
text encoder according to the specified arguments, defining the model architecture. Instantiating a configuration
with the defaults will yield a similar configuration to that of the text encoder of the CLIP
<a href="https://huggingface.co/openai/clip-vit-base-patch32" rel="nofollow">openai/clip-vit-base-patch32</a> architecture.`,Qe,te,He=`Configuration objects inherit from <a href="/docs/transformers/pr_37350/en/main_classes/configuration#transformers.PretrainedConfig">PretrainedConfig</a> and can be used to control the model outputs. Read the
documentation from <a href="/docs/transformers/pr_37350/en/main_classes/configuration#transformers.PretrainedConfig">PretrainedConfig</a> for more information.`,rt,S,oe,Ae,D,ne,Fe,Q,B,We=`This is the configuration class to store the configuration of a <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPVisionModel">CLIPVisionModel</a>. It is used to instantiate a
CLIP vision encoder according to the specified arguments, defining the model architecture. Instantiating a
configuration with the defaults will yield a similar configuration to that of the vision encoder of the CLIP
<a href="https://huggingface.co/openai/clip-vit-base-patch32" rel="nofollow">openai/clip-vit-base-patch32</a> architecture.`,at,Pe,Vt=`Configuration objects inherit from <a href="/docs/transformers/pr_37350/en/main_classes/configuration#transformers.PretrainedConfig">PretrainedConfig</a> and can be used to control the model outputs. Read the
documentation from <a href="/docs/transformers/pr_37350/en/main_classes/configuration#transformers.PretrainedConfig">PretrainedConfig</a> for more information.`,be,pe,Oe,se,C,P,A,q,O,Ze="Construct a CLIP tokenizer. Based on byte-level Byte-Pair-Encoding.",Te,ye,Ve=`This tokenizer inherits from <a href="/docs/transformers/pr_37350/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.`,I,L,E,it,qe,$t=`Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
adding special tokens. A CLIP sequence has the following format:`,R,Ee,ut="<li>single sequence: <code>&lt;|startoftext|&gt; X &lt;|endoftext|&gt;</code></li>",ht,Ye,To="Pairs of sequences are not the expected use case, but they will be handled without a separator.",io,Ot,Kt,lo,yo,Xo=`Retrieve 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> method.`,yt,Mt,eo,co,Mo,Ho=`Create a mask from the two sequences passed. CLIP does not make use of token type ids, therefore a list of
zeros is returned.`,Dt,po,It,Io,xt,vo,re,Pt,Jo,Rt,on=`Construct a “fast” CLIP tokenizer (backed by HuggingFace’s <em>tokenizers</em> library). Based on byte-level
Byte-Pair-Encoding.`,zo,Xt,nn=`This tokenizer inherits from <a href="/docs/transformers/pr_37350/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.`,Uo,we,Lt,Fo,Ht,sn=`Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
adding special tokens. A CLIP sequence has the following format:`,Wo,kt,Yo="<li>single sequence: <code>&lt;|startoftext|&gt; X &lt;|endoftext|&gt;</code></li>",Qt,to,mo="Pairs of sequences are not the expected use case, but they will be handled without a separator.",wo,Le,jt,Zo,Yt,rn=`Create a mask from the two sequences passed. CLIP does not make use of token type ids, therefore a list of
zeros is returned.`,$o,Bt,Vo,Jt,Gt,So,fo,zt="Constructs a CLIP image processor.",oo,At,Nt,Do,uo,no="Preprocess an image or batch of images.",xo,vt,Co,f,x,Ut,gt,qt="Constructs a fast CLIP image processor.",Ft,Ke,Se,Wt,et,Et="Preprocess an image or batch of images.",Ct,lt,_t,dt,ho,Bo,go,ln,Zt,Go,yn,Qo,xn="Constructs a CLIP processor which wraps a CLIP image processor and a CLIP tokenizer into a single processor.",Mn,Ao,Pn=`<a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPProcessor">CLIPProcessor</a> offers all the functionalities of <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPImageProcessor">CLIPImageProcessor</a> and <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPTokenizerFast">CLIPTokenizerFast</a>. See the
<code>__call__()</code> and <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPProcessor.decode">decode()</a> for more information.`,In,Po,No,vn,Oo,Ln=`This method forwards all its arguments to CLIPTokenizerFast’s <a href="/docs/transformers/pr_37350/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.batch_decode">batch_decode()</a>. Please
refer to the docstring of this method for more information.`,Cn,Lo,qo,wn,Ko,kn=`This method forwards all its arguments to CLIPTokenizerFast’s <a href="/docs/transformers/pr_37350/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.decode">decode()</a>. Please refer to
the docstring of this method for more information.`,dn,ko,cn,Eo,pn,an,mn;return Me=new ke({props:{title:"CLIP",local:"clip",headingTag:"h1"}}),ae=new ct({props:{warning:!1,$$slots:{default:[En]},$$scope:{ctx:w}}}),Je=new qn({props:{id:"usage",options:["Pipeline","AutoModel"],$$slots:{default:[Hn]},$$scope:{ctx:w}}}),ze=new ke({props:{title:"Notes",local:"notes",headingTag:"h2"}}),ie=new ke({props:{title:"CLIPConfig",local:"transformers.CLIPConfig",headingTag:"h2"}}),xe=new J({props:{name:"class transformers.CLIPConfig",anchor:"transformers.CLIPConfig",parameters:[{name:"text_config",val:" = None"},{name:"vision_config",val:" = None"},{name:"projection_dim",val:" = 512"},{name:"logit_scale_init_value",val:" = 2.6592"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.CLIPConfig.text_config",description:`<strong>text_config</strong> (<code>dict</code>, <em>optional</em>) &#x2014;
Dictionary of configuration options used to initialize <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPTextConfig">CLIPTextConfig</a>.`,name:"text_config"},{anchor:"transformers.CLIPConfig.vision_config",description:`<strong>vision_config</strong> (<code>dict</code>, <em>optional</em>) &#x2014;
Dictionary of configuration options used to initialize <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPVisionConfig">CLIPVisionConfig</a>.`,name:"vision_config"},{anchor:"transformers.CLIPConfig.projection_dim",description:`<strong>projection_dim</strong> (<code>int</code>, <em>optional</em>, defaults to 512) &#x2014;
Dimensionality of text and vision projection layers.`,name:"projection_dim"},{anchor:"transformers.CLIPConfig.logit_scale_init_value",description:`<strong>logit_scale_init_value</strong> (<code>float</code>, <em>optional</em>, defaults to 2.6592) &#x2014;
The initial value of the <em>logit_scale</em> parameter. Default is used as per the original CLIP implementation.`,name:"logit_scale_init_value"},{anchor:"transformers.CLIPConfig.kwargs",description:`<strong>kwargs</strong> (<em>optional</em>) &#x2014;
Dictionary of keyword arguments.`,name:"kwargs"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/configuration_clip.py#L226"}}),Xe=new Re({props:{anchor:"transformers.CLIPConfig.example",$$slots:{default:[Yn]},$$scope:{ctx:w}}}),le=new J({props:{name:"from_text_vision_configs",anchor:"transformers.CLIPConfig.from_text_vision_configs",parameters:[{name:"text_config",val:": CLIPTextConfig"},{name:"vision_config",val:": CLIPVisionConfig"},{name:"**kwargs",val:""}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/configuration_clip.py#L363",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>An instance of a configuration object</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><a
  href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPConfig"
>CLIPConfig</a></p>
`}}),Ce=new ke({props:{title:"CLIPTextConfig",local:"transformers.CLIPTextConfig",headingTag:"h2"}}),ge=new J({props:{name:"class transformers.CLIPTextConfig",anchor:"transformers.CLIPTextConfig",parameters:[{name:"vocab_size",val:" = 49408"},{name:"hidden_size",val:" = 512"},{name:"intermediate_size",val:" = 2048"},{name:"projection_dim",val:" = 512"},{name:"num_hidden_layers",val:" = 12"},{name:"num_attention_heads",val:" = 8"},{name:"max_position_embeddings",val:" = 77"},{name:"hidden_act",val:" = 'quick_gelu'"},{name:"layer_norm_eps",val:" = 1e-05"},{name:"attention_dropout",val:" = 0.0"},{name:"initializer_range",val:" = 0.02"},{name:"initializer_factor",val:" = 1.0"},{name:"pad_token_id",val:" = 1"},{name:"bos_token_id",val:" = 49406"},{name:"eos_token_id",val:" = 49407"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.CLIPTextConfig.vocab_size",description:`<strong>vocab_size</strong> (<code>int</code>, <em>optional</em>, defaults to 49408) &#x2014;
Vocabulary size of the CLIP text model. Defines the number of different tokens that can be represented by
the <code>inputs_ids</code> passed when calling <a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPModel">CLIPModel</a>.`,name:"vocab_size"},{anchor:"transformers.CLIPTextConfig.hidden_size",description:`<strong>hidden_size</strong> (<code>int</code>, <em>optional</em>, defaults to 512) &#x2014;
Dimensionality of the encoder layers and the pooler layer.`,name:"hidden_size"},{anchor:"transformers.CLIPTextConfig.intermediate_size",description:`<strong>intermediate_size</strong> (<code>int</code>, <em>optional</em>, defaults to 2048) &#x2014;
Dimensionality of the &#x201C;intermediate&#x201D; (i.e., feed-forward) layer in the Transformer encoder.`,name:"intermediate_size"},{anchor:"transformers.CLIPTextConfig.projection_dim",description:`<strong>projection_dim</strong> (<code>int</code>, <em>optional</em>, defaults to 512) &#x2014;
Dimensionality of text and vision projection layers.`,name:"projection_dim"},{anchor:"transformers.CLIPTextConfig.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.CLIPTextConfig.num_attention_heads",description:`<strong>num_attention_heads</strong> (<code>int</code>, <em>optional</em>, defaults to 8) &#x2014;
Number of attention heads for each attention layer in the Transformer encoder.`,name:"num_attention_heads"},{anchor:"transformers.CLIPTextConfig.max_position_embeddings",description:`<strong>max_position_embeddings</strong> (<code>int</code>, <em>optional</em>, defaults to 77) &#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.CLIPTextConfig.hidden_act",description:`<strong>hidden_act</strong> (<code>str</code> or <code>function</code>, <em>optional</em>, defaults to <code>&quot;quick_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;selu&quot;</code> and <code>&quot;gelu_new&quot;</code> <code>&quot;quick_gelu&quot;</code> are supported.`,name:"hidden_act"},{anchor:"transformers.CLIPTextConfig.layer_norm_eps",description:`<strong>layer_norm_eps</strong> (<code>float</code>, <em>optional</em>, defaults to 1e-05) &#x2014;
The epsilon used by the layer normalization layers.`,name:"layer_norm_eps"},{anchor:"transformers.CLIPTextConfig.attention_dropout",description:`<strong>attention_dropout</strong> (<code>float</code>, <em>optional</em>, defaults to 0.0) &#x2014;
The dropout ratio for the attention probabilities.`,name:"attention_dropout"},{anchor:"transformers.CLIPTextConfig.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.CLIPTextConfig.initializer_factor",description:`<strong>initializer_factor</strong> (<code>float</code>, <em>optional</em>, defaults to 1.0) &#x2014;
A factor for initializing all weight matrices (should be kept to 1, used internally for initialization
testing).`,name:"initializer_factor"},{anchor:"transformers.CLIPTextConfig.pad_token_id",description:`<strong>pad_token_id</strong> (<code>int</code>, <em>optional</em>, defaults to 1) &#x2014;
Padding token id.`,name:"pad_token_id"},{anchor:"transformers.CLIPTextConfig.bos_token_id",description:`<strong>bos_token_id</strong> (<code>int</code>, <em>optional</em>, defaults to 49406) &#x2014;
Beginning of stream token id.`,name:"bos_token_id"},{anchor:"transformers.CLIPTextConfig.eos_token_id",description:`<strong>eos_token_id</strong> (<code>int</code>, <em>optional</em>, defaults to 49407) &#x2014;
End of stream token id.`,name:"eos_token_id"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/configuration_clip.py#L33"}}),S=new Re({props:{anchor:"transformers.CLIPTextConfig.example",$$slots:{default:[Sn]},$$scope:{ctx:w}}}),Ae=new ke({props:{title:"CLIPVisionConfig",local:"transformers.CLIPVisionConfig",headingTag:"h2"}}),Fe=new J({props:{name:"class transformers.CLIPVisionConfig",anchor:"transformers.CLIPVisionConfig",parameters:[{name:"hidden_size",val:" = 768"},{name:"intermediate_size",val:" = 3072"},{name:"projection_dim",val:" = 512"},{name:"num_hidden_layers",val:" = 12"},{name:"num_attention_heads",val:" = 12"},{name:"num_channels",val:" = 3"},{name:"image_size",val:" = 224"},{name:"patch_size",val:" = 32"},{name:"hidden_act",val:" = 'quick_gelu'"},{name:"layer_norm_eps",val:" = 1e-05"},{name:"attention_dropout",val:" = 0.0"},{name:"initializer_range",val:" = 0.02"},{name:"initializer_factor",val:" = 1.0"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.CLIPVisionConfig.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.CLIPVisionConfig.intermediate_size",description:`<strong>intermediate_size</strong> (<code>int</code>, <em>optional</em>, defaults to 3072) &#x2014;
Dimensionality of the &#x201C;intermediate&#x201D; (i.e., feed-forward) layer in the Transformer encoder.`,name:"intermediate_size"},{anchor:"transformers.CLIPVisionConfig.projection_dim",description:`<strong>projection_dim</strong> (<code>int</code>, <em>optional</em>, defaults to 512) &#x2014;
Dimensionality of text and vision projection layers.`,name:"projection_dim"},{anchor:"transformers.CLIPVisionConfig.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.CLIPVisionConfig.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.CLIPVisionConfig.num_channels",description:`<strong>num_channels</strong> (<code>int</code>, <em>optional</em>, defaults to 3) &#x2014;
The number of input channels.`,name:"num_channels"},{anchor:"transformers.CLIPVisionConfig.image_size",description:`<strong>image_size</strong> (<code>int</code>, <em>optional</em>, defaults to 224) &#x2014;
The size (resolution) of each image.`,name:"image_size"},{anchor:"transformers.CLIPVisionConfig.patch_size",description:`<strong>patch_size</strong> (<code>int</code>, <em>optional</em>, defaults to 32) &#x2014;
The size (resolution) of each patch.`,name:"patch_size"},{anchor:"transformers.CLIPVisionConfig.hidden_act",description:`<strong>hidden_act</strong> (<code>str</code> or <code>function</code>, <em>optional</em>, defaults to <code>&quot;quick_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;selu&quot;</code> and <code>&quot;gelu_new&quot;</code> <code>&quot;quick_gelu&quot;</code> are supported.`,name:"hidden_act"},{anchor:"transformers.CLIPVisionConfig.layer_norm_eps",description:`<strong>layer_norm_eps</strong> (<code>float</code>, <em>optional</em>, defaults to 1e-05) &#x2014;
The epsilon used by the layer normalization layers.`,name:"layer_norm_eps"},{anchor:"transformers.CLIPVisionConfig.attention_dropout",description:`<strong>attention_dropout</strong> (<code>float</code>, <em>optional</em>, defaults to 0.0) &#x2014;
The dropout ratio for the attention probabilities.`,name:"attention_dropout"},{anchor:"transformers.CLIPVisionConfig.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.CLIPVisionConfig.initializer_factor",description:`<strong>initializer_factor</strong> (<code>float</code>, <em>optional</em>, defaults to 1.0) &#x2014;
A factor for initializing all weight matrices (should be kept to 1, used internally for initialization
testing).`,name:"initializer_factor"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/configuration_clip.py#L134"}}),pe=new Re({props:{anchor:"transformers.CLIPVisionConfig.example",$$slots:{default:[Dn]},$$scope:{ctx:w}}}),se=new ke({props:{title:"CLIPTokenizer",local:"transformers.CLIPTokenizer",headingTag:"h2"}}),A=new J({props:{name:"class transformers.CLIPTokenizer",anchor:"transformers.CLIPTokenizer",parameters:[{name:"vocab_file",val:""},{name:"merges_file",val:""},{name:"errors",val:" = 'replace'"},{name:"unk_token",val:" = '<|endoftext|>'"},{name:"bos_token",val:" = '<|startoftext|>'"},{name:"eos_token",val:" = '<|endoftext|>'"},{name:"pad_token",val:" = '<|endoftext|>'"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.CLIPTokenizer.vocab_file",description:`<strong>vocab_file</strong> (<code>str</code>) &#x2014;
Path to the vocabulary file.`,name:"vocab_file"},{anchor:"transformers.CLIPTokenizer.merges_file",description:`<strong>merges_file</strong> (<code>str</code>) &#x2014;
Path to the merges file.`,name:"merges_file"},{anchor:"transformers.CLIPTokenizer.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.CLIPTokenizer.unk_token",description:`<strong>unk_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;&lt;|endoftext|&gt;&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.CLIPTokenizer.bos_token",description:`<strong>bos_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;&lt;|startoftext|&gt;&quot;</code>) &#x2014;
The beginning of sequence token.`,name:"bos_token"},{anchor:"transformers.CLIPTokenizer.eos_token",description:`<strong>eos_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;&lt;|endoftext|&gt;&quot;</code>) &#x2014;
The end of sequence token.`,name:"eos_token"},{anchor:"transformers.CLIPTokenizer.pad_token",description:`<strong>pad_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;&lt;|endoftext|&gt;&quot;</code>) &#x2014;
The token used for padding, for example when batching sequences of different lengths.`,name:"pad_token"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/tokenization_clip.py#L254"}}),E=new J({props:{name:"build_inputs_with_special_tokens",anchor:"transformers.CLIPTokenizer.build_inputs_with_special_tokens",parameters:[{name:"token_ids_0",val:": typing.List[int]"},{name:"token_ids_1",val:": typing.Optional[typing.List[int]] = None"}],parametersDescription:[{anchor:"transformers.CLIPTokenizer.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.CLIPTokenizer.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/vr_37350/src/transformers/models/clip/tokenization_clip.py#L339",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>
`}}),Kt=new J({props:{name:"get_special_tokens_mask",anchor:"transformers.CLIPTokenizer.get_special_tokens_mask",parameters:[{name:"token_ids_0",val:": typing.List[int]"},{name:"token_ids_1",val:": typing.Optional[typing.List[int]] = None"},{name:"already_has_special_tokens",val:": bool = False"}],parametersDescription:[{anchor:"transformers.CLIPTokenizer.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.CLIPTokenizer.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.CLIPTokenizer.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/vr_37350/src/transformers/models/clip/tokenization_clip.py#L366",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>
`}}),eo=new J({props:{name:"create_token_type_ids_from_sequences",anchor:"transformers.CLIPTokenizer.create_token_type_ids_from_sequences",parameters:[{name:"token_ids_0",val:": typing.List[int]"},{name:"token_ids_1",val:": typing.Optional[typing.List[int]] = None"}],parametersDescription:[{anchor:"transformers.CLIPTokenizer.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.CLIPTokenizer.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/vr_37350/src/transformers/models/clip/tokenization_clip.py#L394",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>List of zeros.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><code>List[int]</code></p>
`}}),It=new J({props:{name:"save_vocabulary",anchor:"transformers.CLIPTokenizer.save_vocabulary",parameters:[{name:"save_directory",val:": str"},{name:"filename_prefix",val:": typing.Optional[str] = None"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/tokenization_clip.py#L489"}}),xt=new ke({props:{title:"CLIPTokenizerFast",local:"transformers.CLIPTokenizerFast",headingTag:"h2"}}),Pt=new J({props:{name:"class transformers.CLIPTokenizerFast",anchor:"transformers.CLIPTokenizerFast",parameters:[{name:"vocab_file",val:" = None"},{name:"merges_file",val:" = None"},{name:"tokenizer_file",val:" = None"},{name:"unk_token",val:" = '<|endoftext|>'"},{name:"bos_token",val:" = '<|startoftext|>'"},{name:"eos_token",val:" = '<|endoftext|>'"},{name:"pad_token",val:" = '<|endoftext|>'"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.CLIPTokenizerFast.vocab_file",description:`<strong>vocab_file</strong> (<code>str</code>, <em>optional</em>) &#x2014;
Path to the vocabulary file.`,name:"vocab_file"},{anchor:"transformers.CLIPTokenizerFast.merges_file",description:`<strong>merges_file</strong> (<code>str</code>, <em>optional</em>) &#x2014;
Path to the merges file.`,name:"merges_file"},{anchor:"transformers.CLIPTokenizerFast.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.CLIPTokenizerFast.unk_token",description:`<strong>unk_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;&lt;|endoftext|&gt;&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.CLIPTokenizerFast.bos_token",description:`<strong>bos_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;&lt;|startoftext|&gt;&quot;</code>) &#x2014;
The beginning of sequence token.`,name:"bos_token"},{anchor:"transformers.CLIPTokenizerFast.eos_token",description:`<strong>eos_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;&lt;|endoftext|&gt;&quot;</code>) &#x2014;
The end of sequence token.`,name:"eos_token"},{anchor:"transformers.CLIPTokenizerFast.pad_token",description:`<strong>pad_token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;&lt;|endoftext|&gt;&quot;</code>) &#x2014;
The token used for padding, for example when batching sequences of different lengths.`,name:"pad_token"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/tokenization_clip_fast.py#L31"}}),Lt=new J({props:{name:"build_inputs_with_special_tokens",anchor:"transformers.CLIPTokenizerFast.build_inputs_with_special_tokens",parameters:[{name:"token_ids_0",val:": typing.List[int]"},{name:"token_ids_1",val:": typing.Optional[typing.List[int]] = None"}],parametersDescription:[{anchor:"transformers.CLIPTokenizerFast.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.CLIPTokenizerFast.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/vr_37350/src/transformers/models/clip/tokenization_clip_fast.py#L109",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>
`}}),jt=new J({props:{name:"create_token_type_ids_from_sequences",anchor:"transformers.CLIPTokenizerFast.create_token_type_ids_from_sequences",parameters:[{name:"token_ids_0",val:": typing.List[int]"},{name:"token_ids_1",val:": typing.Optional[typing.List[int]] = None"}],parametersDescription:[{anchor:"transformers.CLIPTokenizerFast.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.CLIPTokenizerFast.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/vr_37350/src/transformers/models/clip/tokenization_clip_fast.py#L136",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>List of zeros.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><code>List[int]</code></p>
`}}),Bt=new ke({props:{title:"CLIPImageProcessor",local:"transformers.CLIPImageProcessor",headingTag:"h2"}}),Gt=new J({props:{name:"class transformers.CLIPImageProcessor",anchor:"transformers.CLIPImageProcessor",parameters:[{name:"do_resize",val:": bool = True"},{name:"size",val:": typing.Dict[str, int] = None"},{name:"resample",val:": Resampling = <Resampling.BICUBIC: 3>"},{name:"do_center_crop",val:": bool = True"},{name:"crop_size",val:": typing.Dict[str, int] = None"},{name:"do_rescale",val:": bool = True"},{name:"rescale_factor",val:": typing.Union[int, float] = 0.00392156862745098"},{name:"do_normalize",val:": bool = True"},{name:"image_mean",val:": typing.Union[float, typing.List[float], NoneType] = None"},{name:"image_std",val:": typing.Union[float, typing.List[float], NoneType] = None"},{name:"do_convert_rgb",val:": bool = True"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.CLIPImageProcessor.do_resize",description:`<strong>do_resize</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) &#x2014;
Whether to resize the image&#x2019;s (height, width) dimensions to the specified <code>size</code>. Can be overridden by
<code>do_resize</code> in the <code>preprocess</code> method.`,name:"do_resize"},{anchor:"transformers.CLIPImageProcessor.size",description:`<strong>size</strong> (<code>Dict[str, int]</code> <em>optional</em>, defaults to <code>{&quot;shortest_edge&quot; -- 224}</code>):
Size of the image after resizing. The shortest edge of the image is resized to size[&#x201C;shortest_edge&#x201D;], with
the longest edge resized to keep the input aspect ratio. Can be overridden by <code>size</code> in the <code>preprocess</code>
method.`,name:"size"},{anchor:"transformers.CLIPImageProcessor.resample",description:`<strong>resample</strong> (<code>PILImageResampling</code>, <em>optional</em>, defaults to <code>Resampling.BICUBIC</code>) &#x2014;
Resampling filter to use if resizing the image. Can be overridden by <code>resample</code> in the <code>preprocess</code> method.`,name:"resample"},{anchor:"transformers.CLIPImageProcessor.do_center_crop",description:`<strong>do_center_crop</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) &#x2014;
Whether to center crop the image to the specified <code>crop_size</code>. Can be overridden by <code>do_center_crop</code> in the
<code>preprocess</code> method.`,name:"do_center_crop"},{anchor:"transformers.CLIPImageProcessor.crop_size",description:`<strong>crop_size</strong> (<code>Dict[str, int]</code> <em>optional</em>, defaults to 224) &#x2014;
Size of the output image after applying <code>center_crop</code>. Can be overridden by <code>crop_size</code> in the <code>preprocess</code>
method.`,name:"crop_size"},{anchor:"transformers.CLIPImageProcessor.do_rescale",description:`<strong>do_rescale</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) &#x2014;
Whether to rescale the image by the specified scale <code>rescale_factor</code>. Can be overridden by <code>do_rescale</code> in
the <code>preprocess</code> method.`,name:"do_rescale"},{anchor:"transformers.CLIPImageProcessor.rescale_factor",description:`<strong>rescale_factor</strong> (<code>int</code> or <code>float</code>, <em>optional</em>, defaults to <code>1/255</code>) &#x2014;
Scale factor to use if rescaling the image. Can be overridden by <code>rescale_factor</code> in the <code>preprocess</code>
method.`,name:"rescale_factor"},{anchor:"transformers.CLIPImageProcessor.do_normalize",description:`<strong>do_normalize</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) &#x2014;
Whether to normalize the image. Can be overridden by <code>do_normalize</code> in the <code>preprocess</code> method.`,name:"do_normalize"},{anchor:"transformers.CLIPImageProcessor.image_mean",description:`<strong>image_mean</strong> (<code>float</code> or <code>List[float]</code>, <em>optional</em>, defaults to <code>[0.48145466, 0.4578275, 0.40821073]</code>) &#x2014;
Mean to use if normalizing the image. This is a float or list of floats the length of the number of
channels in the image. Can be overridden by the <code>image_mean</code> parameter in the <code>preprocess</code> method.`,name:"image_mean"},{anchor:"transformers.CLIPImageProcessor.image_std",description:`<strong>image_std</strong> (<code>float</code> or <code>List[float]</code>, <em>optional</em>, defaults to <code>[0.26862954, 0.26130258, 0.27577711]</code>) &#x2014;
Standard deviation to use if normalizing the image. This is a float or list of floats the length of the
number of channels in the image. Can be overridden by the <code>image_std</code> parameter in the <code>preprocess</code> method.
Can be overridden by the <code>image_std</code> parameter in the <code>preprocess</code> method.`,name:"image_std"},{anchor:"transformers.CLIPImageProcessor.do_convert_rgb",description:`<strong>do_convert_rgb</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) &#x2014;
Whether to convert the image to RGB.`,name:"do_convert_rgb"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/image_processing_clip.py#L52"}}),Nt=new J({props:{name:"preprocess",anchor:"transformers.CLIPImageProcessor.preprocess",parameters:[{name:"images",val:": typing.Union[ForwardRef('PIL.Image.Image'), numpy.ndarray, ForwardRef('torch.Tensor'), list['PIL.Image.Image'], list[numpy.ndarray], list['torch.Tensor']]"},{name:"do_resize",val:": typing.Optional[bool] = None"},{name:"size",val:": typing.Dict[str, int] = None"},{name:"resample",val:": Resampling = None"},{name:"do_center_crop",val:": typing.Optional[bool] = None"},{name:"crop_size",val:": typing.Optional[int] = None"},{name:"do_rescale",val:": typing.Optional[bool] = None"},{name:"rescale_factor",val:": typing.Optional[float] = None"},{name:"do_normalize",val:": typing.Optional[bool] = None"},{name:"image_mean",val:": typing.Union[float, typing.List[float], NoneType] = None"},{name:"image_std",val:": typing.Union[float, typing.List[float], NoneType] = None"},{name:"do_convert_rgb",val:": typing.Optional[bool] = None"},{name:"return_tensors",val:": typing.Union[str, transformers.utils.generic.TensorType, NoneType] = None"},{name:"data_format",val:": typing.Optional[transformers.image_utils.ChannelDimension] = <ChannelDimension.FIRST: 'channels_first'>"},{name:"input_data_format",val:": typing.Union[transformers.image_utils.ChannelDimension, str, NoneType] = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.CLIPImageProcessor.preprocess.images",description:`<strong>images</strong> (<code>ImageInput</code>) &#x2014;
Image to preprocess. Expects a single or batch of images with pixel values ranging from 0 to 255. If
passing in images with pixel values between 0 and 1, set <code>do_rescale=False</code>.`,name:"images"},{anchor:"transformers.CLIPImageProcessor.preprocess.do_resize",description:`<strong>do_resize</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>self.do_resize</code>) &#x2014;
Whether to resize the image.`,name:"do_resize"},{anchor:"transformers.CLIPImageProcessor.preprocess.size",description:`<strong>size</strong> (<code>Dict[str, int]</code>, <em>optional</em>, defaults to <code>self.size</code>) &#x2014;
Size of the image after resizing. Shortest edge of the image is resized to size[&#x201C;shortest_edge&#x201D;], with
the longest edge resized to keep the input aspect ratio.`,name:"size"},{anchor:"transformers.CLIPImageProcessor.preprocess.resample",description:`<strong>resample</strong> (<code>int</code>, <em>optional</em>, defaults to <code>self.resample</code>) &#x2014;
Resampling filter to use if resizing the image. This can be one of the enum <code>PILImageResampling</code>. Only
has an effect if <code>do_resize</code> is set to <code>True</code>.`,name:"resample"},{anchor:"transformers.CLIPImageProcessor.preprocess.do_center_crop",description:`<strong>do_center_crop</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>self.do_center_crop</code>) &#x2014;
Whether to center crop the image.`,name:"do_center_crop"},{anchor:"transformers.CLIPImageProcessor.preprocess.crop_size",description:`<strong>crop_size</strong> (<code>Dict[str, int]</code>, <em>optional</em>, defaults to <code>self.crop_size</code>) &#x2014;
Size of the center crop. Only has an effect if <code>do_center_crop</code> is set to <code>True</code>.`,name:"crop_size"},{anchor:"transformers.CLIPImageProcessor.preprocess.do_rescale",description:`<strong>do_rescale</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>self.do_rescale</code>) &#x2014;
Whether to rescale the image.`,name:"do_rescale"},{anchor:"transformers.CLIPImageProcessor.preprocess.rescale_factor",description:`<strong>rescale_factor</strong> (<code>float</code>, <em>optional</em>, defaults to <code>self.rescale_factor</code>) &#x2014;
Rescale factor to rescale the image by if <code>do_rescale</code> is set to <code>True</code>.`,name:"rescale_factor"},{anchor:"transformers.CLIPImageProcessor.preprocess.do_normalize",description:`<strong>do_normalize</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>self.do_normalize</code>) &#x2014;
Whether to normalize the image.`,name:"do_normalize"},{anchor:"transformers.CLIPImageProcessor.preprocess.image_mean",description:`<strong>image_mean</strong> (<code>float</code> or <code>List[float]</code>, <em>optional</em>, defaults to <code>self.image_mean</code>) &#x2014;
Image mean to use for normalization. Only has an effect if <code>do_normalize</code> is set to <code>True</code>.`,name:"image_mean"},{anchor:"transformers.CLIPImageProcessor.preprocess.image_std",description:`<strong>image_std</strong> (<code>float</code> or <code>List[float]</code>, <em>optional</em>, defaults to <code>self.image_std</code>) &#x2014;
Image standard deviation to use for normalization. Only has an effect if <code>do_normalize</code> is set to
<code>True</code>.`,name:"image_std"},{anchor:"transformers.CLIPImageProcessor.preprocess.do_convert_rgb",description:`<strong>do_convert_rgb</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>self.do_convert_rgb</code>) &#x2014;
Whether to convert the image to RGB.`,name:"do_convert_rgb"},{anchor:"transformers.CLIPImageProcessor.preprocess.return_tensors",description:`<strong>return_tensors</strong> (<code>str</code> or <code>TensorType</code>, <em>optional</em>) &#x2014;
The type of tensors to return. Can be one of:<ul>
<li>Unset: Return a list of <code>np.ndarray</code>.</li>
<li><code>TensorType.TENSORFLOW</code> or <code>&apos;tf&apos;</code>: Return a batch of type <code>tf.Tensor</code>.</li>
<li><code>TensorType.PYTORCH</code> or <code>&apos;pt&apos;</code>: Return a batch of type <code>torch.Tensor</code>.</li>
<li><code>TensorType.NUMPY</code> or <code>&apos;np&apos;</code>: Return a batch of type <code>np.ndarray</code>.</li>
<li><code>TensorType.JAX</code> or <code>&apos;jax&apos;</code>: Return a batch of type <code>jax.numpy.ndarray</code>.</li>
</ul>`,name:"return_tensors"},{anchor:"transformers.CLIPImageProcessor.preprocess.data_format",description:`<strong>data_format</strong> (<code>ChannelDimension</code> or <code>str</code>, <em>optional</em>, defaults to <code>ChannelDimension.FIRST</code>) &#x2014;
The channel dimension format for the output image. Can be one of:<ul>
<li><code>&quot;channels_first&quot;</code> or <code>ChannelDimension.FIRST</code>: image in (num_channels, height, width) format.</li>
<li><code>&quot;channels_last&quot;</code> or <code>ChannelDimension.LAST</code>: image in (height, width, num_channels) format.</li>
<li>Unset: Use the channel dimension format of the input image.</li>
</ul>`,name:"data_format"},{anchor:"transformers.CLIPImageProcessor.preprocess.input_data_format",description:`<strong>input_data_format</strong> (<code>ChannelDimension</code> or <code>str</code>, <em>optional</em>) &#x2014;
The channel dimension format for the input image. If unset, the channel dimension format is inferred
from the input image. Can be one of:<ul>
<li><code>&quot;channels_first&quot;</code> or <code>ChannelDimension.FIRST</code>: image in (num_channels, height, width) format.</li>
<li><code>&quot;channels_last&quot;</code> or <code>ChannelDimension.LAST</code>: image in (height, width, num_channels) format.</li>
<li><code>&quot;none&quot;</code> or <code>ChannelDimension.NONE</code>: image in (height, width) format.</li>
</ul>`,name:"input_data_format"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/image_processing_clip.py#L200"}}),vt=new ke({props:{title:"CLIPImageProcessorFast",local:"transformers.CLIPImageProcessorFast",headingTag:"h2"}}),x=new J({props:{name:"class transformers.CLIPImageProcessorFast",anchor:"transformers.CLIPImageProcessorFast",parameters:[{name:"**kwargs",val:": typing_extensions.Unpack[transformers.image_processing_utils_fast.DefaultFastImageProcessorKwargs]"}],parametersDescription:[{anchor:"transformers.CLIPImageProcessorFast.do_resize",description:`<strong>do_resize</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>self.do_resize</code>) &#x2014;
Whether to resize the image&#x2019;s (height, width) dimensions to the specified <code>size</code>. Can be overridden by the
<code>do_resize</code> parameter in the <code>preprocess</code> method.`,name:"do_resize"},{anchor:"transformers.CLIPImageProcessorFast.size",description:`<strong>size</strong> (<code>dict</code>, <em>optional</em>, defaults to <code>self.size</code>) &#x2014;
Size of the output image after resizing. Can be overridden by the <code>size</code> parameter in the <code>preprocess</code>
method.`,name:"size"},{anchor:"transformers.CLIPImageProcessorFast.default_to_square",description:`<strong>default_to_square</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>self.default_to_square</code>) &#x2014;
Whether to default to a square image when resizing, if size is an int.`,name:"default_to_square"},{anchor:"transformers.CLIPImageProcessorFast.resample",description:`<strong>resample</strong> (<code>PILImageResampling</code>, <em>optional</em>, defaults to <code>self.resample</code>) &#x2014;
Resampling filter to use if resizing the image. Only has an effect if <code>do_resize</code> is set to <code>True</code>. Can be
overridden by the <code>resample</code> parameter in the <code>preprocess</code> method.`,name:"resample"},{anchor:"transformers.CLIPImageProcessorFast.do_center_crop",description:`<strong>do_center_crop</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>self.do_center_crop</code>) &#x2014;
Whether to center crop the image to the specified <code>crop_size</code>. Can be overridden by <code>do_center_crop</code> in the
<code>preprocess</code> method.`,name:"do_center_crop"},{anchor:"transformers.CLIPImageProcessorFast.crop_size",description:`<strong>crop_size</strong> (<code>Dict[str, int]</code> <em>optional</em>, defaults to <code>self.crop_size</code>) &#x2014;
Size of the output image after applying <code>center_crop</code>. Can be overridden by <code>crop_size</code> in the <code>preprocess</code>
method.`,name:"crop_size"},{anchor:"transformers.CLIPImageProcessorFast.do_rescale",description:`<strong>do_rescale</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>self.do_rescale</code>) &#x2014;
Whether to rescale the image by the specified scale <code>rescale_factor</code>. Can be overridden by the
<code>do_rescale</code> parameter in the <code>preprocess</code> method.`,name:"do_rescale"},{anchor:"transformers.CLIPImageProcessorFast.rescale_factor",description:`<strong>rescale_factor</strong> (<code>int</code> or <code>float</code>, <em>optional</em>, defaults to <code>self.rescale_factor</code>) &#x2014;
Scale factor to use if rescaling the image. Only has an effect if <code>do_rescale</code> is set to <code>True</code>. Can be
overridden by the <code>rescale_factor</code> parameter in the <code>preprocess</code> method.`,name:"rescale_factor"},{anchor:"transformers.CLIPImageProcessorFast.do_normalize",description:`<strong>do_normalize</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>self.do_normalize</code>) &#x2014;
Whether to normalize the image. Can be overridden by the <code>do_normalize</code> parameter in the <code>preprocess</code>
method. Can be overridden by the <code>do_normalize</code> parameter in the <code>preprocess</code> method.`,name:"do_normalize"},{anchor:"transformers.CLIPImageProcessorFast.image_mean",description:`<strong>image_mean</strong> (<code>float</code> or <code>List[float]</code>, <em>optional</em>, defaults to <code>self.image_mean</code>) &#x2014;
Mean to use if normalizing the image. This is a float or list of floats the length of the number of
channels in the image. Can be overridden by the <code>image_mean</code> parameter in the <code>preprocess</code> method. Can be
overridden by the <code>image_mean</code> parameter in the <code>preprocess</code> method.`,name:"image_mean"},{anchor:"transformers.CLIPImageProcessorFast.image_std",description:`<strong>image_std</strong> (<code>float</code> or <code>List[float]</code>, <em>optional</em>, defaults to <code>self.image_std</code>) &#x2014;
Standard deviation to use if normalizing the image. This is a float or list of floats the length of the
number of channels in the image. Can be overridden by the <code>image_std</code> parameter in the <code>preprocess</code> method.
Can be overridden by the <code>image_std</code> parameter in the <code>preprocess</code> method.`,name:"image_std"},{anchor:"transformers.CLIPImageProcessorFast.do_convert_rgb",description:`<strong>do_convert_rgb</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>self.do_convert_rgb</code>) &#x2014;
Whether to convert the image to RGB.`,name:"do_convert_rgb"},{anchor:"transformers.CLIPImageProcessorFast.return_tensors",description:"<strong>return_tensors</strong> (<code>str</code> or <code>TensorType</code>, <em>optional</em>, defaults to <code>self.return_tensors</code>) &#x2014;\nReturns stacked tensors if set to `pt, otherwise returns a list of tensors.",name:"return_tensors"},{anchor:"transformers.CLIPImageProcessorFast.data_format",description:`<strong>data_format</strong> (<code>ChannelDimension</code> or <code>str</code>, <em>optional</em>, defaults to <code>self.data_format</code>) &#x2014;
Only <code>ChannelDimension.FIRST</code> is supported. Added for compatibility with slow processors.`,name:"data_format"},{anchor:"transformers.CLIPImageProcessorFast.input_data_format",description:`<strong>input_data_format</strong> (<code>ChannelDimension</code> or <code>str</code>, <em>optional</em>, defaults to <code>self.input_data_format</code>) &#x2014;
The channel dimension format for the input image. If unset, the channel dimension format is inferred
from the input image. Can be one of:<ul>
<li><code>&quot;channels_first&quot;</code> or <code>ChannelDimension.FIRST</code>: image in (num_channels, height, width) format.</li>
<li><code>&quot;channels_last&quot;</code> or <code>ChannelDimension.LAST</code>: image in (height, width, num_channels) format.</li>
<li><code>&quot;none&quot;</code> or <code>ChannelDimension.NONE</code>: image in (height, width) format.</li>
</ul>`,name:"input_data_format"},{anchor:"transformers.CLIPImageProcessorFast.device",description:`<strong>device</strong> (<code>torch.device</code>, <em>optional</em>, defaults to <code>self.device</code>) &#x2014;
The device to process the images on. If unset, the device is inferred from the input images.`,name:"device"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/image_processing_clip_fast.py#L22"}}),Se=new J({props:{name:"preprocess",anchor:"transformers.CLIPImageProcessorFast.preprocess",parameters:[{name:"images",val:": typing.Union[ForwardRef('PIL.Image.Image'), numpy.ndarray, ForwardRef('torch.Tensor'), list['PIL.Image.Image'], list[numpy.ndarray], list['torch.Tensor']]"},{name:"**kwargs",val:": typing_extensions.Unpack[transformers.image_processing_utils_fast.DefaultFastImageProcessorKwargs]"}],parametersDescription:[{anchor:"transformers.CLIPImageProcessorFast.preprocess.images",description:`<strong>images</strong> (<code>ImageInput</code>) &#x2014;
Image to preprocess. Expects a single or batch of images with pixel values ranging from 0 to 255. If
passing in images with pixel values between 0 and 1, set <code>do_rescale=False</code>.`,name:"images"},{anchor:"transformers.CLIPImageProcessorFast.preprocess.do_resize",description:`<strong>do_resize</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>self.do_resize</code>) &#x2014;
Whether to resize the image.`,name:"do_resize"},{anchor:"transformers.CLIPImageProcessorFast.preprocess.size",description:`<strong>size</strong> (<code>Dict[str, int]</code>, <em>optional</em>, defaults to <code>self.size</code>) &#x2014;
Describes the maximum input dimensions to the model.`,name:"size"},{anchor:"transformers.CLIPImageProcessorFast.preprocess.resample",description:`<strong>resample</strong> (<code>PILImageResampling</code> or <code>InterpolationMode</code>, <em>optional</em>, defaults to <code>self.resample</code>) &#x2014;
Resampling filter to use if resizing the image. This can be one of the enum <code>PILImageResampling</code>. Only
has an effect if <code>do_resize</code> is set to <code>True</code>.`,name:"resample"},{anchor:"transformers.CLIPImageProcessorFast.preprocess.do_center_crop",description:`<strong>do_center_crop</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>self.do_center_crop</code>) &#x2014;
Whether to center crop the image.`,name:"do_center_crop"},{anchor:"transformers.CLIPImageProcessorFast.preprocess.crop_size",description:`<strong>crop_size</strong> (<code>Dict[str, int]</code>, <em>optional</em>, defaults to <code>self.crop_size</code>) &#x2014;
Size of the output image after applying <code>center_crop</code>.`,name:"crop_size"},{anchor:"transformers.CLIPImageProcessorFast.preprocess.do_rescale",description:`<strong>do_rescale</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>self.do_rescale</code>) &#x2014;
Whether to rescale the image.`,name:"do_rescale"},{anchor:"transformers.CLIPImageProcessorFast.preprocess.rescale_factor",description:`<strong>rescale_factor</strong> (<code>float</code>, <em>optional</em>, defaults to <code>self.rescale_factor</code>) &#x2014;
Rescale factor to rescale the image by if <code>do_rescale</code> is set to <code>True</code>.`,name:"rescale_factor"},{anchor:"transformers.CLIPImageProcessorFast.preprocess.do_normalize",description:`<strong>do_normalize</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>self.do_normalize</code>) &#x2014;
Whether to normalize the image.`,name:"do_normalize"},{anchor:"transformers.CLIPImageProcessorFast.preprocess.image_mean",description:`<strong>image_mean</strong> (<code>float</code> or <code>List[float]</code>, <em>optional</em>, defaults to <code>self.image_mean</code>) &#x2014;
Image mean to use for normalization. Only has an effect if <code>do_normalize</code> is set to <code>True</code>.`,name:"image_mean"},{anchor:"transformers.CLIPImageProcessorFast.preprocess.image_std",description:`<strong>image_std</strong> (<code>float</code> or <code>List[float]</code>, <em>optional</em>, defaults to <code>self.image_std</code>) &#x2014;
Image standard deviation to use for normalization. Only has an effect if <code>do_normalize</code> is set to
<code>True</code>.`,name:"image_std"},{anchor:"transformers.CLIPImageProcessorFast.preprocess.do_convert_rgb",description:`<strong>do_convert_rgb</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>self.do_convert_rgb</code>) &#x2014;
Whether to convert the image to RGB.`,name:"do_convert_rgb"},{anchor:"transformers.CLIPImageProcessorFast.preprocess.return_tensors",description:"<strong>return_tensors</strong> (<code>str</code> or <code>TensorType</code>, <em>optional</em>, defaults to <code>self.return_tensors</code>) &#x2014;\nReturns stacked tensors if set to `pt, otherwise returns a list of tensors.",name:"return_tensors"},{anchor:"transformers.CLIPImageProcessorFast.preprocess.data_format",description:`<strong>data_format</strong> (<code>ChannelDimension</code> or <code>str</code>, <em>optional</em>, defaults to <code>self.data_format</code>) &#x2014;
Only <code>ChannelDimension.FIRST</code> is supported. Added for compatibility with slow processors.`,name:"data_format"},{anchor:"transformers.CLIPImageProcessorFast.preprocess.input_data_format",description:`<strong>input_data_format</strong> (<code>ChannelDimension</code> or <code>str</code>, <em>optional</em>, defaults to <code>self.input_data_format</code>) &#x2014;
The channel dimension format for the input image. If unset, the channel dimension format is inferred
from the input image. Can be one of:<ul>
<li><code>&quot;channels_first&quot;</code> or <code>ChannelDimension.FIRST</code>: image in (num_channels, height, width) format.</li>
<li><code>&quot;channels_last&quot;</code> or <code>ChannelDimension.LAST</code>: image in (height, width, num_channels) format.</li>
<li><code>&quot;none&quot;</code> or <code>ChannelDimension.NONE</code>: image in (height, width) format.</li>
</ul>`,name:"input_data_format"},{anchor:"transformers.CLIPImageProcessorFast.preprocess.device",description:`<strong>device</strong> (<code>torch.device</code>, <em>optional</em>, defaults to <code>self.device</code>) &#x2014;
The device to process the images on. If unset, the device is inferred from the input images.`,name:"device"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/image_processing_utils_fast.py#L667"}}),lt=new ke({props:{title:"CLIPFeatureExtractor",local:"transformers.CLIPFeatureExtractor",headingTag:"h2"}}),ho=new J({props:{name:"class transformers.CLIPFeatureExtractor",anchor:"transformers.CLIPFeatureExtractor",parameters:[{name:"*args",val:""},{name:"**kwargs",val:""}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/feature_extraction_clip.py#L26"}}),go=new ke({props:{title:"CLIPProcessor",local:"transformers.CLIPProcessor",headingTag:"h2"}}),Go=new J({props:{name:"class transformers.CLIPProcessor",anchor:"transformers.CLIPProcessor",parameters:[{name:"image_processor",val:" = None"},{name:"tokenizer",val:" = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.CLIPProcessor.image_processor",description:`<strong>image_processor</strong> (<a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPImageProcessor">CLIPImageProcessor</a>, <em>optional</em>) &#x2014;
The image processor is a required input.`,name:"image_processor"},{anchor:"transformers.CLIPProcessor.tokenizer",description:`<strong>tokenizer</strong> (<a href="/docs/transformers/pr_37350/en/model_doc/clip#transformers.CLIPTokenizerFast">CLIPTokenizerFast</a>, <em>optional</em>) &#x2014;
The tokenizer is a required input.`,name:"tokenizer"}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/processing_clip.py#L25"}}),No=new J({props:{name:"batch_decode",anchor:"transformers.CLIPProcessor.batch_decode",parameters:[{name:"*args",val:""},{name:"**kwargs",val:""}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/processing_clip.py#L119"}}),qo=new J({props:{name:"decode",anchor:"transformers.CLIPProcessor.decode",parameters:[{name:"*args",val:""},{name:"**kwargs",val:""}],source:"https://github.com/huggingface/transformers/blob/vr_37350/src/transformers/models/clip/processing_clip.py#L126"}}),ko=new Gn({props:{pytorch:!0,tensorflow:!0,jax:!0,$$slots:{jax:[Vs],tensorflow:[$s],pytorch:[fs]},$$scope:{ctx:w}}}),Eo=new Nn({props:{source:"https://github.com/huggingface/transformers/blob/main/docs/source/en/model_doc/clip.md"}}),{c(){e=c("meta"),d=a(),o=c("p"),n=a(),m=c("div"),m.innerHTML=t,v=a(),h(Me.$$.fragment),X=a(),$e=c("p"),$e.innerHTML=bt,W=a(),H=c("p"),H.innerHTML=me,z=a(),h(ae.$$.fragment),tt=a(),G=c("p"),G.innerHTML=pt,fe=a(),h(Je.$$.fragment),K=a(),h(ze.$$.fragment),Y=a(),ue=c("ul"),ue.innerHTML=mt,ot=a(),h(ie.$$.fragment),De=a(),Z=c("div"),h(xe.$$.fragment),ee=a(),de=c("p"),de.innerHTML=ce,Be=a(),Ge=c("p"),Ge.innerHTML=ft,Ue=a(),h(Xe.$$.fragment),N=a(),Ie=c("div"),h(le.$$.fragment),he=a(),V=c("p"),V.innerHTML=nt,ve=a(),h(Ce.$$.fragment),wt=a(),F=c("div"),h(ge.$$.fragment),st=a(),Ne=c("p"),Ne.innerHTML=_e,Qe=a(),te=c("p"),te.innerHTML=He,rt=a(),h(S.$$.fragment),oe=a(),h(Ae.$$.fragment),D=a(),ne=c("div"),h(Fe.$$.fragment),Q=a(),B=c("p"),B.innerHTML=We,at=a(),Pe=c("p"),Pe.innerHTML=Vt,be=a(),h(pe.$$.fragment),Oe=a(),h(se.$$.fragment),C=a(),P=c("div"),h(A.$$.fragment),q=a(),O=c("p"),O.textContent=Ze,Te=a(),ye=c("p"),ye.innerHTML=Ve,I=a(),L=c("div"),h(E.$$.fragment),it=a(),qe=c("p"),qe.textContent=$t,R=a(),Ee=c("ul"),Ee.innerHTML=ut,ht=a(),Ye=c("p"),Ye.textContent=To,io=a(),Ot=c("div"),h(Kt.$$.fragment),lo=a(),yo=c("p"),yo.innerHTML=Xo,yt=a(),Mt=c("div"),h(eo.$$.fragment),co=a(),Mo=c("p"),Mo.textContent=Ho,Dt=a(),po=c("div"),h(It.$$.fragment),Io=a(),h(xt.$$.fragment),vo=a(),re=c("div"),h(Pt.$$.fragment),Jo=a(),Rt=c("p"),Rt.innerHTML=on,zo=a(),Xt=c("p"),Xt.innerHTML=nn,Uo=a(),we=c("div"),h(Lt.$$.fragment),Fo=a(),Ht=c("p"),Ht.textContent=sn,Wo=a(),kt=c("ul"),kt.innerHTML=Yo,Qt=a(),to=c("p"),to.textContent=mo,wo=a(),Le=c("div"),h(jt.$$.fragment),Zo=a(),Yt=c("p"),Yt.textContent=rn,$o=a(),h(Bt.$$.fragment),Vo=a(),Jt=c("div"),h(Gt.$$.fragment),So=a(),fo=c("p"),fo.textContent=zt,oo=a(),At=c("div"),h(Nt.$$.fragment),Do=a(),uo=c("p"),uo.textContent=no,xo=a(),h(vt.$$.fragment),Co=a(),f=c("div"),h(x.$$.fragment),Ut=a(),gt=c("p"),gt.textContent=qt,Ft=a(),Ke=c("div"),h(Se.$$.fragment),Wt=a(),et=c("p"),et.textContent=Et,Ct=a(),h(lt.$$.fragment),_t=a(),dt=c("div"),h(ho.$$.fragment),Bo=a(),h(go.$$.fragment),ln=a(),Zt=c("div"),h(Go.$$.fragment),yn=a(),Qo=c("p"),Qo.textContent=xn,Mn=a(),Ao=c("p"),Ao.innerHTML=Pn,In=a(),Po=c("div"),h(No.$$.fragment),vn=a(),Oo=c("p"),Oo.innerHTML=Ln,Cn=a(),Lo=c("div"),h(qo.$$.fragment),wn=a(),Ko=c("p"),Ko.innerHTML=kn,dn=a(),h(ko.$$.fragment),cn=a(),h(Eo.$$.fragment),pn=a(),an=c("p"),this.h()},l(r){const $=Vn("svelte-u9bgzb",document.head);e=p($,"META",{name:!0,content:!0}),$.forEach(l),d=i(r),o=p(r,"P",{}),k(o).forEach(l),n=i(r),m=p(r,"DIV",{style:!0,"data-svelte-h":!0}),M(m)!=="svelte-jbowd8"&&(m.innerHTML=t),v=i(r),g(Me.$$.fragment,r),X=i(r),$e=p(r,"P",{"data-svelte-h":!0}),M($e)!=="svelte-ka0a7b"&&($e.innerHTML=bt),W=i(r),H=p(r,"P",{"data-svelte-h":!0}),M(H)!=="svelte-17xaumu"&&(H.innerHTML=me),z=i(r),g(ae.$$.fragment,r),tt=i(r),G=p(r,"P",{"data-svelte-h":!0}),M(G)!=="svelte-lnp4tr"&&(G.innerHTML=pt),fe=i(r),g(Je.$$.fragment,r),K=i(r),g(ze.$$.fragment,r),Y=i(r),ue=p(r,"UL",{"data-svelte-h":!0}),M(ue)!=="svelte-9napk1"&&(ue.innerHTML=mt),ot=i(r),g(ie.$$.fragment,r),De=i(r),Z=p(r,"DIV",{class:!0});var St=k(Z);g(xe.$$.fragment,St),ee=i(St),de=p(St,"P",{"data-svelte-h":!0}),M(de)!=="svelte-1jif1aj"&&(de.innerHTML=ce),Be=i(St),Ge=p(St,"P",{"data-svelte-h":!0}),M(Ge)!=="svelte-1dngyef"&&(Ge.innerHTML=ft),Ue=i(St),g(Xe.$$.fragment,St),N=i(St),Ie=p(St,"DIV",{class:!0});var Ro=k(Ie);g(le.$$.fragment,Ro),he=i(Ro),V=p(Ro,"P",{"data-svelte-h":!0}),M(V)!=="svelte-prybsw"&&(V.innerHTML=nt),Ro.forEach(l),St.forEach(l),ve=i(r),g(Ce.$$.fragment,r),wt=i(r),F=p(r,"DIV",{class:!0});var so=k(F);g(ge.$$.fragment,so),st=i(so),Ne=p(so,"P",{"data-svelte-h":!0}),M(Ne)!=="svelte-1ph6tlg"&&(Ne.innerHTML=_e),Qe=i(so),te=p(so,"P",{"data-svelte-h":!0}),M(te)!=="svelte-1dngyef"&&(te.innerHTML=He),rt=i(so),g(S.$$.fragment,so),so.forEach(l),oe=i(r),g(Ae.$$.fragment,r),D=i(r),ne=p(r,"DIV",{class:!0});var ro=k(ne);g(Fe.$$.fragment,ro),Q=i(ro),B=p(ro,"P",{"data-svelte-h":!0}),M(B)!=="svelte-1u2nal8"&&(B.innerHTML=We),at=i(ro),Pe=p(ro,"P",{"data-svelte-h":!0}),M(Pe)!=="svelte-1dngyef"&&(Pe.innerHTML=Vt),be=i(ro),g(pe.$$.fragment,ro),ro.forEach(l),Oe=i(r),g(se.$$.fragment,r),C=i(r),P=p(r,"DIV",{class:!0});var Tt=k(P);g(A.$$.fragment,Tt),q=i(Tt),O=p(Tt,"P",{"data-svelte-h":!0}),M(O)!=="svelte-vz3r3l"&&(O.textContent=Ze),Te=i(Tt),ye=p(Tt,"P",{"data-svelte-h":!0}),M(ye)!=="svelte-xkxj4"&&(ye.innerHTML=Ve),I=i(Tt),L=p(Tt,"DIV",{class:!0});var ao=k(L);g(E.$$.fragment,ao),it=i(ao),qe=p(ao,"P",{"data-svelte-h":!0}),M(qe)!=="svelte-js0y4v"&&(qe.textContent=$t),R=i(ao),Ee=p(ao,"UL",{"data-svelte-h":!0}),M(Ee)!=="svelte-uzwo54"&&(Ee.innerHTML=ut),ht=i(ao),Ye=p(ao,"P",{"data-svelte-h":!0}),M(Ye)!=="svelte-1bzh7dh"&&(Ye.textContent=To),ao.forEach(l),io=i(Tt),Ot=p(Tt,"DIV",{class:!0});var fn=k(Ot);g(Kt.$$.fragment,fn),lo=i(fn),yo=p(fn,"P",{"data-svelte-h":!0}),M(yo)!=="svelte-1f4f5kp"&&(yo.innerHTML=Xo),fn.forEach(l),yt=i(Tt),Mt=p(Tt,"DIV",{class:!0});var un=k(Mt);g(eo.$$.fragment,un),co=i(un),Mo=p(un,"P",{"data-svelte-h":!0}),M(Mo)!=="svelte-1ctpnmt"&&(Mo.textContent=Ho),un.forEach(l),Dt=i(Tt),po=p(Tt,"DIV",{class:!0});var jn=k(po);g(It.$$.fragment,jn),jn.forEach(l),Tt.forEach(l),Io=i(r),g(xt.$$.fragment,r),vo=i(r),re=p(r,"DIV",{class:!0});var _o=k(re);g(Pt.$$.fragment,_o),Jo=i(_o),Rt=p(_o,"P",{"data-svelte-h":!0}),M(Rt)!=="svelte-g8hwlw"&&(Rt.innerHTML=on),zo=i(_o),Xt=p(_o,"P",{"data-svelte-h":!0}),M(Xt)!=="svelte-18t2z8k"&&(Xt.innerHTML=nn),Uo=i(_o),we=p(_o,"DIV",{class:!0});var jo=k(we);g(Lt.$$.fragment,jo),Fo=i(jo),Ht=p(jo,"P",{"data-svelte-h":!0}),M(Ht)!=="svelte-js0y4v"&&(Ht.textContent=sn),Wo=i(jo),kt=p(jo,"UL",{"data-svelte-h":!0}),M(kt)!=="svelte-uzwo54"&&(kt.innerHTML=Yo),Qt=i(jo),to=p(jo,"P",{"data-svelte-h":!0}),M(to)!=="svelte-1bzh7dh"&&(to.textContent=mo),jo.forEach(l),wo=i(_o),Le=p(_o,"DIV",{class:!0});var hn=k(Le);g(jt.$$.fragment,hn),Zo=i(hn),Yt=p(hn,"P",{"data-svelte-h":!0}),M(Yt)!=="svelte-1ctpnmt"&&(Yt.textContent=rn),hn.forEach(l),_o.forEach(l),$o=i(r),g(Bt.$$.fragment,r),Vo=i(r),Jt=p(r,"DIV",{class:!0});var en=k(Jt);g(Gt.$$.fragment,en),So=i(en),fo=p(en,"P",{"data-svelte-h":!0}),M(fo)!=="svelte-x48p92"&&(fo.textContent=zt),oo=i(en),At=p(en,"DIV",{class:!0});var gn=k(At);g(Nt.$$.fragment,gn),Do=i(gn),uo=p(gn,"P",{"data-svelte-h":!0}),M(uo)!=="svelte-1x3yxsa"&&(uo.textContent=no),gn.forEach(l),en.forEach(l),xo=i(r),g(vt.$$.fragment,r),Co=i(r),f=p(r,"DIV",{class:!0});var tn=k(f);g(x.$$.fragment,tn),Ut=i(tn),gt=p(tn,"P",{"data-svelte-h":!0}),M(gt)!=="svelte-refdl6"&&(gt.textContent=qt),Ft=i(tn),Ke=p(tn,"DIV",{class:!0});var _n=k(Ke);g(Se.$$.fragment,_n),Wt=i(_n),et=p(_n,"P",{"data-svelte-h":!0}),M(et)!=="svelte-1x3yxsa"&&(et.textContent=Et),_n.forEach(l),tn.forEach(l),Ct=i(r),g(lt.$$.fragment,r),_t=i(r),dt=p(r,"DIV",{class:!0});var Jn=k(dt);g(ho.$$.fragment,Jn),Jn.forEach(l),Bo=i(r),g(go.$$.fragment,r),ln=i(r),Zt=p(r,"DIV",{class:!0});var bo=k(Zt);g(Go.$$.fragment,bo),yn=i(bo),Qo=p(bo,"P",{"data-svelte-h":!0}),M(Qo)!=="svelte-ofpncz"&&(Qo.textContent=xn),Mn=i(bo),Ao=p(bo,"P",{"data-svelte-h":!0}),M(Ao)!=="svelte-1kcjk5e"&&(Ao.innerHTML=Pn),In=i(bo),Po=p(bo,"DIV",{class:!0});var bn=k(Po);g(No.$$.fragment,bn),vn=i(bn),Oo=p(bn,"P",{"data-svelte-h":!0}),M(Oo)!=="svelte-1atw87z"&&(Oo.innerHTML=Ln),bn.forEach(l),Cn=i(bo),Lo=p(bo,"DIV",{class:!0});var Tn=k(Lo);g(qo.$$.fragment,Tn),wn=i(Tn),Ko=p(Tn,"P",{"data-svelte-h":!0}),M(Ko)!=="svelte-1qzn2zd"&&(Ko.innerHTML=kn),Tn.forEach(l),bo.forEach(l),dn=i(r),g(ko.$$.fragment,r),cn=i(r),g(Eo.$$.fragment,r),pn=i(r),an=p(r,"P",{}),k(an).forEach(l),this.h()},h(){j(e,"name","hf:doc:metadata"),j(e,"content",Gs),Bn(m,"float","right"),j(Ie,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(Z,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(F,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(ne,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(L,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(Ot,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(Mt,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(po,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(P,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(we,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(Le,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(re,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(At,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(Jt,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(Ke,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(f,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(dt,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(Po,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(Lo,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),j(Zt,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8")},m(r,$){s(document.head,e),u(r,d,$),u(r,o,$),u(r,n,$),u(r,m,$),u(r,v,$),_(Me,r,$),u(r,X,$),u(r,$e,$),u(r,W,$),u(r,H,$),u(r,z,$),_(ae,r,$),u(r,tt,$),u(r,G,$),u(r,fe,$),_(Je,r,$),u(r,K,$),_(ze,r,$),u(r,Y,$),u(r,ue,$),u(r,ot,$),_(ie,r,$),u(r,De,$),u(r,Z,$),_(xe,Z,null),s(Z,ee),s(Z,de),s(Z,Be),s(Z,Ge),s(Z,Ue),_(Xe,Z,null),s(Z,N),s(Z,Ie),_(le,Ie,null),s(Ie,he),s(Ie,V),u(r,ve,$),_(Ce,r,$),u(r,wt,$),u(r,F,$),_(ge,F,null),s(F,st),s(F,Ne),s(F,Qe),s(F,te),s(F,rt),_(S,F,null),u(r,oe,$),_(Ae,r,$),u(r,D,$),u(r,ne,$),_(Fe,ne,null),s(ne,Q),s(ne,B),s(ne,at),s(ne,Pe),s(ne,be),_(pe,ne,null),u(r,Oe,$),_(se,r,$),u(r,C,$),u(r,P,$),_(A,P,null),s(P,q),s(P,O),s(P,Te),s(P,ye),s(P,I),s(P,L),_(E,L,null),s(L,it),s(L,qe),s(L,R),s(L,Ee),s(L,ht),s(L,Ye),s(P,io),s(P,Ot),_(Kt,Ot,null),s(Ot,lo),s(Ot,yo),s(P,yt),s(P,Mt),_(eo,Mt,null),s(Mt,co),s(Mt,Mo),s(P,Dt),s(P,po),_(It,po,null),u(r,Io,$),_(xt,r,$),u(r,vo,$),u(r,re,$),_(Pt,re,null),s(re,Jo),s(re,Rt),s(re,zo),s(re,Xt),s(re,Uo),s(re,we),_(Lt,we,null),s(we,Fo),s(we,Ht),s(we,Wo),s(we,kt),s(we,Qt),s(we,to),s(re,wo),s(re,Le),_(jt,Le,null),s(Le,Zo),s(Le,Yt),u(r,$o,$),_(Bt,r,$),u(r,Vo,$),u(r,Jt,$),_(Gt,Jt,null),s(Jt,So),s(Jt,fo),s(Jt,oo),s(Jt,At),_(Nt,At,null),s(At,Do),s(At,uo),u(r,xo,$),_(vt,r,$),u(r,Co,$),u(r,f,$),_(x,f,null),s(f,Ut),s(f,gt),s(f,Ft),s(f,Ke),_(Se,Ke,null),s(Ke,Wt),s(Ke,et),u(r,Ct,$),_(lt,r,$),u(r,_t,$),u(r,dt,$),_(ho,dt,null),u(r,Bo,$),_(go,r,$),u(r,ln,$),u(r,Zt,$),_(Go,Zt,null),s(Zt,yn),s(Zt,Qo),s(Zt,Mn),s(Zt,Ao),s(Zt,In),s(Zt,Po),_(No,Po,null),s(Po,vn),s(Po,Oo),s(Zt,Cn),s(Zt,Lo),_(qo,Lo,null),s(Lo,wn),s(Lo,Ko),u(r,dn,$),_(ko,r,$),u(r,cn,$),_(Eo,r,$),u(r,pn,$),u(r,an,$),mn=!0},p(r,[$]){const St={};$&2&&(St.$$scope={dirty:$,ctx:r}),ae.$set(St);const Ro={};$&2&&(Ro.$$scope={dirty:$,ctx:r}),Je.$set(Ro);const so={};$&2&&(so.$$scope={dirty:$,ctx:r}),Xe.$set(so);const ro={};$&2&&(ro.$$scope={dirty:$,ctx:r}),S.$set(ro);const Tt={};$&2&&(Tt.$$scope={dirty:$,ctx:r}),pe.$set(Tt);const ao={};$&2&&(ao.$$scope={dirty:$,ctx:r}),ko.$set(ao)},i(r){mn||(b(Me.$$.fragment,r),b(ae.$$.fragment,r),b(Je.$$.fragment,r),b(ze.$$.fragment,r),b(ie.$$.fragment,r),b(xe.$$.fragment,r),b(Xe.$$.fragment,r),b(le.$$.fragment,r),b(Ce.$$.fragment,r),b(ge.$$.fragment,r),b(S.$$.fragment,r),b(Ae.$$.fragment,r),b(Fe.$$.fragment,r),b(pe.$$.fragment,r),b(se.$$.fragment,r),b(A.$$.fragment,r),b(E.$$.fragment,r),b(Kt.$$.fragment,r),b(eo.$$.fragment,r),b(It.$$.fragment,r),b(xt.$$.fragment,r),b(Pt.$$.fragment,r),b(Lt.$$.fragment,r),b(jt.$$.fragment,r),b(Bt.$$.fragment,r),b(Gt.$$.fragment,r),b(Nt.$$.fragment,r),b(vt.$$.fragment,r),b(x.$$.fragment,r),b(Se.$$.fragment,r),b(lt.$$.fragment,r),b(ho.$$.fragment,r),b(go.$$.fragment,r),b(Go.$$.fragment,r),b(No.$$.fragment,r),b(qo.$$.fragment,r),b(ko.$$.fragment,r),b(Eo.$$.fragment,r),mn=!0)},o(r){T(Me.$$.fragment,r),T(ae.$$.fragment,r),T(Je.$$.fragment,r),T(ze.$$.fragment,r),T(ie.$$.fragment,r),T(xe.$$.fragment,r),T(Xe.$$.fragment,r),T(le.$$.fragment,r),T(Ce.$$.fragment,r),T(ge.$$.fragment,r),T(S.$$.fragment,r),T(Ae.$$.fragment,r),T(Fe.$$.fragment,r),T(pe.$$.fragment,r),T(se.$$.fragment,r),T(A.$$.fragment,r),T(E.$$.fragment,r),T(Kt.$$.fragment,r),T(eo.$$.fragment,r),T(It.$$.fragment,r),T(xt.$$.fragment,r),T(Pt.$$.fragment,r),T(Lt.$$.fragment,r),T(jt.$$.fragment,r),T(Bt.$$.fragment,r),T(Gt.$$.fragment,r),T(Nt.$$.fragment,r),T(vt.$$.fragment,r),T(x.$$.fragment,r),T(Se.$$.fragment,r),T(lt.$$.fragment,r),T(ho.$$.fragment,r),T(go.$$.fragment,r),T(Go.$$.fragment,r),T(No.$$.fragment,r),T(qo.$$.fragment,r),T(ko.$$.fragment,r),T(Eo.$$.fragment,r),mn=!1},d(r){r&&(l(d),l(o),l(n),l(m),l(v),l(X),l($e),l(W),l(H),l(z),l(tt),l(G),l(fe),l(K),l(Y),l(ue),l(ot),l(De),l(Z),l(ve),l(wt),l(F),l(oe),l(D),l(ne),l(Oe),l(C),l(P),l(Io),l(vo),l(re),l($o),l(Vo),l(Jt),l(xo),l(Co),l(f),l(Ct),l(_t),l(dt),l(Bo),l(ln),l(Zt),l(dn),l(cn),l(pn),l(an)),l(e),y(Me,r),y(ae,r),y(Je,r),y(ze,r),y(ie,r),y(xe),y(Xe),y(le),y(Ce,r),y(ge),y(S),y(Ae,r),y(Fe),y(pe),y(se,r),y(A),y(E),y(Kt),y(eo),y(It),y(xt,r),y(Pt),y(Lt),y(jt),y(Bt,r),y(Gt),y(Nt),y(vt,r),y(x),y(Se),y(lt,r),y(ho),y(go,r),y(Go),y(No),y(qo),y(ko,r),y(Eo,r)}}}const Gs='{"title":"CLIP","local":"clip","sections":[{"title":"Notes","local":"notes","sections":[],"depth":2},{"title":"CLIPConfig","local":"transformers.CLIPConfig","sections":[],"depth":2},{"title":"CLIPTextConfig","local":"transformers.CLIPTextConfig","sections":[],"depth":2},{"title":"CLIPVisionConfig","local":"transformers.CLIPVisionConfig","sections":[],"depth":2},{"title":"CLIPTokenizer","local":"transformers.CLIPTokenizer","sections":[],"depth":2},{"title":"CLIPTokenizerFast","local":"transformers.CLIPTokenizerFast","sections":[],"depth":2},{"title":"CLIPImageProcessor","local":"transformers.CLIPImageProcessor","sections":[],"depth":2},{"title":"CLIPImageProcessorFast","local":"transformers.CLIPImageProcessorFast","sections":[],"depth":2},{"title":"CLIPFeatureExtractor","local":"transformers.CLIPFeatureExtractor","sections":[],"depth":2},{"title":"CLIPProcessor","local":"transformers.CLIPProcessor","sections":[],"depth":2},{"title":"CLIPModel","local":"transformers.CLIPModel","sections":[],"depth":2},{"title":"CLIPTextModel","local":"transformers.CLIPTextModel","sections":[],"depth":2},{"title":"CLIPTextModelWithProjection","local":"transformers.CLIPTextModelWithProjection","sections":[],"depth":2},{"title":"CLIPVisionModelWithProjection","local":"transformers.CLIPVisionModelWithProjection","sections":[],"depth":2},{"title":"CLIPVisionModel","local":"transformers.CLIPVisionModel","sections":[],"depth":2},{"title":"CLIPForImageClassification","local":"transformers.CLIPForImageClassification","sections":[],"depth":2},{"title":"TFCLIPModel","local":"transformers.TFCLIPModel","sections":[],"depth":2},{"title":"TFCLIPTextModel","local":"transformers.TFCLIPTextModel","sections":[],"depth":2},{"title":"TFCLIPVisionModel","local":"transformers.TFCLIPVisionModel","sections":[],"depth":2},{"title":"FlaxCLIPModel","local":"transformers.FlaxCLIPModel","sections":[],"depth":2},{"title":"FlaxCLIPTextModel","local":"transformers.FlaxCLIPTextModel","sections":[],"depth":2},{"title":"FlaxCLIPTextModelWithProjection","local":"transformers.FlaxCLIPTextModelWithProjection","sections":[],"depth":2},{"title":"FlaxCLIPVisionModel","local":"transformers.FlaxCLIPVisionModel","sections":[],"depth":2}],"depth":1}';function Ns(w){return Fn(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class As extends Wn{constructor(e){super(),Zn(this,e,Ns,Bs,Un,{})}}export{As as component};