Buckets:

hf-doc-build
/

doc-dev

hf-doc-build/doc-dev / transformers /pr_34362 /en /_app /immutable /nodes /178.e7c4e4b3.js

rtrm's picture

3 months ago

74.9 kB

	import{s as Rn,f as Hn,o as Sn,n as nt}from"../chunks/scheduler.25b97de1.js";import{S as En,i as Xn,g as c,s as o,r as u,A as Yn,h as d,f as t,c as r,j as w,u as f,x as y,k as $,y as a,a as l,v as g,d as h,t as _,w as b}from"../chunks/index.d9030fc9.js";import{T as Nn}from"../chunks/Tip.baa67368.js";import{D as k}from"../chunks/Docstring.ffac8efa.js";import{C as Ut}from"../chunks/CodeBlock.e6cd0d95.js";import{E as Jt}from"../chunks/ExampleCodeBlock.22dfe688.js";import{H as V,E as An}from"../chunks/EditOnGithub.91d95064.js";function Dn(x){let i,I="Example:",m,p,v;return p=new Ut({props:{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> (
	<span class="hljs-meta">... </span> InstructBlipVisionConfig,
	<span class="hljs-meta">... </span> InstructBlipQFormerConfig,
	<span class="hljs-meta">... </span> OPTConfig,
	<span class="hljs-meta">... </span> InstructBlipConfig,
	<span class="hljs-meta">... </span> InstructBlipForConditionalGeneration,
	<span class="hljs-meta">... </span>)

	<span class="hljs-meta">>>> </span><span class="hljs-comment"># Initializing a InstructBlipConfig with Salesforce/instruct-blip-flan-t5 style configuration</span>
	<span class="hljs-meta">>>> </span>configuration = InstructBlipConfig()

	<span class="hljs-meta">>>> </span><span class="hljs-comment"># Initializing a InstructBlipForConditionalGeneration (with random weights) from the Salesforce/instruct-blip-flan-t5 style configuration</span>
	<span class="hljs-meta">>>> </span>model = InstructBlipForConditionalGeneration(configuration)

	<span class="hljs-meta">>>> </span><span class="hljs-comment"># Accessing the model configuration</span>
	<span class="hljs-meta">>>> </span>configuration = model.config

	<span class="hljs-meta">>>> </span><span class="hljs-comment"># We can also initialize a InstructBlipConfig from a InstructBlipVisionConfig, InstructBlipQFormerConfig and any PretrainedConfig</span>

	<span class="hljs-meta">>>> </span><span class="hljs-comment"># Initializing InstructBLIP vision, InstructBLIP Q-Former and language model configurations</span>
	<span class="hljs-meta">>>> </span>vision_config = InstructBlipVisionConfig()
	<span class="hljs-meta">>>> </span>qformer_config = InstructBlipQFormerConfig()
	<span class="hljs-meta">>>> </span>text_config = OPTConfig()

	<span class="hljs-meta">>>> </span>config = InstructBlipConfig.from_text_vision_configs(vision_config, qformer_config, text_config)`,wrap:!1}}),{c(){i=c("p"),i.textContent=I,m=o(),u(p.$$.fragment)},l(s){i=d(s,"P",{"data-svelte-h":!0}),y(i)!=="svelte-11lpom8"&&(i.textContent=I),m=r(s),f(p.$$.fragment,s)},m(s,M){l(s,i,M),l(s,m,M),g(p,s,M),v=!0},p:nt,i(s){v\|\|(h(p.$$.fragment,s),v=!0)},o(s){_(p.$$.fragment,s),v=!1},d(s){s&&(t(i),t(m)),b(p,s)}}}function On(x){let i,I="Example:",m,p,v;return p=new Ut({props:{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> InstructBlipVisionConfig, InstructBlipVisionModel

	<span class="hljs-meta">>>> </span><span class="hljs-comment"># Initializing a InstructBlipVisionConfig with Salesforce/instruct-blip-flan-t5 style configuration</span>
	<span class="hljs-meta">>>> </span>configuration = InstructBlipVisionConfig()

	<span class="hljs-meta">>>> </span><span class="hljs-comment"># Initializing a InstructBlipVisionModel (with random weights) from the Salesforce/instruct-blip-flan-t5 style configuration</span>
	<span class="hljs-meta">>>> </span>model = InstructBlipVisionModel(configuration)

	<span class="hljs-meta">>>> </span><span class="hljs-comment"># Accessing the model configuration</span>
	<span class="hljs-meta">>>> </span>configuration = model.config`,wrap:!1}}),{c(){i=c("p"),i.textContent=I,m=o(),u(p.$$.fragment)},l(s){i=d(s,"P",{"data-svelte-h":!0}),y(i)!=="svelte-11lpom8"&&(i.textContent=I),m=r(s),f(p.$$.fragment,s)},m(s,M){l(s,i,M),l(s,m,M),g(p,s,M),v=!0},p:nt,i(s){v\|\|(h(p.$$.fragment,s),v=!0)},o(s){_(p.$$.fragment,s),v=!1},d(s){s&&(t(i),t(m)),b(p,s)}}}function Kn(x){let i,I="Examples:",m,p,v;return p=new Ut({props:{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> InstructBlipQFormerConfig, InstructBlipQFormerModel

	<span class="hljs-meta">>>> </span><span class="hljs-comment"># Initializing a InstructBLIP Salesforce/instruct-blip-flan-t5 style configuration</span>
	<span class="hljs-meta">>>> </span>configuration = InstructBlipQFormerConfig()

	<span class="hljs-meta">>>> </span><span class="hljs-comment"># Initializing a model (with random weights) from the Salesforce/instruct-blip-flan-t5 style configuration</span>
	<span class="hljs-meta">>>> </span>model = InstructBlipQFormerModel(configuration)
	<span class="hljs-meta">>>> </span><span class="hljs-comment"># Accessing the model configuration</span>
	<span class="hljs-meta">>>> </span>configuration = model.config`,wrap:!1}}),{c(){i=c("p"),i.textContent=I,m=o(),u(p.$$.fragment)},l(s){i=d(s,"P",{"data-svelte-h":!0}),y(i)!=="svelte-kvfsh7"&&(i.textContent=I),m=r(s),f(p.$$.fragment,s)},m(s,M){l(s,i,M),l(s,m,M),g(p,s,M),v=!0},p:nt,i(s){v\|\|(h(p.$$.fragment,s),v=!0)},o(s){_(p.$$.fragment,s),v=!1},d(s){s&&(t(i),t(m)),b(p,s)}}}function eo(x){let i,I=`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(){i=c("p"),i.innerHTML=I},l(m){i=d(m,"P",{"data-svelte-h":!0}),y(i)!=="svelte-fincs2"&&(i.innerHTML=I)},m(m,p){l(m,i,p)},p:nt,d(m){m&&t(i)}}}function to(x){let i,I=`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(){i=c("p"),i.innerHTML=I},l(m){i=d(m,"P",{"data-svelte-h":!0}),y(i)!=="svelte-fincs2"&&(i.innerHTML=I)},m(m,p){l(m,i,p)},p:nt,d(m){m&&t(i)}}}function no(x){let i,I="Examples:",m,p,v;return p=new Ut({props:{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> InstructBlipProcessor, InstructBlipForConditionalGeneration
	<span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch
	<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image
	<span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> requests

	<span class="hljs-meta">>>> </span>model = InstructBlipForConditionalGeneration.from_pretrained(<span class="hljs-string">"Salesforce/instructblip-vicuna-7b"</span>)
	<span class="hljs-meta">>>> </span>processor = InstructBlipProcessor.from_pretrained(<span class="hljs-string">"Salesforce/instructblip-vicuna-7b"</span>)

	<span class="hljs-meta">>>> </span>device = <span class="hljs-string">"cuda"</span> <span class="hljs-keyword">if</span> torch.cuda.is_available() <span class="hljs-keyword">else</span> <span class="hljs-string">"cpu"</span>
	<span class="hljs-meta">>>> </span>model.to(device)
	<span class="hljs-meta">>>> </span>url = <span class="hljs-string">"https://raw.githubusercontent.com/salesforce/LAVIS/main/docs/_static/Confusing-Pictures.jpg"</span>
	<span class="hljs-meta">>>> </span>image = Image.<span class="hljs-built_in">open</span>(requests.get(url, stream=<span class="hljs-literal">True</span>).raw).convert(<span class="hljs-string">"RGB"</span>)
	<span class="hljs-meta">>>> </span>prompt = <span class="hljs-string">"What is unusual about this image?"</span>
	<span class="hljs-meta">>>> </span>inputs = processor(images=image, text=prompt, return_tensors=<span class="hljs-string">"pt"</span>).to(device)

	<span class="hljs-meta">>>> </span>outputs = model.generate(
	<span class="hljs-meta">... </span> **inputs,
	<span class="hljs-meta">... </span> do_sample=<span class="hljs-literal">False</span>,
	<span class="hljs-meta">... </span> num_beams=<span class="hljs-number">5</span>,
	<span class="hljs-meta">... </span> max_length=<span class="hljs-number">256</span>,
	<span class="hljs-meta">... </span> min_length=<span class="hljs-number">1</span>,
	<span class="hljs-meta">... </span> top_p=<span class="hljs-number">0.9</span>,
	<span class="hljs-meta">... </span> repetition_penalty=<span class="hljs-number">1.5</span>,
	<span class="hljs-meta">... </span> length_penalty=<span class="hljs-number">1.0</span>,
	<span class="hljs-meta">... </span> temperature=<span class="hljs-number">1</span>,
	<span class="hljs-meta">... </span>)
	<span class="hljs-meta">>>> </span>generated_text = processor.batch_decode(outputs, skip_special_tokens=<span class="hljs-literal">True</span>)[<span class="hljs-number">0</span>].strip()
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(generated_text)
	The unusual aspect of this image <span class="hljs-keyword">is</span> that a man <span class="hljs-keyword">is</span> ironing clothes on the back of a yellow SUV, which <span class="hljs-keyword">is</span> parked <span class="hljs-keyword">in</span> the middle of a busy city street. This <span class="hljs-keyword">is</span> an unconventional approach to ironing clothes, <span class="hljs-keyword">as</span> it requires the man to balance himself <span class="hljs-keyword">and</span> his ironing equipment on top of the vehicle <span class="hljs-keyword">while</span> navigating through traffic. Additionally, the presence of taxis <span class="hljs-keyword">and</span> other vehicles <span class="hljs-keyword">in</span> the scene further emphasizes the unusual nature of this situation.`,wrap:!1}}),{c(){i=c("p"),i.textContent=I,m=o(),u(p.$$.fragment)},l(s){i=d(s,"P",{"data-svelte-h":!0}),y(i)!=="svelte-kvfsh7"&&(i.textContent=I),m=r(s),f(p.$$.fragment,s)},m(s,M){l(s,i,M),l(s,m,M),g(p,s,M),v=!0},p:nt,i(s){v\|\|(h(p.$$.fragment,s),v=!0)},o(s){_(p.$$.fragment,s),v=!1},d(s){s&&(t(i),t(m)),b(p,s)}}}function oo(x){let i,I,m,p,v,s,M,rt,te,gn=`The InstructBLIP model was proposed in <a href="https://arxiv.org/abs/2305.06500" rel="nofollow">InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning</a> by Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi.
	InstructBLIP leverages the <a href="blip2">BLIP-2</a> architecture for visual instruction tuning.`,st,ne,hn="The abstract from the paper is the following:",at,oe,_n="<em>General-purpose language models that can solve various language-domain tasks have emerged driven by the pre-training and instruction-tuning pipeline. However, building general-purpose vision-language models is challenging due to the increased task discrepancy introduced by the additional visual input. Although vision-language pre-training has been widely studied, vision-language instruction tuning remains relatively less explored. In this paper, we conduct a systematic and comprehensive study on vision-language instruction tuning based on the pre-trained BLIP-2 models. We gather a wide variety of 26 publicly available datasets, transform them into instruction tuning format and categorize them into two clusters for held-in instruction tuning and held-out zero-shot evaluation. Additionally, we introduce instruction-aware visual feature extraction, a crucial method that enables the model to extract informative features tailored to the given instruction. The resulting InstructBLIP models achieve state-of-the-art zero-shot performance across all 13 held-out datasets, substantially outperforming BLIP-2 and the larger Flamingo. Our models also lead to state-of-the-art performance when finetuned on individual downstream tasks (e.g., 90.7% accuracy on ScienceQA IMG). Furthermore, we qualitatively demonstrate the advantages of InstructBLIP over concurrent multimodal models.</em>",it,N,bn,lt,re,yn='InstructBLIP architecture. Taken from the <a href="https://arxiv.org/abs/2305.06500">original paper.</a>',ct,se,vn=`This model was contributed by <a href="https://huggingface.co/nielsr" rel="nofollow">nielsr</a>.
	The original code can be found <a href="https://github.com/salesforce/LAVIS/tree/main/projects/instructblip" rel="nofollow">here</a>.`,dt,ae,pt,ie,Mn='InstructBLIP uses the same architecture as <a href="blip2">BLIP-2</a> with a tiny but important difference: it also feeds the text prompt (instruction) to the Q-Former.',mt,le,ut,T,ce,zt,Je,In=`<a href="/docs/transformers/pr_34362/en/model_doc/instructblip#transformers.InstructBlipConfig">InstructBlipConfig</a> is the configuration class to store the configuration of a
	<a href="/docs/transformers/pr_34362/en/model_doc/instructblip#transformers.InstructBlipForConditionalGeneration">InstructBlipForConditionalGeneration</a>. It is used to instantiate a InstructBLIP model according to the specified
	arguments, defining the vision model, Q-Former model and language model configs. Instantiating a configuration with
	the defaults will yield a similar configuration to that of the InstructBLIP
	<a href="https://huggingface.co/Salesforce/instruct-blip-flan-t5" rel="nofollow">Salesforce/instruct-blip-flan-t5</a> architecture.`,Wt,Ue,Bn=`Configuration objects inherit from <a href="/docs/transformers/pr_34362/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_34362/en/main_classes/configuration#transformers.PretrainedConfig">PretrainedConfig</a> for more information.`,Zt,R,Pt,H,de,qt,ze,$n=`Instantiate a <a href="/docs/transformers/pr_34362/en/model_doc/instructblip#transformers.InstructBlipConfig">InstructBlipConfig</a> (or a derived class) from a InstructBLIP vision model, Q-Former and
	language model configurations.`,ft,pe,gt,F,me,Qt,We,wn=`This is the configuration class to store the configuration of a <a href="/docs/transformers/pr_34362/en/model_doc/instructblip#transformers.InstructBlipVisionModel">InstructBlipVisionModel</a>. It is used to
	instantiate a InstructBLIP vision encoder according to the specified arguments, defining the model architecture.
	Instantiating a configuration defaults will yield a similar configuration to that of the InstructBLIP
	<a href="https://huggingface.co/Salesforce/instruct-blip-flan-t5" rel="nofollow">Salesforce/instruct-blip-flan-t5</a> architecture.`,Gt,Ze,Tn=`Configuration objects inherit from <a href="/docs/transformers/pr_34362/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_34362/en/main_classes/configuration#transformers.PretrainedConfig">PretrainedConfig</a> for more information.`,Vt,S,ht,ue,_t,J,fe,Lt,Pe,Cn=`This is the configuration class to store the configuration of a <a href="/docs/transformers/pr_34362/en/model_doc/instructblip#transformers.InstructBlipQFormerModel">InstructBlipQFormerModel</a>. It is used to
	instantiate a InstructBLIP Querying Transformer (Q-Former) model according to the specified arguments, defining the
	model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of
	the InstructBLIP <a href="https://huggingface.co/Salesforce/instruct-blip-flan-t5" rel="nofollow">Salesforce/instruct-blip-flan-t5</a>
	architecture. Configuration objects inherit from <a href="/docs/transformers/pr_34362/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_34362/en/main_classes/configuration#transformers.PretrainedConfig">PretrainedConfig</a> for more information.`,Nt,qe,xn='Note that <a href="/docs/transformers/pr_34362/en/model_doc/instructblip#transformers.InstructBlipQFormerModel">InstructBlipQFormerModel</a> is very similar to <a href="/docs/transformers/pr_34362/en/model_doc/bert#transformers.BertLMHeadModel">BertLMHeadModel</a> with interleaved cross-attention.',Rt,E,bt,ge,yt,C,he,Ht,Qe,jn=`Constructs an InstructBLIP processor which wraps a BLIP image processor and a LLaMa/T5 tokenizer into a single
	processor.`,St,Ge,kn=`<a href="/docs/transformers/pr_34362/en/model_doc/instructblip#transformers.InstructBlipProcessor">InstructBlipProcessor</a> offers all the functionalities of <a href="/docs/transformers/pr_34362/en/model_doc/blip#transformers.BlipImageProcessor">BlipImageProcessor</a> and <a href="/docs/transformers/pr_34362/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See the
	docstring of <code>__call__()</code> and <a href="/docs/transformers/pr_34362/en/model_doc/blip#transformers.BlipProcessor.decode">decode()</a> for more information.`,Et,X,_e,Xt,Ve,Fn=`This method forwards all its arguments to PreTrainedTokenizer’s <a href="/docs/transformers/pr_34362/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.batch_decode">batch_decode()</a>. Please
	refer to the docstring of this method for more information.`,Yt,Y,be,At,Le,Jn=`This method forwards all its arguments to PreTrainedTokenizer’s <a href="/docs/transformers/pr_34362/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.decode">decode()</a>. Please refer to
	the docstring of this method for more information.`,vt,ye,Mt,L,ve,Dt,Q,Me,Ot,Ne,Un='The <a href="/docs/transformers/pr_34362/en/model_doc/instructblip#transformers.InstructBlipVisionModel">InstructBlipVisionModel</a> forward method, overrides the <code>__call__</code> special method.',Kt,A,It,Ie,Bt,Z,Be,en,Re,zn=`Querying Transformer (Q-Former), used in InstructBLIP. Slightly modified from BLIP-2 as it also takes the
	instruction as input.`,tn,G,$e,nn,He,Wn=`encoder_hidden_states (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, hidden_size)</code>, <em>optional</em>):
	Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
	the model is configured as a decoder.
	encoder_attention_mask (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>):
	Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
	the cross-attention if the model is configured as a decoder. Mask values selected in <code>[0, 1]</code>:`,on,Se,Zn=`<li>1 for tokens that are <strong>not masked</strong>,</li> <li>0 for tokens that are <strong>masked</strong>.
	past_key_values (<code>tuple(tuple(torch.FloatTensor))</code> of length <code>config.n_layers</code> with each tuple having 4 tensors of:
	shape <code>(batch_size, num_heads, sequence_length - 1, embed_size_per_head)</code>): Contains precomputed key and
	value hidden states of the attention blocks. Can be used to speed up decoding. If <code>past_key_values</code> are
	used, the user can optionally input only the last <code>decoder_input_ids</code> (those that don’t have their past key
	value states given to this model) of shape <code>(batch_size, 1)</code> instead of all <code>decoder_input_ids</code> of shape
	<code>(batch_size, sequence_length)</code>.
	use_cache (<code>bool</code>, <em>optional</em>):
	If set to <code>True</code>, <code>past_key_values</code> key value states are returned and can be used to speed up decoding (see
	<code>past_key_values</code>).</li>`,$t,we,wt,B,Te,rn,Ee,Pn=`InstructBLIP Model for generating text given an image and an optional text prompt. The model consists of a vision
	encoder, Querying Transformer (Q-Former) and a language model.`,sn,Xe,qn=`One can optionally pass <code>input_ids</code> to the model, which serve as a text prompt, to make the language model continue
	the prompt. Otherwise, the language model starts generating text from the [BOS] (beginning-of-sequence) token.`,an,Ye,Qn=`This model inherits from <a href="/docs/transformers/pr_34362/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.)`,ln,Ae,Gn=`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.`,cn,W,Ce,dn,De,Vn='The <a href="/docs/transformers/pr_34362/en/model_doc/instructblip#transformers.InstructBlipForConditionalGeneration">InstructBlipForConditionalGeneration</a> forward method, overrides the <code>__call__</code> special method.',pn,D,mn,O,un,K,xe,fn,Oe,Ln="Overrides <code>generate</code> function to be able to use the model as a conditional generator.",Tt,je,Ct,ot,xt;return v=new V({props:{title:"InstructBLIP",local:"instructblip",headingTag:"h1"}}),M=new V({props:{title:"Overview",local:"overview",headingTag:"h2"}}),ae=new V({props:{title:"Usage tips",local:"usage-tips",headingTag:"h2"}}),le=new V({props:{title:"InstructBlipConfig",local:"transformers.InstructBlipConfig",headingTag:"h2"}}),ce=new k({props:{name:"class transformers.InstructBlipConfig",anchor:"transformers.InstructBlipConfig",parameters:[{name:"vision_config",val:" = None"},{name:"qformer_config",val:" = None"},{name:"text_config",val:" = None"},{name:"num_query_tokens",val:" = 32"},{name:"image_token_index",val:" = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.InstructBlipConfig.vision_config",description:`<strong>vision_config</strong> (<code>dict</code>, <em>optional</em>) —
	Dictionary of configuration options used to initialize <a href="/docs/transformers/pr_34362/en/model_doc/instructblip#transformers.InstructBlipVisionConfig">InstructBlipVisionConfig</a>.`,name:"vision_config"},{anchor:"transformers.InstructBlipConfig.qformer_config",description:`<strong>qformer_config</strong> (<code>dict</code>, <em>optional</em>) —
	Dictionary of configuration options used to initialize <a href="/docs/transformers/pr_34362/en/model_doc/instructblip#transformers.InstructBlipQFormerConfig">InstructBlipQFormerConfig</a>.`,name:"qformer_config"},{anchor:"transformers.InstructBlipConfig.text_config",description:`<strong>text_config</strong> (<code>dict</code>, <em>optional</em>) —
	Dictionary of configuration options used to initialize any <a href="/docs/transformers/pr_34362/en/main_classes/configuration#transformers.PretrainedConfig">PretrainedConfig</a>.`,name:"text_config"},{anchor:"transformers.InstructBlipConfig.num_query_tokens",description:`<strong>num_query_tokens</strong> (<code>int</code>, <em>optional</em>, defaults to 32) —
	The number of query tokens passed through the Transformer.`,name:"num_query_tokens"},{anchor:"transformers.InstructBlipConfig.image_token_index",description:`<strong>image_token_index</strong> (<code>int</code>, <em>optional</em>) —
	Token index of special image token.`,name:"image_token_index"},{anchor:"transformers.InstructBlipConfig.kwargs",description:`<strong>kwargs</strong> (<em>optional</em>) —
	Dictionary of keyword arguments.`,name:"kwargs"}],source:"https://github.com/huggingface/transformers/blob/vr_34362/src/transformers/models/instructblip/configuration_instructblip.py#L251"}}),R=new Jt({props:{anchor:"transformers.InstructBlipConfig.example",$$slots:{default:[Dn]},$$scope:{ctx:x}}}),de=new k({props:{name:"from_vision_qformer_text_configs",anchor:"transformers.InstructBlipConfig.from_vision_qformer_text_configs",parameters:[{name:"vision_config",val:": InstructBlipVisionConfig"},{name:"qformer_config",val:": InstructBlipQFormerConfig"},{name:"text_config",val:": PretrainedConfig"},{name:"**kwargs",val:""}],source:"https://github.com/huggingface/transformers/blob/vr_34362/src/transformers/models/instructblip/configuration_instructblip.py#L347",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_34362/en/model_doc/instructblip#transformers.InstructBlipConfig"
	>InstructBlipConfig</a></p>
	`}}),pe=new V({props:{title:"InstructBlipVisionConfig",local:"transformers.InstructBlipVisionConfig",headingTag:"h2"}}),me=new k({props:{name:"class transformers.InstructBlipVisionConfig",anchor:"transformers.InstructBlipVisionConfig",parameters:[{name:"hidden_size",val:" = 1408"},{name:"intermediate_size",val:" = 6144"},{name:"num_hidden_layers",val:" = 39"},{name:"num_attention_heads",val:" = 16"},{name:"image_size",val:" = 224"},{name:"patch_size",val:" = 14"},{name:"hidden_act",val:" = 'gelu'"},{name:"layer_norm_eps",val:" = 1e-06"},{name:"attention_dropout",val:" = 0.0"},{name:"initializer_range",val:" = 1e-10"},{name:"qkv_bias",val:" = True"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.InstructBlipVisionConfig.hidden_size",description:`<strong>hidden_size</strong> (<code>int</code>, <em>optional</em>, defaults to 1408) —
	Dimensionality of the encoder layers and the pooler layer.`,name:"hidden_size"},{anchor:"transformers.InstructBlipVisionConfig.intermediate_size",description:`<strong>intermediate_size</strong> (<code>int</code>, <em>optional</em>, defaults to 6144) —
	Dimensionality of the “intermediate” (i.e., feed-forward) layer in the Transformer encoder.`,name:"intermediate_size"},{anchor:"transformers.InstructBlipVisionConfig.num_hidden_layers",description:`<strong>num_hidden_layers</strong> (<code>int</code>, <em>optional</em>, defaults to 39) —
	Number of hidden layers in the Transformer encoder.`,name:"num_hidden_layers"},{anchor:"transformers.InstructBlipVisionConfig.num_attention_heads",description:`<strong>num_attention_heads</strong> (<code>int</code>, <em>optional</em>, defaults to 16) —
	Number of attention heads for each attention layer in the Transformer encoder.`,name:"num_attention_heads"},{anchor:"transformers.InstructBlipVisionConfig.image_size",description:`<strong>image_size</strong> (<code>int</code>, <em>optional</em>, defaults to 224) —
	The size (resolution) of each image.`,name:"image_size"},{anchor:"transformers.InstructBlipVisionConfig.patch_size",description:`<strong>patch_size</strong> (<code>int</code>, <em>optional</em>, defaults to 14) —
	The size (resolution) of each patch.`,name:"patch_size"},{anchor:"transformers.InstructBlipVisionConfig.hidden_act",description:`<strong>hidden_act</strong> (<code>str</code> or <code>function</code>, <em>optional</em>, defaults to <code>"gelu"</code>) —
	The non-linear activation function (function or string) in the encoder and pooler. If string, <code>"gelu"</code>,
	<code>"relu"</code>, <code>"selu"</code> and <code>"gelu_new"</code> <code>"gelu"</code> are supported. to 1e-5): The epsilon used by the layer
	normalization layers.`,name:"hidden_act"},{anchor:"transformers.InstructBlipVisionConfig.layer_norm_eps",description:`<strong>layer_norm_eps</strong> (<code>float</code>, <em>optional</em>, defaults to 1e-06) —
	The epsilon used by the layer normalization layers.`,name:"layer_norm_eps"},{anchor:"transformers.InstructBlipVisionConfig.attention_dropout",description:`<strong>attention_dropout</strong> (<code>float</code>, <em>optional</em>, defaults to 0.0) —
	The dropout ratio for the attention probabilities.`,name:"attention_dropout"},{anchor:"transformers.InstructBlipVisionConfig.initializer_range",description:`<strong>initializer_range</strong> (<code>float</code>, <em>optional</em>, defaults to 1e-10) —
	The standard deviation of the truncated_normal_initializer for initializing all weight matrices.`,name:"initializer_range"},{anchor:"transformers.InstructBlipVisionConfig.qkv_bias",description:`<strong>qkv_bias</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) —
	Whether to add a bias to the queries and values in the self-attention layers.`,name:"qkv_bias"}],source:"https://github.com/huggingface/transformers/blob/vr_34362/src/transformers/models/instructblip/configuration_instructblip.py#L29"}}),S=new Jt({props:{anchor:"transformers.InstructBlipVisionConfig.example",$$slots:{default:[On]},$$scope:{ctx:x}}}),ue=new V({props:{title:"InstructBlipQFormerConfig",local:"transformers.InstructBlipQFormerConfig",headingTag:"h2"}}),fe=new k({props:{name:"class transformers.InstructBlipQFormerConfig",anchor:"transformers.InstructBlipQFormerConfig",parameters:[{name:"vocab_size",val:" = 30522"},{name:"hidden_size",val:" = 768"},{name:"num_hidden_layers",val:" = 12"},{name:"num_attention_heads",val:" = 12"},{name:"intermediate_size",val:" = 3072"},{name:"hidden_act",val:" = 'gelu'"},{name:"hidden_dropout_prob",val:" = 0.1"},{name:"attention_probs_dropout_prob",val:" = 0.1"},{name:"max_position_embeddings",val:" = 512"},{name:"initializer_range",val:" = 0.02"},{name:"layer_norm_eps",val:" = 1e-12"},{name:"pad_token_id",val:" = 0"},{name:"position_embedding_type",val:" = 'absolute'"},{name:"cross_attention_frequency",val:" = 2"},{name:"encoder_hidden_size",val:" = 1408"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.InstructBlipQFormerConfig.vocab_size",description:`<strong>vocab_size</strong> (<code>int</code>, <em>optional</em>, defaults to 30522) —
	Vocabulary size of the Q-Former model. Defines the number of different tokens that can be represented by
	the <code>inputs_ids</code> passed when calling the model.`,name:"vocab_size"},{anchor:"transformers.InstructBlipQFormerConfig.hidden_size",description:`<strong>hidden_size</strong> (<code>int</code>, <em>optional</em>, defaults to 768) —
	Dimensionality of the encoder layers and the pooler layer.`,name:"hidden_size"},{anchor:"transformers.InstructBlipQFormerConfig.num_hidden_layers",description:`<strong>num_hidden_layers</strong> (<code>int</code>, <em>optional</em>, defaults to 12) —
	Number of hidden layers in the Transformer encoder.`,name:"num_hidden_layers"},{anchor:"transformers.InstructBlipQFormerConfig.num_attention_heads",description:`<strong>num_attention_heads</strong> (<code>int</code>, <em>optional</em>, defaults to 12) —
	Number of attention heads for each attention layer in the Transformer encoder.`,name:"num_attention_heads"},{anchor:"transformers.InstructBlipQFormerConfig.intermediate_size",description:`<strong>intermediate_size</strong> (<code>int</code>, <em>optional</em>, defaults to 3072) —
	Dimensionality of the “intermediate” (often named feed-forward) layer in the Transformer encoder.`,name:"intermediate_size"},{anchor:"transformers.InstructBlipQFormerConfig.hidden_act",description:`<strong>hidden_act</strong> (<code>str</code> or <code>Callable</code>, <em>optional</em>, defaults to <code>"gelu"</code>) —
	The non-linear activation function (function or string) in the encoder and pooler. If string, <code>"gelu"</code>,
	<code>"relu"</code>, <code>"silu"</code> and <code>"gelu_new"</code> are supported.`,name:"hidden_act"},{anchor:"transformers.InstructBlipQFormerConfig.hidden_dropout_prob",description:`<strong>hidden_dropout_prob</strong> (<code>float</code>, <em>optional</em>, defaults to 0.1) —
	The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.`,name:"hidden_dropout_prob"},{anchor:"transformers.InstructBlipQFormerConfig.attention_probs_dropout_prob",description:`<strong>attention_probs_dropout_prob</strong> (<code>float</code>, <em>optional</em>, defaults to 0.1) —
	The dropout ratio for the attention probabilities.`,name:"attention_probs_dropout_prob"},{anchor:"transformers.InstructBlipQFormerConfig.max_position_embeddings",description:`<strong>max_position_embeddings</strong> (<code>int</code>, <em>optional</em>, defaults to 512) —
	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.InstructBlipQFormerConfig.initializer_range",description:`<strong>initializer_range</strong> (<code>float</code>, <em>optional</em>, defaults to 0.02) —
	The standard deviation of the truncated_normal_initializer for initializing all weight matrices.`,name:"initializer_range"},{anchor:"transformers.InstructBlipQFormerConfig.layer_norm_eps",description:`<strong>layer_norm_eps</strong> (<code>float</code>, <em>optional</em>, defaults to 1e-12) —
	The epsilon used by the layer normalization layers.`,name:"layer_norm_eps"},{anchor:"transformers.InstructBlipQFormerConfig.pad_token_id",description:`<strong>pad_token_id</strong> (<code>int</code>, <em>optional</em>, defaults to 0) —
	Token id used for padding sequences.`,name:"pad_token_id"},{anchor:"transformers.InstructBlipQFormerConfig.position_embedding_type",description:`<strong>position_embedding_type</strong> (<code>str</code>, <em>optional</em>, defaults to <code>"absolute"</code>) —
	Type of position embedding. Choose one of <code>"absolute"</code>, <code>"relative_key"</code>, <code>"relative_key_query"</code>. For
	positional embeddings use <code>"absolute"</code>. For more information on <code>"relative_key"</code>, please refer to
	<a href="https://arxiv.org/abs/1803.02155" rel="nofollow">Self-Attention with Relative Position Representations (Shaw et al.)</a>.
	For more information on <code>"relative_key_query"</code>, please refer to <em>Method 4</em> in <a href="https://arxiv.org/abs/2009.13658" rel="nofollow">Improve Transformer Models
	with Better Relative Position Embeddings (Huang et al.)</a>.`,name:"position_embedding_type"},{anchor:"transformers.InstructBlipQFormerConfig.cross_attention_frequency",description:`<strong>cross_attention_frequency</strong> (<code>int</code>, <em>optional</em>, defaults to 2) —
	The frequency of adding cross-attention to the Transformer layers.`,name:"cross_attention_frequency"},{anchor:"transformers.InstructBlipQFormerConfig.encoder_hidden_size",description:`<strong>encoder_hidden_size</strong> (<code>int</code>, <em>optional</em>, defaults to 1408) —
	The hidden size of the hidden states for cross-attention.`,name:"encoder_hidden_size"}],source:"https://github.com/huggingface/transformers/blob/vr_34362/src/transformers/models/instructblip/configuration_instructblip.py#L130"}}),E=new Jt({props:{anchor:"transformers.InstructBlipQFormerConfig.example",$$slots:{default:[Kn]},$$scope:{ctx:x}}}),ge=new V({props:{title:"InstructBlipProcessor",local:"transformers.InstructBlipProcessor",headingTag:"h2"}}),he=new k({props:{name:"class transformers.InstructBlipProcessor",anchor:"transformers.InstructBlipProcessor",parameters:[{name:"image_processor",val:""},{name:"tokenizer",val:""},{name:"qformer_tokenizer",val:""},{name:"num_query_tokens",val:" = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.InstructBlipProcessor.image_processor",description:`<strong>image_processor</strong> (<code>BlipImageProcessor</code>) —
	An instance of <a href="/docs/transformers/pr_34362/en/model_doc/blip#transformers.BlipImageProcessor">BlipImageProcessor</a>. The image processor is a required input.`,name:"image_processor"},{anchor:"transformers.InstructBlipProcessor.tokenizer",description:"<strong>tokenizer</strong> (<code>AutoTokenizer</code>) —\nAn instance of [‘PreTrainedTokenizer`]. The tokenizer is a required input.",name:"tokenizer"},{anchor:"transformers.InstructBlipProcessor.qformer_tokenizer",description:"<strong>qformer_tokenizer</strong> (<code>AutoTokenizer</code>) —\nAn instance of [‘PreTrainedTokenizer`]. The Q-Former tokenizer is a required input.",name:"qformer_tokenizer"},{anchor:"transformers.InstructBlipProcessor.num_query_tokens",description:`<strong>num_query_tokens</strong> (<code>int</code>, <em>optional</em>) —”
	Number of tokens used by the Qformer as queries, should be same as in model’s config.`,name:"num_query_tokens"}],source:"https://github.com/huggingface/transformers/blob/vr_34362/src/transformers/models/instructblip/processing_instructblip.py#L55"}}),_e=new k({props:{name:"batch_decode",anchor:"transformers.InstructBlipProcessor.batch_decode",parameters:[{name:"args",val:""},{name:"kwargs",val:""}],source:"https://github.com/huggingface/transformers/blob/vr_34362/src/transformers/models/instructblip/processing_instructblip.py#L166"}}),be=new k({props:{name:"decode",anchor:"transformers.InstructBlipProcessor.decode",parameters:[{name:"args",val:""},{name:"**kwargs",val:""}],source:"https://github.com/huggingface/transformers/blob/vr_34362/src/transformers/models/instructblip/processing_instructblip.py#L174"}}),ye=new V({props:{title:"InstructBlipVisionModel",local:"transformers.InstructBlipVisionModel",headingTag:"h2"}}),ve=new k({props:{name:"class transformers.InstructBlipVisionModel",anchor:"transformers.InstructBlipVisionModel",parameters:[{name:"config",val:": InstructBlipVisionConfig"}],source:"https://github.com/huggingface/transformers/blob/vr_34362/src/transformers/models/instructblip/modeling_instructblip.py#L535"}}),Me=new k({props:{name:"forward",anchor:"transformers.InstructBlipVisionModel.forward",parameters:[{name:"pixel_values",val:": Optional = None"},{name:"output_attentions",val:": Optional = None"},{name:"output_hidden_states",val:": Optional = None"},{name:"return_dict",val:": Optional = None"},{name:"interpolate_pos_encoding",val:": bool = False"}],parametersDescription:[{anchor:"transformers.InstructBlipVisionModel.forward.pixel_values",description:`<strong>pixel_values</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, num_channels, height, width)</code>) —
	Pixel values. Pixel values can be obtained using <a href="/docs/transformers/pr_34362/en/model_doc/instructblip#transformers.InstructBlipProcessor">InstructBlipProcessor</a>. See
	<code>InstructBlipProcessor.__call__()</code> for details.`,name:"pixel_values"},{anchor:"transformers.InstructBlipVisionModel.forward.output_attentions",description:`<strong>output_attentions</strong> (<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.`,name:"output_attentions"},{anchor:"transformers.InstructBlipVisionModel.forward.output_hidden_states",description:`<strong>output_hidden_states</strong> (<code>bool</code>, <em>optional</em>) —
	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.InstructBlipVisionModel.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) —
	Whether or not to return a <a href="/docs/transformers/pr_34362/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"},{anchor:"transformers.InstructBlipVisionModel.forward.interpolate_pos_encoding",description:`<strong>interpolate_pos_encoding</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) —
	Whether to interpolate the pre-trained position encodings.`,name:"interpolate_pos_encoding"}],source:"https://github.com/huggingface/transformers/blob/vr_34362/src/transformers/models/instructblip/modeling_instructblip.py#L550",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


	<p>A <a
	href="/docs/transformers/pr_34362/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><class 'transformers.models.instructblip.configuration_instructblip.InstructBlipVisionConfig'></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_34362/en/main_classes/output#transformers.modeling_outputs.BaseModelOutputWithPooling"
	>transformers.modeling_outputs.BaseModelOutputWithPooling</a> or <code>tuple(torch.FloatTensor)</code></p>
	`}}),A=new Nn({props:{$$slots:{default:[eo]},$$scope:{ctx:x}}}),Ie=new V({props:{title:"InstructBlipQFormerModel",local:"transformers.InstructBlipQFormerModel",headingTag:"h2"}}),Be=new k({props:{name:"class transformers.InstructBlipQFormerModel",anchor:"transformers.InstructBlipQFormerModel",parameters:[{name:"config",val:": InstructBlipQFormerConfig"}],source:"https://github.com/huggingface/transformers/blob/vr_34362/src/transformers/models/instructblip/modeling_instructblip.py#L1081"}}),$e=new k({props:{name:"forward",anchor:"transformers.InstructBlipQFormerModel.forward",parameters:[{name:"input_ids",val:": LongTensor"},{name:"attention_mask",val:": Optional = None"},{name:"position_ids",val:": Optional = None"},{name:"query_embeds",val:": Optional = None"},{name:"head_mask",val:": Optional = None"},{name:"encoder_hidden_states",val:": Optional = None"},{name:"encoder_attention_mask",val:": Optional = None"},{name:"past_key_values",val:": Optional = None"},{name:"use_cache",val:": Optional = None"},{name:"output_attentions",val:": Optional = None"},{name:"output_hidden_states",val:": Optional = None"},{name:"return_dict",val:": Optional = None"}],source:"https://github.com/huggingface/transformers/blob/vr_34362/src/transformers/models/instructblip/modeling_instructblip.py#L1154"}}),we=new V({props:{title:"InstructBlipForConditionalGeneration",local:"transformers.InstructBlipForConditionalGeneration",headingTag:"h2"}}),Te=new k({props:{name:"class transformers.InstructBlipForConditionalGeneration",anchor:"transformers.InstructBlipForConditionalGeneration",parameters:[{name:"config",val:": InstructBlipConfig"}],parametersDescription:[{anchor:"transformers.InstructBlipForConditionalGeneration.config",description:`<strong>config</strong> (<a href="/docs/transformers/pr_34362/en/model_doc/instructblip#transformers.InstructBlipConfig">InstructBlipConfig</a>) — 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_34362/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_34362/src/transformers/models/instructblip/modeling_instructblip.py#L1277"}}),Ce=new k({props:{name:"forward",anchor:"transformers.InstructBlipForConditionalGeneration.forward",parameters:[{name:"pixel_values",val:": FloatTensor"},{name:"qformer_input_ids",val:": FloatTensor"},{name:"qformer_attention_mask",val:": Optional = None"},{name:"input_ids",val:": Optional = None"},{name:"attention_mask",val:": Optional = None"},{name:"decoder_input_ids",val:": Optional = None"},{name:"decoder_attention_mask",val:": Optional = None"},{name:"output_attentions",val:": Optional = None"},{name:"output_hidden_states",val:": Optional = None"},{name:"labels",val:": Optional = None"},{name:"return_dict",val:": Optional = None"},{name:"interpolate_pos_encoding",val:": bool = False"}],parametersDescription:[{anchor:"transformers.InstructBlipForConditionalGeneration.forward.pixel_values",description:`<strong>pixel_values</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, num_channels, height, width)</code>) —
	Pixel values. Pixel values can be obtained using <a href="/docs/transformers/pr_34362/en/model_doc/instructblip#transformers.InstructBlipProcessor">InstructBlipProcessor</a>. See
	<code>InstructBlipProcessor.__call__()</code> for details.`,name:"pixel_values"},{anchor:"transformers.InstructBlipForConditionalGeneration.forward.qformer_input_ids",description:`<strong>qformer_input_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) —
	Indices of input sequence tokens in the vocabulary of the Q-Former. Input tokens can optionally be provided
	to serve as text prompt, which the Q-Former model will encode.</p>
	<p>Indices can be obtained using <a href="/docs/transformers/pr_34362/en/model_doc/instructblip#transformers.InstructBlipProcessor">InstructBlipProcessor</a>. See <code>InstructBlipProcessor.__call__()</code> for
	details.</p>
	<p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"qformer_input_ids"},{anchor:"transformers.InstructBlipForConditionalGeneration.forward.qformer_attention_mask",description:`<strong>qformer_attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) —
	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:"qformer_attention_mask"},{anchor:"transformers.InstructBlipForConditionalGeneration.forward.input_ids",description:`<strong>input_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) —
	Indices of input sequence tokens in the vocabulary of the language model. Input tokens can optionally be
	provided to serve as text prompt, which the language model can continue.</p>
	<p>Indices can be obtained using <a href="/docs/transformers/pr_34362/en/model_doc/instructblip#transformers.InstructBlipProcessor">InstructBlipProcessor</a>. See <code>InstructBlipProcessor.__call__()</code> for
	details.</p>
	<p><a href="../glossary#input-ids">What are input IDs?</a>`,name:"input_ids"},{anchor:"transformers.InstructBlipForConditionalGeneration.forward.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_length)</code>, <em>optional</em>) —
	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.InstructBlipForConditionalGeneration.forward.decoder_input_ids",description:`<strong>decoder_input_ids</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size, target_sequence_length)</code>, <em>optional</em>) —
	Indices of decoder input sequence tokens in the vocabulary of the language model. Only relevant in case an
	encoder-decoder language model (like T5) is used.</p>
	<p>Indices can be obtained using <a href="/docs/transformers/pr_34362/en/model_doc/auto#transformers.AutoTokenizer">AutoTokenizer</a>. See <a href="/docs/transformers/pr_34362/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.encode">PreTrainedTokenizer.encode()</a> and
	<a href="/docs/transformers/pr_34362/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase.__call__">PreTrainedTokenizer.<strong>call</strong>()</a> for details. <a href="../glossary#decoder-input-ids">What are decoder input IDs?</a>`,name:"decoder_input_ids"},{anchor:"transformers.InstructBlipForConditionalGeneration.forward.decoder_attention_mask",description:`<strong>decoder_attention_mask</strong> (<code>torch.BoolTensor</code> of shape <code>(batch_size, target_sequence_length)</code>, <em>optional</em>) —
	Default behavior: generate a tensor that ignores pad tokens in <code>decoder_input_ids</code>. Causal mask will also
	be used by default.</p>
	<p>Only relevant in case an encoder-decoder language model (like T5) is used.`,name:"decoder_attention_mask"},{anchor:"transformers.InstructBlipForConditionalGeneration.forward.output_attentions",description:`<strong>output_attentions</strong> (<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.`,name:"output_attentions"},{anchor:"transformers.InstructBlipForConditionalGeneration.forward.output_hidden_states",description:`<strong>output_hidden_states</strong> (<code>bool</code>, <em>optional</em>) —
	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.InstructBlipForConditionalGeneration.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>) —
	Whether or not to return a <a href="/docs/transformers/pr_34362/en/main_classes/output#transformers.utils.ModelOutput">ModelOutput</a> instead of a plain tuple.`,name:"return_dict"},{anchor:"transformers.InstructBlipForConditionalGeneration.forward.interpolate_pos_encoding",description:`<strong>interpolate_pos_encoding</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) —
	Whether to interpolate the pre-trained position encodings.`,name:"interpolate_pos_encoding"},{anchor:"transformers.InstructBlipForConditionalGeneration.forward.labels",description:`<strong>labels</strong> (<code>torch.LongTensor</code> of shape <code>(batch_size,)</code>, <em>optional</em>) —
	Labels for computing the language modeling loss. Indices should be in <code>[-100, 0, ..., config.vocab_size - 1]</code>. All labels set to <code>-100</code> are ignored (masked), the loss is only computed for labels in <code>[0, ..., config.vocab_size]</code>`,name:"labels"}],source:"https://github.com/huggingface/transformers/blob/vr_34362/src/transformers/models/instructblip/modeling_instructblip.py#L1360",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


	<p>A <code>transformers.models.instructblip.modeling_instructblip.InstructBlipForConditionalGenerationModelOutput</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><class 'transformers.models.instructblip.configuration_instructblip.InstructBlipVisionConfig'></code>) and inputs.</p>
	<ul>
	<li><strong>loss</strong> (<code>torch.FloatTensor</code>, <em>optional</em>, returned when <code>labels</code> is provided, <code>torch.FloatTensor</code> of shape <code>(1,)</code>) — Language modeling loss from the language model.</li>
	<li><strong>logits</strong> (<code>torch.FloatTensor</code> of shape <code>(batch_size, sequence_length, config.vocab_size)</code>) — Prediction scores of the language modeling head of the language model.</li>
	<li><strong>vision_outputs</strong> (<code>BaseModelOutputWithPooling</code>) — Outputs of the vision encoder.</li>
	<li><strong>qformer_outputs</strong> (<code>BaseModelOutputWithPoolingAndCrossAttentions</code>) — Outputs of the Q-Former (Querying Transformer).</li>
	<li><strong>language_model_outputs</strong> (<code>CausalLMOutputWithPast</code> or <code>Seq2SeqLMOutput</code>) — Outputs of the language model.</li>
	</ul>
	`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


	<p><code>transformers.models.instructblip.modeling_instructblip.InstructBlipForConditionalGenerationModelOutput</code> or <code>tuple(torch.FloatTensor)</code></p>
	`}}),D=new Nn({props:{$$slots:{default:[to]},$$scope:{ctx:x}}}),O=new Jt({props:{anchor:"transformers.InstructBlipForConditionalGeneration.forward.example",$$slots:{default:[no]},$$scope:{ctx:x}}}),xe=new k({props:{name:"generate",anchor:"transformers.InstructBlipForConditionalGeneration.generate",parameters:[{name:"pixel_values",val:": FloatTensor"},{name:"qformer_input_ids",val:": Optional = None"},{name:"qformer_attention_mask",val:": Optional = None"},{name:"input_ids",val:": Optional = None"},{name:"attention_mask",val:": Optional = None"},{name:"interpolate_pos_encoding",val:": bool = False"},{name:"**generate_kwargs",val:""}],parametersDescription:[{anchor:"transformers.InstructBlipForConditionalGeneration.generate.pixel_values",description:`<strong>pixel_values</strong> (<code>torch.FloatTensor</code> of shape (batch_size, num_channels, height, width)) —
	Input images to be processed.`,name:"pixel_values"},{anchor:"transformers.InstructBlipForConditionalGeneration.generate.qformer_input_ids",description:`<strong>qformer_input_ids</strong> (<code>torch.LongTensor</code> of shape (batch_size, sequence_length), <em>optional</em>) —
	The sequence used as a prompt to be fed to the Q-Former module.`,name:"qformer_input_ids"},{anchor:"transformers.InstructBlipForConditionalGeneration.generate.qformer_attention_mask",description:`<strong>qformer_attention_mask</strong> (<code>torch.LongTensor</code> of shape (batch_size, sequence_length), <em>optional</em>) —
	Mask to avoid performing attention on padding token indices.`,name:"qformer_attention_mask"},{anchor:"transformers.InstructBlipForConditionalGeneration.generate.input_ids",description:`<strong>input_ids</strong> (<code>torch.LongTensor</code> of shape (batch_size, sequence_length), <em>optional</em>) —
	The sequence used as a prompt for the generation.`,name:"input_ids"},{anchor:"transformers.InstructBlipForConditionalGeneration.generate.attention_mask",description:`<strong>attention_mask</strong> (<code>torch.LongTensor</code> of shape (batch_size, sequence_length), <em>optional</em>) —
	Mask to avoid performing attention on padding token indices.`,name:"attention_mask"},{anchor:"transformers.InstructBlipForConditionalGeneration.generate.interpolate_pos_encoding",description:`<strong>interpolate_pos_encoding</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) —
	Whether to interpolate the positional encoding of the image embeddings.`,name:"interpolate_pos_encoding"}],source:"https://github.com/huggingface/transformers/blob/vr_34362/src/transformers/models/instructblip/modeling_instructblip.py#L1529",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


	<p>A list of strings of length batch_size * num_captions.</p>
	`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


	<p>captions (list)</p>
	`}}),je=new An({props:{source:"https://github.com/huggingface/transformers/blob/main/docs/source/en/model_doc/instructblip.md"}}),{c(){i=c("meta"),I=o(),m=c("p"),p=o(),u(v.$$.fragment),s=o(),u(M.$$.fragment),rt=o(),te=c("p"),te.innerHTML=gn,st=o(),ne=c("p"),ne.textContent=hn,at=o(),oe=c("p"),oe.innerHTML=_n,it=o(),N=c("img"),lt=o(),re=c("small"),re.innerHTML=yn,ct=o(),se=c("p"),se.innerHTML=vn,dt=o(),u(ae.$$.fragment),pt=o(),ie=c("p"),ie.innerHTML=Mn,mt=o(),u(le.$$.fragment),ut=o(),T=c("div"),u(ce.$$.fragment),zt=o(),Je=c("p"),Je.innerHTML=In,Wt=o(),Ue=c("p"),Ue.innerHTML=Bn,Zt=o(),u(R.$$.fragment),Pt=o(),H=c("div"),u(de.$$.fragment),qt=o(),ze=c("p"),ze.innerHTML=$n,ft=o(),u(pe.$$.fragment),gt=o(),F=c("div"),u(me.$$.fragment),Qt=o(),We=c("p"),We.innerHTML=wn,Gt=o(),Ze=c("p"),Ze.innerHTML=Tn,Vt=o(),u(S.$$.fragment),ht=o(),u(ue.$$.fragment),_t=o(),J=c("div"),u(fe.$$.fragment),Lt=o(),Pe=c("p"),Pe.innerHTML=Cn,Nt=o(),qe=c("p"),qe.innerHTML=xn,Rt=o(),u(E.$$.fragment),bt=o(),u(ge.$$.fragment),yt=o(),C=c("div"),u(he.$$.fragment),Ht=o(),Qe=c("p"),Qe.textContent=jn,St=o(),Ge=c("p"),Ge.innerHTML=kn,Et=o(),X=c("div"),u(_e.$$.fragment),Xt=o(),Ve=c("p"),Ve.innerHTML=Fn,Yt=o(),Y=c("div"),u(be.$$.fragment),At=o(),Le=c("p"),Le.innerHTML=Jn,vt=o(),u(ye.$$.fragment),Mt=o(),L=c("div"),u(ve.$$.fragment),Dt=o(),Q=c("div"),u(Me.$$.fragment),Ot=o(),Ne=c("p"),Ne.innerHTML=Un,Kt=o(),u(A.$$.fragment),It=o(),u(Ie.$$.fragment),Bt=o(),Z=c("div"),u(Be.$$.fragment),en=o(),Re=c("p"),Re.textContent=zn,tn=o(),G=c("div"),u($e.$$.fragment),nn=o(),He=c("p"),He.innerHTML=Wn,on=o(),Se=c("ul"),Se.innerHTML=Zn,$t=o(),u(we.$$.fragment),wt=o(),B=c("div"),u(Te.$$.fragment),rn=o(),Ee=c("p"),Ee.textContent=Pn,sn=o(),Xe=c("p"),Xe.innerHTML=qn,an=o(),Ye=c("p"),Ye.innerHTML=Qn,ln=o(),Ae=c("p"),Ae.innerHTML=Gn,cn=o(),W=c("div"),u(Ce.$$.fragment),dn=o(),De=c("p"),De.innerHTML=Vn,pn=o(),u(D.$$.fragment),mn=o(),u(O.$$.fragment),un=o(),K=c("div"),u(xe.$$.fragment),fn=o(),Oe=c("p"),Oe.innerHTML=Ln,Tt=o(),u(je.$$.fragment),Ct=o(),ot=c("p"),this.h()},l(e){const n=Yn("svelte-u9bgzb",document.head);i=d(n,"META",{name:!0,content:!0}),n.forEach(t),I=r(e),m=d(e,"P",{}),w(m).forEach(t),p=r(e),f(v.$$.fragment,e),s=r(e),f(M.$$.fragment,e),rt=r(e),te=d(e,"P",{"data-svelte-h":!0}),y(te)!=="svelte-15j8z3m"&&(te.innerHTML=gn),st=r(e),ne=d(e,"P",{"data-svelte-h":!0}),y(ne)!=="svelte-vfdo9a"&&(ne.textContent=hn),at=r(e),oe=d(e,"P",{"data-svelte-h":!0}),y(oe)!=="svelte-g3w4hv"&&(oe.innerHTML=_n),it=r(e),N=d(e,"IMG",{src:!0,alt:!0,width:!0}),lt=r(e),re=d(e,"SMALL",{"data-svelte-h":!0}),y(re)!=="svelte-gvupf8"&&(re.innerHTML=yn),ct=r(e),se=d(e,"P",{"data-svelte-h":!0}),y(se)!=="svelte-1b5u17p"&&(se.innerHTML=vn),dt=r(e),f(ae.$$.fragment,e),pt=r(e),ie=d(e,"P",{"data-svelte-h":!0}),y(ie)!=="svelte-14dqybh"&&(ie.innerHTML=Mn),mt=r(e),f(le.$$.fragment,e),ut=r(e),T=d(e,"DIV",{class:!0});var U=w(T);f(ce.$$.fragment,U),zt=r(U),Je=d(U,"P",{"data-svelte-h":!0}),y(Je)!=="svelte-1pfd63i"&&(Je.innerHTML=In),Wt=r(U),Ue=d(U,"P",{"data-svelte-h":!0}),y(Ue)!=="svelte-24w6m3"&&(Ue.innerHTML=Bn),Zt=r(U),f(R.$$.fragment,U),Pt=r(U),H=d(U,"DIV",{class:!0});var ke=w(H);f(de.$$.fragment,ke),qt=r(ke),ze=d(ke,"P",{"data-svelte-h":!0}),y(ze)!=="svelte-1vz38xu"&&(ze.innerHTML=$n),ke.forEach(t),U.forEach(t),ft=r(e),f(pe.$$.fragment,e),gt=r(e),F=d(e,"DIV",{class:!0});var P=w(F);f(me.$$.fragment,P),Qt=r(P),We=d(P,"P",{"data-svelte-h":!0}),y(We)!=="svelte-j8f2u"&&(We.innerHTML=wn),Gt=r(P),Ze=d(P,"P",{"data-svelte-h":!0}),y(Ze)!=="svelte-24w6m3"&&(Ze.innerHTML=Tn),Vt=r(P),f(S.$$.fragment,P),P.forEach(t),ht=r(e),f(ue.$$.fragment,e),_t=r(e),J=d(e,"DIV",{class:!0});var q=w(J);f(fe.$$.fragment,q),Lt=r(q),Pe=d(q,"P",{"data-svelte-h":!0}),y(Pe)!=="svelte-3mikwu"&&(Pe.innerHTML=Cn),Nt=r(q),qe=d(q,"P",{"data-svelte-h":!0}),y(qe)!=="svelte-12zbptb"&&(qe.innerHTML=xn),Rt=r(q),f(E.$$.fragment,q),q.forEach(t),bt=r(e),f(ge.$$.fragment,e),yt=r(e),C=d(e,"DIV",{class:!0});var z=w(C);f(he.$$.fragment,z),Ht=r(z),Qe=d(z,"P",{"data-svelte-h":!0}),y(Qe)!=="svelte-1o16zs2"&&(Qe.textContent=jn),St=r(z),Ge=d(z,"P",{"data-svelte-h":!0}),y(Ge)!=="svelte-jvhl3v"&&(Ge.innerHTML=kn),Et=r(z),X=d(z,"DIV",{class:!0});var Fe=w(X);f(_e.$$.fragment,Fe),Xt=r(Fe),Ve=d(Fe,"P",{"data-svelte-h":!0}),y(Ve)!=="svelte-td3ljx"&&(Ve.innerHTML=Fn),Fe.forEach(t),Yt=r(z),Y=d(z,"DIV",{class:!0});var jt=w(Y);f(be.$$.fragment,jt),At=r(jt),Le=d(jt,"P",{"data-svelte-h":!0}),y(Le)!=="svelte-1a6khgr"&&(Le.innerHTML=Jn),jt.forEach(t),z.forEach(t),vt=r(e),f(ye.$$.fragment,e),Mt=r(e),L=d(e,"DIV",{class:!0});var kt=w(L);f(ve.$$.fragment,kt),Dt=r(kt),Q=d(kt,"DIV",{class:!0});var Ke=w(Q);f(Me.$$.fragment,Ke),Ot=r(Ke),Ne=d(Ke,"P",{"data-svelte-h":!0}),y(Ne)!=="svelte-1wqm69s"&&(Ne.innerHTML=Un),Kt=r(Ke),f(A.$$.fragment,Ke),Ke.forEach(t),kt.forEach(t),It=r(e),f(Ie.$$.fragment,e),Bt=r(e),Z=d(e,"DIV",{class:!0});var et=w(Z);f(Be.$$.fragment,et),en=r(et),Re=d(et,"P",{"data-svelte-h":!0}),y(Re)!=="svelte-1ta0rcm"&&(Re.textContent=zn),tn=r(et),G=d(et,"DIV",{class:!0});var tt=w(G);f($e.$$.fragment,tt),nn=r(tt),He=d(tt,"P",{"data-svelte-h":!0}),y(He)!=="svelte-1h74cdd"&&(He.innerHTML=Wn),on=r(tt),Se=d(tt,"UL",{"data-svelte-h":!0}),y(Se)!=="svelte-1yyv1jy"&&(Se.innerHTML=Zn),tt.forEach(t),et.forEach(t),$t=r(e),f(we.$$.fragment,e),wt=r(e),B=d(e,"DIV",{class:!0});var j=w(B);f(Te.$$.fragment,j),rn=r(j),Ee=d(j,"P",{"data-svelte-h":!0}),y(Ee)!=="svelte-18txus8"&&(Ee.textContent=Pn),sn=r(j),Xe=d(j,"P",{"data-svelte-h":!0}),y(Xe)!=="svelte-1ks26sg"&&(Xe.innerHTML=qn),an=r(j),Ye=d(j,"P",{"data-svelte-h":!0}),y(Ye)!=="svelte-9fxjcw"&&(Ye.innerHTML=Qn),ln=r(j),Ae=d(j,"P",{"data-svelte-h":!0}),y(Ae)!=="svelte-hswkmf"&&(Ae.innerHTML=Gn),cn=r(j),W=d(j,"DIV",{class:!0});var ee=w(W);f(Ce.$$.fragment,ee),dn=r(ee),De=d(ee,"P",{"data-svelte-h":!0}),y(De)!=="svelte-zcbang"&&(De.innerHTML=Vn),pn=r(ee),f(D.$$.fragment,ee),mn=r(ee),f(O.$$.fragment,ee),ee.forEach(t),un=r(j),K=d(j,"DIV",{class:!0});var Ft=w(K);f(xe.$$.fragment,Ft),fn=r(Ft),Oe=d(Ft,"P",{"data-svelte-h":!0}),y(Oe)!=="svelte-eq620n"&&(Oe.innerHTML=Ln),Ft.forEach(t),j.forEach(t),Tt=r(e),f(je.$$.fragment,e),Ct=r(e),ot=d(e,"P",{}),w(ot).forEach(t),this.h()},h(){$(i,"name","hf:doc:metadata"),$(i,"content",ro),Hn(N.src,bn="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/instructblip_architecture.jpg")\|\|$(N,"src",bn),$(N,"alt","drawing"),$(N,"width","600"),$(H,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),$(T,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),$(F,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),$(J,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),$(X,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),$(Y,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),$(C,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),$(Q,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),$(L,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),$(G,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),$(Z,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),$(W,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),$(K,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),$(B,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8")},m(e,n){a(document.head,i),l(e,I,n),l(e,m,n),l(e,p,n),g(v,e,n),l(e,s,n),g(M,e,n),l(e,rt,n),l(e,te,n),l(e,st,n),l(e,ne,n),l(e,at,n),l(e,oe,n),l(e,it,n),l(e,N,n),l(e,lt,n),l(e,re,n),l(e,ct,n),l(e,se,n),l(e,dt,n),g(ae,e,n),l(e,pt,n),l(e,ie,n),l(e,mt,n),g(le,e,n),l(e,ut,n),l(e,T,n),g(ce,T,null),a(T,zt),a(T,Je),a(T,Wt),a(T,Ue),a(T,Zt),g(R,T,null),a(T,Pt),a(T,H),g(de,H,null),a(H,qt),a(H,ze),l(e,ft,n),g(pe,e,n),l(e,gt,n),l(e,F,n),g(me,F,null),a(F,Qt),a(F,We),a(F,Gt),a(F,Ze),a(F,Vt),g(S,F,null),l(e,ht,n),g(ue,e,n),l(e,_t,n),l(e,J,n),g(fe,J,null),a(J,Lt),a(J,Pe),a(J,Nt),a(J,qe),a(J,Rt),g(E,J,null),l(e,bt,n),g(ge,e,n),l(e,yt,n),l(e,C,n),g(he,C,null),a(C,Ht),a(C,Qe),a(C,St),a(C,Ge),a(C,Et),a(C,X),g(_e,X,null),a(X,Xt),a(X,Ve),a(C,Yt),a(C,Y),g(be,Y,null),a(Y,At),a(Y,Le),l(e,vt,n),g(ye,e,n),l(e,Mt,n),l(e,L,n),g(ve,L,null),a(L,Dt),a(L,Q),g(Me,Q,null),a(Q,Ot),a(Q,Ne),a(Q,Kt),g(A,Q,null),l(e,It,n),g(Ie,e,n),l(e,Bt,n),l(e,Z,n),g(Be,Z,null),a(Z,en),a(Z,Re),a(Z,tn),a(Z,G),g($e,G,null),a(G,nn),a(G,He),a(G,on),a(G,Se),l(e,$t,n),g(we,e,n),l(e,wt,n),l(e,B,n),g(Te,B,null),a(B,rn),a(B,Ee),a(B,sn),a(B,Xe),a(B,an),a(B,Ye),a(B,ln),a(B,Ae),a(B,cn),a(B,W),g(Ce,W,null),a(W,dn),a(W,De),a(W,pn),g(D,W,null),a(W,mn),g(O,W,null),a(B,un),a(B,K),g(xe,K,null),a(K,fn),a(K,Oe),l(e,Tt,n),g(je,e,n),l(e,Ct,n),l(e,ot,n),xt=!0},p(e,[n]){const U={};n&2&&(U.$$scope={dirty:n,ctx:e}),R.$set(U);const ke={};n&2&&(ke.$$scope={dirty:n,ctx:e}),S.$set(ke);const P={};n&2&&(P.$$scope={dirty:n,ctx:e}),E.$set(P);const q={};n&2&&(q.$$scope={dirty:n,ctx:e}),A.$set(q);const z={};n&2&&(z.$$scope={dirty:n,ctx:e}),D.$set(z);const Fe={};n&2&&(Fe.$$scope={dirty:n,ctx:e}),O.$set(Fe)},i(e){xt\|\|(h(v.$$.fragment,e),h(M.$$.fragment,e),h(ae.$$.fragment,e),h(le.$$.fragment,e),h(ce.$$.fragment,e),h(R.$$.fragment,e),h(de.$$.fragment,e),h(pe.$$.fragment,e),h(me.$$.fragment,e),h(S.$$.fragment,e),h(ue.$$.fragment,e),h(fe.$$.fragment,e),h(E.$$.fragment,e),h(ge.$$.fragment,e),h(he.$$.fragment,e),h(_e.$$.fragment,e),h(be.$$.fragment,e),h(ye.$$.fragment,e),h(ve.$$.fragment,e),h(Me.$$.fragment,e),h(A.$$.fragment,e),h(Ie.$$.fragment,e),h(Be.$$.fragment,e),h($e.$$.fragment,e),h(we.$$.fragment,e),h(Te.$$.fragment,e),h(Ce.$$.fragment,e),h(D.$$.fragment,e),h(O.$$.fragment,e),h(xe.$$.fragment,e),h(je.$$.fragment,e),xt=!0)},o(e){_(v.$$.fragment,e),_(M.$$.fragment,e),_(ae.$$.fragment,e),_(le.$$.fragment,e),_(ce.$$.fragment,e),_(R.$$.fragment,e),_(de.$$.fragment,e),_(pe.$$.fragment,e),_(me.$$.fragment,e),_(S.$$.fragment,e),_(ue.$$.fragment,e),_(fe.$$.fragment,e),_(E.$$.fragment,e),_(ge.$$.fragment,e),_(he.$$.fragment,e),_(_e.$$.fragment,e),_(be.$$.fragment,e),_(ye.$$.fragment,e),_(ve.$$.fragment,e),_(Me.$$.fragment,e),_(A.$$.fragment,e),_(Ie.$$.fragment,e),_(Be.$$.fragment,e),_($e.$$.fragment,e),_(we.$$.fragment,e),_(Te.$$.fragment,e),_(Ce.$$.fragment,e),_(D.$$.fragment,e),_(O.$$.fragment,e),_(xe.$$.fragment,e),_(je.$$.fragment,e),xt=!1},d(e){e&&(t(I),t(m),t(p),t(s),t(rt),t(te),t(st),t(ne),t(at),t(oe),t(it),t(N),t(lt),t(re),t(ct),t(se),t(dt),t(pt),t(ie),t(mt),t(ut),t(T),t(ft),t(gt),t(F),t(ht),t(_t),t(J),t(bt),t(yt),t(C),t(vt),t(Mt),t(L),t(It),t(Bt),t(Z),t($t),t(wt),t(B),t(Tt),t(Ct),t(ot)),t(i),b(v,e),b(M,e),b(ae,e),b(le,e),b(ce),b(R),b(de),b(pe,e),b(me),b(S),b(ue,e),b(fe),b(E),b(ge,e),b(he),b(_e),b(be),b(ye,e),b(ve),b(Me),b(A),b(Ie,e),b(Be),b($e),b(we,e),b(Te),b(Ce),b(D),b(O),b(xe),b(je,e)}}}const ro='{"title":"InstructBLIP","local":"instructblip","sections":[{"title":"Overview","local":"overview","sections":[],"depth":2},{"title":"Usage tips","local":"usage-tips","sections":[],"depth":2},{"title":"InstructBlipConfig","local":"transformers.InstructBlipConfig","sections":[],"depth":2},{"title":"InstructBlipVisionConfig","local":"transformers.InstructBlipVisionConfig","sections":[],"depth":2},{"title":"InstructBlipQFormerConfig","local":"transformers.InstructBlipQFormerConfig","sections":[],"depth":2},{"title":"InstructBlipProcessor","local":"transformers.InstructBlipProcessor","sections":[],"depth":2},{"title":"InstructBlipVisionModel","local":"transformers.InstructBlipVisionModel","sections":[],"depth":2},{"title":"InstructBlipQFormerModel","local":"transformers.InstructBlipQFormerModel","sections":[],"depth":2},{"title":"InstructBlipForConditionalGeneration","local":"transformers.InstructBlipForConditionalGeneration","sections":[],"depth":2}],"depth":1}';function so(x){return Sn(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class fo extends En{constructor(i){super(),Xn(this,i,so,oo,Rn,{})}}export{fo as component};

Xet Storage Details

Size:: 74.9 kB
Xet hash:: a2ccd16123604aaed49d22087723e4a98ef20695ccf873c8a0f0f4cb49e9e935

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.