diffusers/pr_13921/en/_app/immutable/nodes/81.310fb9cc.js

import{s as fe,n as ue,o as he}from"../chunks/scheduler.53228c21.js";import{S as pe,i as _e,e as i,s as r,c as l,h as ge,a as d,d as o,b as s,f as P,g as f,j as E,k as I,l as u,m as n,n as h,t as p,o as _,p as g}from"../chunks/index.cac5d66a.js";import{C as Me}from"../chunks/CopyLLMTxtMenu.0ef49226.js";import{D as re}from"../chunks/Docstring.9de32ff4.js";import{C as Te}from"../chunks/CodeBlock.606cbaf4.js";import{H as se,E as be}from"../chunks/MermaidChart.svelte_svelte_type_style_lang.48d5cb47.js";function $e(ae){let m,J,j,X,T,B,b,O,$,ie='A Diffusion Transformer model for 3D video-like data was introduced in <a href="https://huggingface.co/genmo/mochi-1-preview" rel="nofollow">Mochi-1 Preview</a> by Genmo.',V,v,de="The model can be loaded with the following code snippet.",R,x,U,y,A,a,D,ee,Z,me='A Transformer model for video-like data introduced in <a href="https://huggingface.co/genmo/mochi-1-preview" rel="nofollow">Mochi</a>.',te,M,w,oe,z,ce='The <a href="/docs/diffusers/pr_13921/en/api/models/mochi_transformer3d#diffusers.MochiTransformer3DModel">MochiTransformer3DModel</a> forward method.',G,k,S,c,q,ne,C,le='The output of <a href="/docs/diffusers/pr_13921/en/api/models/transformer2d#diffusers.Transformer2DModel">Transformer2DModel</a>.',W,L,Y,H,Q;return T=new Me({props:{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"}}),b=new se({props:{title:"MochiTransformer3DModel",local:"mochitransformer3dmodel",headingTag:"h1"}}),x=new Te({props:{code:"ZnJvbSUyMGRpZmZ1c2VycyUyMGltcG9ydCUyME1vY2hpVHJhbnNmb3JtZXIzRE1vZGVsJTBBJTBBdHJhbnNmb3JtZXIlMjAlM0QlMjBNb2NoaVRyYW5zZm9ybWVyM0RNb2RlbC5mcm9tX3ByZXRyYWluZWQoJTIyZ2VubW8lMkZtb2NoaS0xLXByZXZpZXclMjIlMkMlMjBzdWJmb2xkZXIlM0QlMjJ0cmFuc2Zvcm1lciUyMiUyQyUyMHRvcmNoX2R0eXBlJTNEdG9yY2guZmxvYXQxNikudG8oJTIyY3VkYSUyMik=",highlighted:`<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> MochiTransformer3DModel

transformer = MochiTransformer3DModel.from_pretrained(<span class="hljs-string">&quot;genmo/mochi-1-preview&quot;</span>, subfolder=<span class="hljs-string">&quot;transformer&quot;</span>, torch_dtype=torch.float16).to(<span class="hljs-string">&quot;cuda&quot;</span>)`,lang:"python",wrap:!1}}),y=new se({props:{title:"MochiTransformer3DModel",local:"diffusers.MochiTransformer3DModel",headingTag:"h2"}}),D=new re({props:{name:"class diffusers.MochiTransformer3DModel",anchor:"diffusers.MochiTransformer3DModel",parameters:[{name:"patch_size",val:": int = 2"},{name:"num_attention_heads",val:": int = 24"},{name:"attention_head_dim",val:": int = 128"},{name:"num_layers",val:": int = 48"},{name:"pooled_projection_dim",val:": int = 1536"},{name:"in_channels",val:": int = 12"},{name:"out_channels",val:": int | None = None"},{name:"qk_norm",val:": str = 'rms_norm'"},{name:"text_embed_dim",val:": int = 4096"},{name:"time_embed_dim",val:": int = 256"},{name:"activation_fn",val:": str = 'swiglu'"},{name:"max_sequence_length",val:": int = 256"}],parametersDescription:[{anchor:"diffusers.MochiTransformer3DModel.patch_size",description:`<strong>patch_size</strong> (<code>int</code>, defaults to <code>2</code>) &#x2014;
The size of the patches to use in the patch embedding layer.`,name:"patch_size"},{anchor:"diffusers.MochiTransformer3DModel.num_attention_heads",description:`<strong>num_attention_heads</strong> (<code>int</code>, defaults to <code>24</code>) &#x2014;
The number of heads to use for multi-head attention.`,name:"num_attention_heads"},{anchor:"diffusers.MochiTransformer3DModel.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.MochiTransformer3DModel.num_layers",description:`<strong>num_layers</strong> (<code>int</code>, defaults to <code>48</code>) &#x2014;
The number of layers of Transformer blocks to use.`,name:"num_layers"},{anchor:"diffusers.MochiTransformer3DModel.in_channels",description:`<strong>in_channels</strong> (<code>int</code>, defaults to <code>12</code>) &#x2014;
The number of channels in the input.`,name:"in_channels"},{anchor:"diffusers.MochiTransformer3DModel.out_channels",description:`<strong>out_channels</strong> (<code>int</code>, <em>optional</em>, defaults to <code>None</code>) &#x2014;
The number of channels in the output.`,name:"out_channels"},{anchor:"diffusers.MochiTransformer3DModel.qk_norm",description:`<strong>qk_norm</strong> (<code>str</code>, defaults to <code>&quot;rms_norm&quot;</code>) &#x2014;
The normalization layer to use.`,name:"qk_norm"},{anchor:"diffusers.MochiTransformer3DModel.text_embed_dim",description:`<strong>text_embed_dim</strong> (<code>int</code>, defaults to <code>4096</code>) &#x2014;
Input dimension of text embeddings from the text encoder.`,name:"text_embed_dim"},{anchor:"diffusers.MochiTransformer3DModel.time_embed_dim",description:`<strong>time_embed_dim</strong> (<code>int</code>, defaults to <code>256</code>) &#x2014;
Output dimension of timestep embeddings.`,name:"time_embed_dim"},{anchor:"diffusers.MochiTransformer3DModel.activation_fn",description:`<strong>activation_fn</strong> (<code>str</code>, defaults to <code>&quot;swiglu&quot;</code>) &#x2014;
Activation function to use in feed-forward.`,name:"activation_fn"},{anchor:"diffusers.MochiTransformer3DModel.max_sequence_length",description:`<strong>max_sequence_length</strong> (<code>int</code>, defaults to <code>256</code>) &#x2014;
The maximum sequence length of text embeddings supported.`,name:"max_sequence_length"}],source:"https://github.com/huggingface/diffusers/blob/vr_13921/src/diffusers/models/transformers/transformer_mochi.py#L309"}}),w=new re({props:{name:"forward",anchor:"diffusers.MochiTransformer3DModel.forward",parameters:[{name:"hidden_states",val:": Tensor"},{name:"encoder_hidden_states",val:": Tensor"},{name:"timestep",val:": LongTensor"},{name:"encoder_attention_mask",val:": Tensor"},{name:"attention_kwargs",val:": dict[str, typing.Any] | None = None"},{name:"return_dict",val:": bool = True"}],parametersDescription:[{anchor:"diffusers.MochiTransformer3DModel.forward.hidden_states",description:`<strong>hidden_states</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, num_channels, num_frames, height, width)</code>) &#x2014;
Input <code>hidden_states</code>.`,name:"hidden_states"},{anchor:"diffusers.MochiTransformer3DModel.forward.encoder_hidden_states",description:`<strong>encoder_hidden_states</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, sequence_len, embed_dims)</code>) &#x2014;
Conditional embeddings (embeddings computed from the input conditions such as prompts) to use.`,name:"encoder_hidden_states"},{anchor:"diffusers.MochiTransformer3DModel.forward.timestep",description:`<strong>timestep</strong> (<code>torch.LongTensor</code>) &#x2014;
Used to indicate denoising step.`,name:"timestep"},{anchor:"diffusers.MochiTransformer3DModel.forward.encoder_attention_mask",description:`<strong>encoder_attention_mask</strong> (<code>torch.Tensor</code>) &#x2014;
Mask applied to <code>encoder_hidden_states</code> during attention.`,name:"encoder_attention_mask"},{anchor:"diffusers.MochiTransformer3DModel.forward.attention_kwargs",description:`<strong>attention_kwargs</strong> (<code>dict</code>, <em>optional</em>) &#x2014;
A kwargs dictionary that if specified is passed along to the <code>AttentionProcessor</code> as defined under
<code>self.processor</code> in
<a href="https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py" rel="nofollow">diffusers.models.attention_processor</a>.`,name:"attention_kwargs"},{anchor:"diffusers.MochiTransformer3DModel.forward.return_dict",description:`<strong>return_dict</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) &#x2014;
Whether or not to return a <code>~models.transformer_2d.Transformer2DModelOutput</code> instead of a plain
tuple.`,name:"return_dict"}],source:"https://github.com/huggingface/diffusers/blob/vr_13921/src/diffusers/models/transformers/transformer_mochi.py#L407",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>The denoised output tensor of shape <code>(batch_size, out_channels, num_frames, height, width)</code>.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><code>torch.Tensor</code></p>
`}}),k=new se({props:{title:"Transformer2DModelOutput",local:"diffusers.models.modeling_outputs.Transformer2DModelOutput",headingTag:"h2"}}),q=new re({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_13921/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_13921/src/diffusers/models/modeling_outputs.py#L21"}}),L=new be({props:{source:"https://github.com/huggingface/diffusers/blob/main/docs/source/en/api/models/mochi_transformer3d.md"}}),{c(){m=i("meta"),J=r(),j=i("p"),X=r(),l(T.$$.fragment),B=r(),l(b.$$.fragment),O=r(),$=i("p"),$.innerHTML=ie,V=r(),v=i("p"),v.textContent=de,R=r(),l(x.$$.fragment),U=r(),l(y.$$.fragment),A=r(),a=i("div"),l(D.$$.fragment),ee=r(),Z=i("p"),Z.innerHTML=me,te=r(),M=i("div"),l(w.$$.fragment),oe=r(),z=i("p"),z.innerHTML=ce,G=r(),l(k.$$.fragment),S=r(),c=i("div"),l(q.$$.fragment),ne=r(),C=i("p"),C.innerHTML=le,W=r(),l(L.$$.fragment),Y=r(),H=i("p"),this.h()},l(e){const t=ge("svelte-u9bgzb",document.head);m=d(t,"META",{name:!0,content:!0}),t.forEach(o),J=s(e),j=d(e,"P",{}),P(j).forEach(o),X=s(e),f(T.$$.fragment,e),B=s(e),f(b.$$.fragment,e),O=s(e),$=d(e,"P",{"data-svelte-h":!0}),E($)!=="svelte-1up2xls"&&($.innerHTML=ie),V=s(e),v=d(e,"P",{"data-svelte-h":!0}),E(v)!=="svelte-1vuni30"&&(v.textContent=de),R=s(e),f(x.$$.fragment,e),U=s(e),f(y.$$.fragment,e),A=s(e),a=d(e,"DIV",{class:!0});var N=P(a);f(D.$$.fragment,N),ee=s(N),Z=d(N,"P",{"data-svelte-h":!0}),E(Z)!=="svelte-133wd4y"&&(Z.innerHTML=me),te=s(N),M=d(N,"DIV",{class:!0});var F=P(M);f(w.$$.fragment,F),oe=s(F),z=d(F,"P",{"data-svelte-h":!0}),E(z)!=="svelte-1my0jn6"&&(z.innerHTML=ce),F.forEach(o),N.forEach(o),G=s(e),f(k.$$.fragment,e),S=s(e),c=d(e,"DIV",{class:!0});var K=P(c);f(q.$$.fragment,K),ne=s(K),C=d(K,"P",{"data-svelte-h":!0}),E(C)!=="svelte-2clpd6"&&(C.innerHTML=le),K.forEach(o),W=s(e),f(L.$$.fragment,e),Y=s(e),H=d(e,"P",{}),P(H).forEach(o),this.h()},h(){I(m,"name","hf:doc:metadata"),I(m,"content",ve),I(M,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),I(a,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),I(c,"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,t){u(document.head,m),n(e,J,t),n(e,j,t),n(e,X,t),h(T,e,t),n(e,B,t),h(b,e,t),n(e,O,t),n(e,$,t),n(e,V,t),n(e,v,t),n(e,R,t),h(x,e,t),n(e,U,t),h(y,e,t),n(e,A,t),n(e,a,t),h(D,a,null),u(a,ee),u(a,Z),u(a,te),u(a,M),h(w,M,null),u(M,oe),u(M,z),n(e,G,t),h(k,e,t),n(e,S,t),n(e,c,t),h(q,c,null),u(c,ne),u(c,C),n(e,W,t),h(L,e,t),n(e,Y,t),n(e,H,t),Q=!0},p:ue,i(e){Q||(p(T.$$.fragment,e),p(b.$$.fragment,e),p(x.$$.fragment,e),p(y.$$.fragment,e),p(D.$$.fragment,e),p(w.$$.fragment,e),p(k.$$.fragment,e),p(q.$$.fragment,e),p(L.$$.fragment,e),Q=!0)},o(e){_(T.$$.fragment,e),_(b.$$.fragment,e),_(x.$$.fragment,e),_(y.$$.fragment,e),_(D.$$.fragment,e),_(w.$$.fragment,e),_(k.$$.fragment,e),_(q.$$.fragment,e),_(L.$$.fragment,e),Q=!1},d(e){e&&(o(J),o(j),o(X),o(B),o(O),o($),o(V),o(v),o(R),o(U),o(A),o(a),o(G),o(S),o(c),o(W),o(Y),o(H)),o(m),g(T,e),g(b,e),g(x,e),g(y,e),g(D),g(w),g(k,e),g(q),g(L,e)}}}const ve='{"title":"MochiTransformer3DModel","local":"mochitransformer3dmodel","sections":[{"title":"MochiTransformer3DModel","local":"diffusers.MochiTransformer3DModel","sections":[],"depth":2},{"title":"Transformer2DModelOutput","local":"diffusers.models.modeling_outputs.Transformer2DModelOutput","sections":[],"depth":2}],"depth":1}';function xe(ae){return he(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class Ze extends pe{constructor(m){super(),_e(this,m,xe,$e,fe,{})}}export{Ze as component};