diffusers/pr_13921/en/_app/immutable/nodes/92.321f14b2.js

import{s as ce,n as ue,o as pe}from"../chunks/scheduler.53228c21.js";import{S as _e,i as he,e as a,s as r,c as f,h as ge,a as i,d as n,b as s,f as E,g as c,j as Z,k as J,l as u,m as t,n as p,t as _,o as h,p as g}from"../chunks/index.cac5d66a.js";import{C as Te}from"../chunks/CopyLLMTxtMenu.0ef49226.js";import{D as re}from"../chunks/Docstring.9de32ff4.js";import{C as ye}from"../chunks/CodeBlock.606cbaf4.js";import{H as se,E as be}from"../chunks/MermaidChart.svelte_svelte_type_style_lang.48d5cb47.js";function ke(de){let l,L,N,F,y,I,b,P,k,ae='A Diffusion Transformer model for 3D video-like data was introduced in <a href="https://github.com/SkyworkAI/SkyReels-V2" rel="nofollow">SkyReels-V2</a> by the Skywork AI.',U,v,ie="The model can be loaded with the following code snippet.",W,M,H,$,O,d,D,ee,R,le="A Transformer model for video-like data used in the Wan-based SkyReels-V2 model.",oe,T,x,ne,q,me='The <a href="/docs/diffusers/pr_13921/en/api/models/skyreels_v2_transformer_3d#diffusers.SkyReelsV2Transformer3DModel">SkyReelsV2Transformer3DModel</a> forward method.',A,w,G,m,V,te,z,fe='The output of <a href="/docs/diffusers/pr_13921/en/api/models/transformer2d#diffusers.Transformer2DModel">Transformer2DModel</a>.',B,S,X,C,K;return y=new Te({props:{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"}}),b=new se({props:{title:"SkyReelsV2Transformer3DModel",local:"skyreelsv2transformer3dmodel",headingTag:"h1"}}),M=new ye({props:{code:"ZnJvbSUyMGRpZmZ1c2VycyUyMGltcG9ydCUyMFNreVJlZWxzVjJUcmFuc2Zvcm1lcjNETW9kZWwlMEElMEF0cmFuc2Zvcm1lciUyMCUzRCUyMFNreVJlZWxzVjJUcmFuc2Zvcm1lcjNETW9kZWwuZnJvbV9wcmV0cmFpbmVkKCUyMlNreXdvcmslMkZTa3lSZWVscy1WMi1ERi0xLjNCLTU0MFAtRGlmZnVzZXJzJTIyJTJDJTIwc3ViZm9sZGVyJTNEJTIydHJhbnNmb3JtZXIlMjIlMkMlMjB0b3JjaF9kdHlwZSUzRHRvcmNoLmJmbG9hdDE2KQ==",highlighted:`<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> SkyReelsV2Transformer3DModel

transformer = SkyReelsV2Transformer3DModel.from_pretrained(<span class="hljs-string">&quot;Skywork/SkyReels-V2-DF-1.3B-540P-Diffusers&quot;</span>, subfolder=<span class="hljs-string">&quot;transformer&quot;</span>, torch_dtype=torch.bfloat16)`,lang:"python",wrap:!1}}),$=new se({props:{title:"SkyReelsV2Transformer3DModel",local:"diffusers.SkyReelsV2Transformer3DModel",headingTag:"h2"}}),D=new re({props:{name:"class diffusers.SkyReelsV2Transformer3DModel",anchor:"diffusers.SkyReelsV2Transformer3DModel",parameters:[{name:"patch_size",val:": tuple = (1, 2, 2)"},{name:"num_attention_heads",val:": int = 16"},{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 = 8192"},{name:"num_layers",val:": int = 32"},{name:"cross_attn_norm",val:": bool = True"},{name:"qk_norm",val:": str | None = 'rms_norm_across_heads'"},{name:"eps",val:": float = 1e-06"},{name:"image_dim",val:": int | None = None"},{name:"added_kv_proj_dim",val:": int | None = None"},{name:"rope_max_seq_len",val:": int = 1024"},{name:"pos_embed_seq_len",val:": int | None = None"},{name:"inject_sample_info",val:": bool = False"},{name:"num_frame_per_block",val:": int = 1"}],parametersDescription:[{anchor:"diffusers.SkyReelsV2Transformer3DModel.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.SkyReelsV2Transformer3DModel.num_attention_heads",description:`<strong>num_attention_heads</strong> (<code>int</code>, defaults to <code>16</code>) &#x2014;
Fixed length for text embeddings.`,name:"num_attention_heads"},{anchor:"diffusers.SkyReelsV2Transformer3DModel.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.SkyReelsV2Transformer3DModel.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.SkyReelsV2Transformer3DModel.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.SkyReelsV2Transformer3DModel.text_dim",description:`<strong>text_dim</strong> (<code>int</code>, defaults to <code>4096</code>) &#x2014;
Input dimension for text embeddings.`,name:"text_dim"},{anchor:"diffusers.SkyReelsV2Transformer3DModel.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.SkyReelsV2Transformer3DModel.ffn_dim",description:`<strong>ffn_dim</strong> (<code>int</code>, defaults to <code>8192</code>) &#x2014;
Intermediate dimension in feed-forward network.`,name:"ffn_dim"},{anchor:"diffusers.SkyReelsV2Transformer3DModel.num_layers",description:`<strong>num_layers</strong> (<code>int</code>, defaults to <code>32</code>) &#x2014;
The number of layers of transformer blocks to use.`,name:"num_layers"},{anchor:"diffusers.SkyReelsV2Transformer3DModel.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.SkyReelsV2Transformer3DModel.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.SkyReelsV2Transformer3DModel.qk_norm",description:`<strong>qk_norm</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;rms_norm_across_heads&quot;</code>) &#x2014;
Enable query/key normalization.`,name:"qk_norm"},{anchor:"diffusers.SkyReelsV2Transformer3DModel.eps",description:`<strong>eps</strong> (<code>float</code>, defaults to <code>1e-6</code>) &#x2014;
Epsilon value for normalization layers.`,name:"eps"},{anchor:"diffusers.SkyReelsV2Transformer3DModel.inject_sample_info",description:`<strong>inject_sample_info</strong> (<code>bool</code>, defaults to <code>False</code>) &#x2014;
Whether to inject sample information into the model.`,name:"inject_sample_info"},{anchor:"diffusers.SkyReelsV2Transformer3DModel.image_dim",description:`<strong>image_dim</strong> (<code>int</code>, <em>optional</em>) &#x2014;
The dimension of the image embeddings.`,name:"image_dim"},{anchor:"diffusers.SkyReelsV2Transformer3DModel.added_kv_proj_dim",description:`<strong>added_kv_proj_dim</strong> (<code>int</code>, <em>optional</em>) &#x2014;
The dimension of the added key/value projection.`,name:"added_kv_proj_dim"},{anchor:"diffusers.SkyReelsV2Transformer3DModel.rope_max_seq_len",description:`<strong>rope_max_seq_len</strong> (<code>int</code>, defaults to <code>1024</code>) &#x2014;
The maximum sequence length for the rotary embeddings.`,name:"rope_max_seq_len"},{anchor:"diffusers.SkyReelsV2Transformer3DModel.pos_embed_seq_len",description:`<strong>pos_embed_seq_len</strong> (<code>int</code>, <em>optional</em>) &#x2014;
The sequence length for the positional embeddings.`,name:"pos_embed_seq_len"}],source:"https://github.com/huggingface/diffusers/blob/vr_13921/src/diffusers/models/transformers/transformer_skyreels_v2.py#L518"}}),x=new re({props:{name:"forward",anchor:"diffusers.SkyReelsV2Transformer3DModel.forward",parameters:[{name:"hidden_states",val:": Tensor"},{name:"timestep",val:": LongTensor"},{name:"encoder_hidden_states",val:": Tensor"},{name:"encoder_hidden_states_image",val:": torch.Tensor | None = None"},{name:"enable_diffusion_forcing",val:": bool = False"},{name:"fps",val:": torch.Tensor | None = None"},{name:"return_dict",val:": bool = True"},{name:"attention_kwargs",val:": dict[str, typing.Any] | None = None"}],parametersDescription:[{anchor:"diffusers.SkyReelsV2Transformer3DModel.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.SkyReelsV2Transformer3DModel.forward.timestep",description:`<strong>timestep</strong> (<code>torch.LongTensor</code>) &#x2014;
Used to indicate denoising step.`,name:"timestep"},{anchor:"diffusers.SkyReelsV2Transformer3DModel.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.SkyReelsV2Transformer3DModel.forward.encoder_hidden_states_image",description:`<strong>encoder_hidden_states_image</strong> (<code>torch.Tensor</code>, <em>optional</em>) &#x2014;
Conditional image embeddings for image-conditioned generation.`,name:"encoder_hidden_states_image"},{anchor:"diffusers.SkyReelsV2Transformer3DModel.forward.enable_diffusion_forcing",description:`<strong>enable_diffusion_forcing</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) &#x2014;
Whether to enable diffusion forcing (per-block causal masking).`,name:"enable_diffusion_forcing"},{anchor:"diffusers.SkyReelsV2Transformer3DModel.forward.fps",description:`<strong>fps</strong> (<code>torch.Tensor</code>, <em>optional</em>) &#x2014;
FPS conditioning embedding.`,name:"fps"},{anchor:"diffusers.SkyReelsV2Transformer3DModel.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"},{anchor:"diffusers.SkyReelsV2Transformer3DModel.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"}],source:"https://github.com/huggingface/diffusers/blob/vr_13921/src/diffusers/models/transformers/transformer_skyreels_v2.py#L633",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>If <code>return_dict</code> is True, an <code>~models.transformer_2d.Transformer2DModelOutput</code> is returned, otherwise a
<code>tuple</code> where the first element is the sample tensor.</p>
`}}),w=new se({props:{title:"Transformer2DModelOutput",local:"diffusers.models.modeling_outputs.Transformer2DModelOutput",headingTag:"h2"}}),V=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"}}),S=new be({props:{source:"https://github.com/huggingface/diffusers/blob/main/docs/source/en/api/models/skyreels_v2_transformer_3d.md"}}),{c(){l=a("meta"),L=r(),N=a("p"),F=r(),f(y.$$.fragment),I=r(),f(b.$$.fragment),P=r(),k=a("p"),k.innerHTML=ae,U=r(),v=a("p"),v.textContent=ie,W=r(),f(M.$$.fragment),H=r(),f($.$$.fragment),O=r(),d=a("div"),f(D.$$.fragment),ee=r(),R=a("p"),R.textContent=le,oe=r(),T=a("div"),f(x.$$.fragment),ne=r(),q=a("p"),q.innerHTML=me,A=r(),f(w.$$.fragment),G=r(),m=a("div"),f(V.$$.fragment),te=r(),z=a("p"),z.innerHTML=fe,B=r(),f(S.$$.fragment),X=r(),C=a("p"),this.h()},l(e){const o=ge("svelte-u9bgzb",document.head);l=i(o,"META",{name:!0,content:!0}),o.forEach(n),L=s(e),N=i(e,"P",{}),E(N).forEach(n),F=s(e),c(y.$$.fragment,e),I=s(e),c(b.$$.fragment,e),P=s(e),k=i(e,"P",{"data-svelte-h":!0}),Z(k)!=="svelte-14k6y22"&&(k.innerHTML=ae),U=s(e),v=i(e,"P",{"data-svelte-h":!0}),Z(v)!=="svelte-1vuni30"&&(v.textContent=ie),W=s(e),c(M.$$.fragment,e),H=s(e),c($.$$.fragment,e),O=s(e),d=i(e,"DIV",{class:!0});var j=E(d);c(D.$$.fragment,j),ee=s(j),R=i(j,"P",{"data-svelte-h":!0}),Z(R)!=="svelte-om4ddc"&&(R.textContent=le),oe=s(j),T=i(j,"DIV",{class:!0});var Q=E(T);c(x.$$.fragment,Q),ne=s(Q),q=i(Q,"P",{"data-svelte-h":!0}),Z(q)!=="svelte-1182zuo"&&(q.innerHTML=me),Q.forEach(n),j.forEach(n),A=s(e),c(w.$$.fragment,e),G=s(e),m=i(e,"DIV",{class:!0});var Y=E(m);c(V.$$.fragment,Y),te=s(Y),z=i(Y,"P",{"data-svelte-h":!0}),Z(z)!=="svelte-2clpd6"&&(z.innerHTML=fe),Y.forEach(n),B=s(e),c(S.$$.fragment,e),X=s(e),C=i(e,"P",{}),E(C).forEach(n),this.h()},h(){J(l,"name","hf:doc:metadata"),J(l,"content",ve),J(T,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),J(d,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),J(m,"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){u(document.head,l),t(e,L,o),t(e,N,o),t(e,F,o),p(y,e,o),t(e,I,o),p(b,e,o),t(e,P,o),t(e,k,o),t(e,U,o),t(e,v,o),t(e,W,o),p(M,e,o),t(e,H,o),p($,e,o),t(e,O,o),t(e,d,o),p(D,d,null),u(d,ee),u(d,R),u(d,oe),u(d,T),p(x,T,null),u(T,ne),u(T,q),t(e,A,o),p(w,e,o),t(e,G,o),t(e,m,o),p(V,m,null),u(m,te),u(m,z),t(e,B,o),p(S,e,o),t(e,X,o),t(e,C,o),K=!0},p:ue,i(e){K||(_(y.$$.fragment,e),_(b.$$.fragment,e),_(M.$$.fragment,e),_($.$$.fragment,e),_(D.$$.fragment,e),_(x.$$.fragment,e),_(w.$$.fragment,e),_(V.$$.fragment,e),_(S.$$.fragment,e),K=!0)},o(e){h(y.$$.fragment,e),h(b.$$.fragment,e),h(M.$$.fragment,e),h($.$$.fragment,e),h(D.$$.fragment,e),h(x.$$.fragment,e),h(w.$$.fragment,e),h(V.$$.fragment,e),h(S.$$.fragment,e),K=!1},d(e){e&&(n(L),n(N),n(F),n(I),n(P),n(k),n(U),n(v),n(W),n(H),n(O),n(d),n(A),n(G),n(m),n(B),n(X),n(C)),n(l),g(y,e),g(b,e),g(M,e),g($,e),g(D),g(x),g(w,e),g(V),g(S,e)}}}const ve='{"title":"SkyReelsV2Transformer3DModel","local":"skyreelsv2transformer3dmodel","sections":[{"title":"SkyReelsV2Transformer3DModel","local":"diffusers.SkyReelsV2Transformer3DModel","sections":[],"depth":2},{"title":"Transformer2DModelOutput","local":"diffusers.models.modeling_outputs.Transformer2DModelOutput","sections":[],"depth":2}],"depth":1}';function Me(de){return pe(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class Re extends _e{constructor(l){super(),he(this,l,Me,ke,ce,{})}}export{Re as component};