diffusers/pr_13331/en/_app/immutable/nodes/63.d9e4b19d.js

import{s as re,n as ae,o as ie}from"../chunks/scheduler.53228c21.js";import{S as le,i as de,e as d,s as n,c as f,h as me,a as m,d as t,b as r,f as F,g as c,j as L,k as P,l as W,m as s,n as u,t as p,o as h,p as _}from"../chunks/index.100fac89.js";import{C as fe}from"../chunks/CopyLLMTxtMenu.67e413d2.js";import{D as ne}from"../chunks/Docstring.60584164.js";import{C as ce}from"../chunks/CodeBlock.d30a6509.js";import{H as A,E as ue}from"../chunks/MermaidChart.svelte_svelte_type_style_lang.debde53c.js";function pe(K){let a,k,Z,U,g,z,b,N,M,ee='A 14B Real-Time Autogressive Diffusion Transformer model (support T2V, I2V and V2V) for 3D video-like data from <a href="https://github.com/PKU-YuanGroup/Helios" rel="nofollow">Helios</a> was introduced in <a href="https://huggingface.co/papers/2603.04379" rel="nofollow">Helios: Real Real-Time Long Video Generation Model</a> by Peking University &amp; ByteDance &amp; etc.',R,T,oe="The model can be loaded with the following code snippet.",I,y,V,v,C,i,$,S,w,te="A Transformer model for video-like data used in the Helios model.",B,x,E,l,D,O,J,se='The output of <a href="/docs/diffusers/pr_13331/en/api/models/transformer2d#diffusers.Transformer2DModel">Transformer2DModel</a>.',X,H,q,j,G;return g=new fe({props:{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"}}),b=new A({props:{title:"HeliosTransformer3DModel",local:"heliostransformer3dmodel",headingTag:"h1"}}),y=new ce({props:{code:"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",highlighted:`<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> HeliosTransformer3DModel

<span class="hljs-comment"># Best Quality</span>
transformer = HeliosTransformer3DModel.from_pretrained(<span class="hljs-string">&quot;BestWishYsh/Helios-Base&quot;</span>, subfolder=<span class="hljs-string">&quot;transformer&quot;</span>, torch_dtype=torch.bfloat16)
<span class="hljs-comment"># Intermediate Weight</span>
transformer = HeliosTransformer3DModel.from_pretrained(<span class="hljs-string">&quot;BestWishYsh/Helios-Mid&quot;</span>, subfolder=<span class="hljs-string">&quot;transformer&quot;</span>, torch_dtype=torch.bfloat16)
<span class="hljs-comment"># Best Efficiency</span>
transformer = HeliosTransformer3DModel.from_pretrained(<span class="hljs-string">&quot;BestWishYsh/Helios-Distilled&quot;</span>, subfolder=<span class="hljs-string">&quot;transformer&quot;</span>, torch_dtype=torch.bfloat16)`,wrap:!1}}),v=new A({props:{title:"HeliosTransformer3DModel",local:"diffusers.HeliosTransformer3DModel",headingTag:"h2"}}),$=new ne({props:{name:"class diffusers.HeliosTransformer3DModel",anchor:"diffusers.HeliosTransformer3DModel",parameters:[{name:"patch_size",val:": tuple = (1, 2, 2)"},{name:"num_attention_heads",val:": int = 40"},{name:"attention_head_dim",val:": int = 128"},{name:"in_channels",val:": int = 16"},{name:"out_channels",val:": int = 16"},{name:"text_dim",val:": int = 4096"},{name:"freq_dim",val:": int = 256"},{name:"ffn_dim",val:": int = 13824"},{name:"num_layers",val:": int = 40"},{name:"cross_attn_norm",val:": bool = True"},{name:"qk_norm",val:": str | None = 'rms_norm_across_heads'"},{name:"eps",val:": float = 1e-06"},{name:"added_kv_proj_dim",val:": int | None = None"},{name:"rope_dim",val:": tuple = (44, 42, 42)"},{name:"rope_theta",val:": float = 10000.0"},{name:"guidance_cross_attn",val:": bool = True"},{name:"zero_history_timestep",val:": bool = True"},{name:"has_multi_term_memory_patch",val:": bool = True"},{name:"is_amplify_history",val:": bool = False"},{name:"history_scale_mode",val:": str = 'per_head'"}],parametersDescription:[{anchor:"diffusers.HeliosTransformer3DModel.patch_size",description:`<strong>patch_size</strong> (<code>tuple[int]</code>, defaults to <code>(1, 2, 2)</code>) &#x2014;
3D patch dimensions for video embedding (t_patch, h_patch, w_patch).`,name:"patch_size"},{anchor:"diffusers.HeliosTransformer3DModel.num_attention_heads",description:`<strong>num_attention_heads</strong> (<code>int</code>, defaults to <code>40</code>) &#x2014;
Fixed length for text embeddings.`,name:"num_attention_heads"},{anchor:"diffusers.HeliosTransformer3DModel.attention_head_dim",description:`<strong>attention_head_dim</strong> (<code>int</code>, defaults to <code>128</code>) &#x2014;
The number of channels in each head.`,name:"attention_head_dim"},{anchor:"diffusers.HeliosTransformer3DModel.in_channels",description:`<strong>in_channels</strong> (<code>int</code>, defaults to <code>16</code>) &#x2014;
The number of channels in the input.`,name:"in_channels"},{anchor:"diffusers.HeliosTransformer3DModel.out_channels",description:`<strong>out_channels</strong> (<code>int</code>, defaults to <code>16</code>) &#x2014;
The number of channels in the output.`,name:"out_channels"},{anchor:"diffusers.HeliosTransformer3DModel.text_dim",description:`<strong>text_dim</strong> (<code>int</code>, defaults to <code>512</code>) &#x2014;
Input dimension for text embeddings.`,name:"text_dim"},{anchor:"diffusers.HeliosTransformer3DModel.freq_dim",description:`<strong>freq_dim</strong> (<code>int</code>, defaults to <code>256</code>) &#x2014;
Dimension for sinusoidal time embeddings.`,name:"freq_dim"},{anchor:"diffusers.HeliosTransformer3DModel.ffn_dim",description:`<strong>ffn_dim</strong> (<code>int</code>, defaults to <code>13824</code>) &#x2014;
Intermediate dimension in feed-forward network.`,name:"ffn_dim"},{anchor:"diffusers.HeliosTransformer3DModel.num_layers",description:`<strong>num_layers</strong> (<code>int</code>, defaults to <code>40</code>) &#x2014;
The number of layers of transformer blocks to use.`,name:"num_layers"},{anchor:"diffusers.HeliosTransformer3DModel.window_size",description:`<strong>window_size</strong> (<code>tuple[int]</code>, defaults to <code>(-1, -1)</code>) &#x2014;
Window size for local attention (-1 indicates global attention).`,name:"window_size"},{anchor:"diffusers.HeliosTransformer3DModel.cross_attn_norm",description:`<strong>cross_attn_norm</strong> (<code>bool</code>, defaults to <code>True</code>) &#x2014;
Enable cross-attention normalization.`,name:"cross_attn_norm"},{anchor:"diffusers.HeliosTransformer3DModel.qk_norm",description:`<strong>qk_norm</strong> (<code>bool</code>, defaults to <code>True</code>) &#x2014;
Enable query/key normalization.`,name:"qk_norm"},{anchor:"diffusers.HeliosTransformer3DModel.eps",description:`<strong>eps</strong> (<code>float</code>, defaults to <code>1e-6</code>) &#x2014;
Epsilon value for normalization layers.`,name:"eps"},{anchor:"diffusers.HeliosTransformer3DModel.add_img_emb",description:`<strong>add_img_emb</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether to use img_emb.`,name:"add_img_emb"},{anchor:"diffusers.HeliosTransformer3DModel.added_kv_proj_dim",description:`<strong>added_kv_proj_dim</strong> (<code>int</code>, <em>optional</em>, defaults to <code>None</code>) &#x2014;
The number of channels to use for the added key and value projections. If <code>None</code>, no projection is used.`,name:"added_kv_proj_dim"}],source:"https://github.com/huggingface/diffusers/blob/vr_13331/src/diffusers/models/transformers/transformer_helios.py#L497"}}),x=new A({props:{title:"Transformer2DModelOutput",local:"diffusers.models.modeling_outputs.Transformer2DModelOutput",headingTag:"h2"}}),D=new ne({props:{name:"class diffusers.models.modeling_outputs.Transformer2DModelOutput",anchor:"diffusers.models.modeling_outputs.Transformer2DModelOutput",parameters:[{name:"sample",val:": torch.Tensor"}],parametersDescription:[{anchor:"diffusers.models.modeling_outputs.Transformer2DModelOutput.sample",description:`<strong>sample</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, num_channels, height, width)</code> or <code>(batch size, num_vector_embeds - 1, num_latent_pixels)</code> if <a href="/docs/diffusers/pr_13331/en/api/models/transformer2d#diffusers.Transformer2DModel">Transformer2DModel</a> is discrete) &#x2014;
The hidden states output conditioned on the <code>encoder_hidden_states</code> input. If discrete, returns probability
distributions for the unnoised latent pixels.`,name:"sample"}],source:"https://github.com/huggingface/diffusers/blob/vr_13331/src/diffusers/models/modeling_outputs.py#L21"}}),H=new ue({props:{source:"https://github.com/huggingface/diffusers/blob/main/docs/source/en/api/models/helios_transformer3d.md"}}),{c(){a=d("meta"),k=n(),Z=d("p"),U=n(),f(g.$$.fragment),z=n(),f(b.$$.fragment),N=n(),M=d("p"),M.innerHTML=ee,R=n(),T=d("p"),T.textContent=oe,I=n(),f(y.$$.fragment),V=n(),f(v.$$.fragment),C=n(),i=d("div"),f($.$$.fragment),S=n(),w=d("p"),w.textContent=te,B=n(),f(x.$$.fragment),E=n(),l=d("div"),f(D.$$.fragment),O=n(),J=d("p"),J.innerHTML=se,X=n(),f(H.$$.fragment),q=n(),j=d("p"),this.h()},l(e){const o=me("svelte-u9bgzb",document.head);a=m(o,"META",{name:!0,content:!0}),o.forEach(t),k=r(e),Z=m(e,"P",{}),F(Z).forEach(t),U=r(e),c(g.$$.fragment,e),z=r(e),c(b.$$.fragment,e),N=r(e),M=m(e,"P",{"data-svelte-h":!0}),L(M)!=="svelte-148zs2y"&&(M.innerHTML=ee),R=r(e),T=m(e,"P",{"data-svelte-h":!0}),L(T)!=="svelte-1vuni30"&&(T.textContent=oe),I=r(e),c(y.$$.fragment,e),V=r(e),c(v.$$.fragment,e),C=r(e),i=m(e,"DIV",{class:!0});var Y=F(i);c($.$$.fragment,Y),S=r(Y),w=m(Y,"P",{"data-svelte-h":!0}),L(w)!=="svelte-1um4p8x"&&(w.textContent=te),Y.forEach(t),B=r(e),c(x.$$.fragment,e),E=r(e),l=m(e,"DIV",{class:!0});var Q=F(l);c(D.$$.fragment,Q),O=r(Q),J=m(Q,"P",{"data-svelte-h":!0}),L(J)!=="svelte-1460eox"&&(J.innerHTML=se),Q.forEach(t),X=r(e),c(H.$$.fragment,e),q=r(e),j=m(e,"P",{}),F(j).forEach(t),this.h()},h(){P(a,"name","hf:doc:metadata"),P(a,"content",he),P(i,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),P(l,"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,o){W(document.head,a),s(e,k,o),s(e,Z,o),s(e,U,o),u(g,e,o),s(e,z,o),u(b,e,o),s(e,N,o),s(e,M,o),s(e,R,o),s(e,T,o),s(e,I,o),u(y,e,o),s(e,V,o),u(v,e,o),s(e,C,o),s(e,i,o),u($,i,null),W(i,S),W(i,w),s(e,B,o),u(x,e,o),s(e,E,o),s(e,l,o),u(D,l,null),W(l,O),W(l,J),s(e,X,o),u(H,e,o),s(e,q,o),s(e,j,o),G=!0},p:ae,i(e){G||(p(g.$$.fragment,e),p(b.$$.fragment,e),p(y.$$.fragment,e),p(v.$$.fragment,e),p($.$$.fragment,e),p(x.$$.fragment,e),p(D.$$.fragment,e),p(H.$$.fragment,e),G=!0)},o(e){h(g.$$.fragment,e),h(b.$$.fragment,e),h(y.$$.fragment,e),h(v.$$.fragment,e),h($.$$.fragment,e),h(x.$$.fragment,e),h(D.$$.fragment,e),h(H.$$.fragment,e),G=!1},d(e){e&&(t(k),t(Z),t(U),t(z),t(N),t(M),t(R),t(T),t(I),t(V),t(C),t(i),t(B),t(E),t(l),t(X),t(q),t(j)),t(a),_(g,e),_(b,e),_(y,e),_(v,e),_($),_(x,e),_(D),_(H,e)}}}const he='{"title":"HeliosTransformer3DModel","local":"heliostransformer3dmodel","sections":[{"title":"HeliosTransformer3DModel","local":"diffusers.HeliosTransformer3DModel","sections":[],"depth":2},{"title":"Transformer2DModelOutput","local":"diffusers.models.modeling_outputs.Transformer2DModelOutput","sections":[],"depth":2}],"depth":1}';function _e(K){return ie(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class $e extends le{constructor(a){super(),de(this,a,_e,pe,re,{})}}export{$e as component};