diffusers/pr_12595/en/_app/immutable/nodes/56.d60ec41f.js

import{s as ue,n as me,o as _e}from"../chunks/scheduler.53228c21.js";import{S as pe,i as he,e as i,s as r,c as l,h as Ae,a,d as s,b as n,f as I,g as u,j as J,k,l as m,m as t,n as _,t as p,o as h,p as A}from"../chunks/index.100fac89.js";import{C as ge}from"../chunks/CopyLLMTxtMenu.7aefc1a4.js";import{D as re}from"../chunks/Docstring.d6cb35e8.js";import{C as Pe}from"../chunks/CodeBlock.d30a6509.js";import{H as ne,E as ye}from"../chunks/MermaidChart.svelte_svelte_type_style_lang.3722da43.js";function Te(de){let c,X,L,z,P,R,y,Z,T,ie='A Diffusion Transformer model for 3D video-like data was introduced in <a href="https://huggingface.co/papers/2412.03603" rel="nofollow">HunyuanVideo: A Systematic Framework For Large Video Generative Models</a> by Tencent.',j,b,ae="The model can be loaded with the following code snippet.",E,$,N,M,U,d,V,ee,F,ce='A Transformer model for video-like data used in <a href="https://huggingface.co/tencent/HunyuanVideo" rel="nofollow">HunyuanVideo</a>.',oe,g,v,se,w,fe="Sets the attention processor to use to compute attention.",W,x,K,f,D,te,C,le='The output of <a href="/docs/diffusers/pr_12595/en/api/models/transformer2d#diffusers.Transformer2DModel">Transformer2DModel</a>.',q,H,O,S,Y;return P=new ge({props:{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"}}),y=new ne({props:{title:"HunyuanVideoTransformer3DModel",local:"hunyuanvideotransformer3dmodel",headingTag:"h1"}}),$=new Pe({props:{code:"ZnJvbSUyMGRpZmZ1c2VycyUyMGltcG9ydCUyMEh1bnl1YW5WaWRlb1RyYW5zZm9ybWVyM0RNb2RlbCUwQSUwQXRyYW5zZm9ybWVyJTIwJTNEJTIwSHVueXVhblZpZGVvVHJhbnNmb3JtZXIzRE1vZGVsLmZyb21fcHJldHJhaW5lZCglMjJodW55dWFudmlkZW8tY29tbXVuaXR5JTJGSHVueXVhblZpZGVvJTIyJTJDJTIwc3ViZm9sZGVyJTNEJTIydHJhbnNmb3JtZXIlMjIlMkMlMjB0b3JjaF9kdHlwZSUzRHRvcmNoLmJmbG9hdDE2KQ==",highlighted:`<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> HunyuanVideoTransformer3DModel

transformer = HunyuanVideoTransformer3DModel.from_pretrained(<span class="hljs-string">&quot;hunyuanvideo-community/HunyuanVideo&quot;</span>, subfolder=<span class="hljs-string">&quot;transformer&quot;</span>, torch_dtype=torch.bfloat16)`,wrap:!1}}),M=new ne({props:{title:"HunyuanVideoTransformer3DModel",local:"diffusers.HunyuanVideoTransformer3DModel",headingTag:"h2"}}),V=new re({props:{name:"class diffusers.HunyuanVideoTransformer3DModel",anchor:"diffusers.HunyuanVideoTransformer3DModel",parameters:[{name:"in_channels",val:": int = 16"},{name:"out_channels",val:": int = 16"},{name:"num_attention_heads",val:": int = 24"},{name:"attention_head_dim",val:": int = 128"},{name:"num_layers",val:": int = 20"},{name:"num_single_layers",val:": int = 40"},{name:"num_refiner_layers",val:": int = 2"},{name:"mlp_ratio",val:": float = 4.0"},{name:"patch_size",val:": int = 2"},{name:"patch_size_t",val:": int = 1"},{name:"qk_norm",val:": str = 'rms_norm'"},{name:"guidance_embeds",val:": bool = True"},{name:"text_embed_dim",val:": int = 4096"},{name:"pooled_projection_dim",val:": int = 768"},{name:"rope_theta",val:": float = 256.0"},{name:"rope_axes_dim",val:": typing.Tuple[int] = (16, 56, 56)"},{name:"image_condition_type",val:": typing.Optional[str] = None"}],parametersDescription:[{anchor:"diffusers.HunyuanVideoTransformer3DModel.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.HunyuanVideoTransformer3DModel.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.HunyuanVideoTransformer3DModel.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.HunyuanVideoTransformer3DModel.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.HunyuanVideoTransformer3DModel.num_layers",description:`<strong>num_layers</strong> (<code>int</code>, defaults to <code>20</code>) &#x2014;
The number of layers of dual-stream blocks to use.`,name:"num_layers"},{anchor:"diffusers.HunyuanVideoTransformer3DModel.num_single_layers",description:`<strong>num_single_layers</strong> (<code>int</code>, defaults to <code>40</code>) &#x2014;
The number of layers of single-stream blocks to use.`,name:"num_single_layers"},{anchor:"diffusers.HunyuanVideoTransformer3DModel.num_refiner_layers",description:`<strong>num_refiner_layers</strong> (<code>int</code>, defaults to <code>2</code>) &#x2014;
The number of layers of refiner blocks to use.`,name:"num_refiner_layers"},{anchor:"diffusers.HunyuanVideoTransformer3DModel.mlp_ratio",description:`<strong>mlp_ratio</strong> (<code>float</code>, defaults to <code>4.0</code>) &#x2014;
The ratio of the hidden layer size to the input size in the feedforward network.`,name:"mlp_ratio"},{anchor:"diffusers.HunyuanVideoTransformer3DModel.patch_size",description:`<strong>patch_size</strong> (<code>int</code>, defaults to <code>2</code>) &#x2014;
The size of the spatial patches to use in the patch embedding layer.`,name:"patch_size"},{anchor:"diffusers.HunyuanVideoTransformer3DModel.patch_size_t",description:`<strong>patch_size_t</strong> (<code>int</code>, defaults to <code>1</code>) &#x2014;
The size of the tmeporal patches to use in the patch embedding layer.`,name:"patch_size_t"},{anchor:"diffusers.HunyuanVideoTransformer3DModel.qk_norm",description:`<strong>qk_norm</strong> (<code>str</code>, defaults to <code>rms_norm</code>) &#x2014;
The normalization to use for the query and key projections in the attention layers.`,name:"qk_norm"},{anchor:"diffusers.HunyuanVideoTransformer3DModel.guidance_embeds",description:`<strong>guidance_embeds</strong> (<code>bool</code>, defaults to <code>True</code>) &#x2014;
Whether to use guidance embeddings in the model.`,name:"guidance_embeds"},{anchor:"diffusers.HunyuanVideoTransformer3DModel.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.HunyuanVideoTransformer3DModel.pooled_projection_dim",description:`<strong>pooled_projection_dim</strong> (<code>int</code>, defaults to <code>768</code>) &#x2014;
The dimension of the pooled projection of the text embeddings.`,name:"pooled_projection_dim"},{anchor:"diffusers.HunyuanVideoTransformer3DModel.rope_theta",description:`<strong>rope_theta</strong> (<code>float</code>, defaults to <code>256.0</code>) &#x2014;
The value of theta to use in the RoPE layer.`,name:"rope_theta"},{anchor:"diffusers.HunyuanVideoTransformer3DModel.rope_axes_dim",description:`<strong>rope_axes_dim</strong> (<code>Tuple[int]</code>, defaults to <code>(16, 56, 56)</code>) &#x2014;
The dimensions of the axes to use in the RoPE layer.`,name:"rope_axes_dim"},{anchor:"diffusers.HunyuanVideoTransformer3DModel.image_condition_type",description:`<strong>image_condition_type</strong> (<code>str</code>, <em>optional</em>, defaults to <code>None</code>) &#x2014;
The type of image conditioning to use. If <code>None</code>, no image conditioning is used. If <code>latent_concat</code>, the
image is concatenated to the latent stream. If <code>token_replace</code>, the image is used to replace first-frame
tokens in the latent stream and apply conditioning.`,name:"image_condition_type"}],source:"https://github.com/huggingface/diffusers/blob/vr_12595/src/diffusers/models/transformers/transformer_hunyuan_video.py#L841"}}),v=new re({props:{name:"set_attn_processor",anchor:"diffusers.HunyuanVideoTransformer3DModel.set_attn_processor",parameters:[{name:"processor",val:": typing.Union[diffusers.models.attention_processor.AttnProcessor, diffusers.models.attention_processor.CustomDiffusionAttnProcessor, diffusers.models.attention_processor.AttnAddedKVProcessor, diffusers.models.attention_processor.AttnAddedKVProcessor2_0, diffusers.models.attention_processor.JointAttnProcessor2_0, diffusers.models.attention_processor.PAGJointAttnProcessor2_0, diffusers.models.attention_processor.PAGCFGJointAttnProcessor2_0, diffusers.models.attention_processor.FusedJointAttnProcessor2_0, diffusers.models.attention_processor.AllegroAttnProcessor2_0, diffusers.models.attention_processor.AuraFlowAttnProcessor2_0, diffusers.models.attention_processor.FusedAuraFlowAttnProcessor2_0, diffusers.models.attention_processor.FluxAttnProcessor2_0, diffusers.models.attention_processor.FluxAttnProcessor2_0_NPU, diffusers.models.attention_processor.FusedFluxAttnProcessor2_0, diffusers.models.attention_processor.FusedFluxAttnProcessor2_0_NPU, diffusers.models.attention_processor.CogVideoXAttnProcessor2_0, diffusers.models.attention_processor.FusedCogVideoXAttnProcessor2_0, diffusers.models.attention_processor.XFormersAttnAddedKVProcessor, diffusers.models.attention_processor.XFormersAttnProcessor, diffusers.models.attention_processor.XLAFlashAttnProcessor2_0, diffusers.models.attention_processor.AttnProcessorNPU, diffusers.models.attention_processor.AttnProcessor2_0, diffusers.models.attention_processor.MochiVaeAttnProcessor2_0, diffusers.models.attention_processor.MochiAttnProcessor2_0, diffusers.models.attention_processor.StableAudioAttnProcessor2_0, diffusers.models.attention_processor.HunyuanAttnProcessor2_0, diffusers.models.attention_processor.FusedHunyuanAttnProcessor2_0, diffusers.models.attention_processor.PAGHunyuanAttnProcessor2_0, diffusers.models.attention_processor.PAGCFGHunyuanAttnProcessor2_0, diffusers.models.attention_processor.LuminaAttnProcessor2_0, diffusers.models.attention_processor.FusedAttnProcessor2_0, diffusers.models.attention_processor.CustomDiffusionXFormersAttnProcessor, diffusers.models.attention_processor.CustomDiffusionAttnProcessor2_0, diffusers.models.attention_processor.SlicedAttnProcessor, diffusers.models.attention_processor.SlicedAttnAddedKVProcessor, diffusers.models.attention_processor.SanaLinearAttnProcessor2_0, diffusers.models.attention_processor.PAGCFGSanaLinearAttnProcessor2_0, diffusers.models.attention_processor.PAGIdentitySanaLinearAttnProcessor2_0, diffusers.models.attention_processor.SanaMultiscaleLinearAttention, diffusers.models.attention_processor.SanaMultiscaleAttnProcessor2_0, diffusers.models.attention_processor.SanaMultiscaleAttentionProjection, diffusers.models.attention_processor.IPAdapterAttnProcessor, diffusers.models.attention_processor.IPAdapterAttnProcessor2_0, diffusers.models.attention_processor.IPAdapterXFormersAttnProcessor, diffusers.models.attention_processor.SD3IPAdapterJointAttnProcessor2_0, diffusers.models.attention_processor.PAGIdentitySelfAttnProcessor2_0, diffusers.models.attention_processor.PAGCFGIdentitySelfAttnProcessor2_0, diffusers.models.attention_processor.LoRAAttnProcessor, diffusers.models.attention_processor.LoRAAttnProcessor2_0, diffusers.models.attention_processor.LoRAXFormersAttnProcessor, diffusers.models.attention_processor.LoRAAttnAddedKVProcessor, typing.Dict[str, typing.Union[diffusers.models.attention_processor.AttnProcessor, diffusers.models.attention_processor.CustomDiffusionAttnProcessor, diffusers.models.attention_processor.AttnAddedKVProcessor, diffusers.models.attention_processor.AttnAddedKVProcessor2_0, diffusers.models.attention_processor.JointAttnProcessor2_0, diffusers.models.attention_processor.PAGJointAttnProcessor2_0, diffusers.models.attention_processor.PAGCFGJointAttnProcessor2_0, diffusers.models.attention_processor.FusedJointAttnProcessor2_0, diffusers.models.attention_processor.AllegroAttnProcessor2_0, diffusers.models.attention_processor.AuraFlowAttnProcessor2_0, diffusers.models.attention_processor.FusedAuraFlowAttnProcessor2_0, diffusers.models.attention_processor.FluxAttnProcessor2_0, diffusers.models.attention_processor.FluxAttnProcessor2_0_NPU, diffusers.models.attention_processor.FusedFluxAttnProcessor2_0, diffusers.models.attention_processor.FusedFluxAttnProcessor2_0_NPU, diffusers.models.attention_processor.CogVideoXAttnProcessor2_0, diffusers.models.attention_processor.FusedCogVideoXAttnProcessor2_0, diffusers.models.attention_processor.XFormersAttnAddedKVProcessor, diffusers.models.attention_processor.XFormersAttnProcessor, diffusers.models.attention_processor.XLAFlashAttnProcessor2_0, diffusers.models.attention_processor.AttnProcessorNPU, diffusers.models.attention_processor.AttnProcessor2_0, diffusers.models.attention_processor.MochiVaeAttnProcessor2_0, diffusers.models.attention_processor.MochiAttnProcessor2_0, diffusers.models.attention_processor.StableAudioAttnProcessor2_0, diffusers.models.attention_processor.HunyuanAttnProcessor2_0, diffusers.models.attention_processor.FusedHunyuanAttnProcessor2_0, diffusers.models.attention_processor.PAGHunyuanAttnProcessor2_0, diffusers.models.attention_processor.PAGCFGHunyuanAttnProcessor2_0, diffusers.models.attention_processor.LuminaAttnProcessor2_0, diffusers.models.attention_processor.FusedAttnProcessor2_0, diffusers.models.attention_processor.CustomDiffusionXFormersAttnProcessor, diffusers.models.attention_processor.CustomDiffusionAttnProcessor2_0, diffusers.models.attention_processor.SlicedAttnProcessor, diffusers.models.attention_processor.SlicedAttnAddedKVProcessor, diffusers.models.attention_processor.SanaLinearAttnProcessor2_0, diffusers.models.attention_processor.PAGCFGSanaLinearAttnProcessor2_0, diffusers.models.attention_processor.PAGIdentitySanaLinearAttnProcessor2_0, diffusers.models.attention_processor.SanaMultiscaleLinearAttention, diffusers.models.attention_processor.SanaMultiscaleAttnProcessor2_0, diffusers.models.attention_processor.SanaMultiscaleAttentionProjection, diffusers.models.attention_processor.IPAdapterAttnProcessor, diffusers.models.attention_processor.IPAdapterAttnProcessor2_0, diffusers.models.attention_processor.IPAdapterXFormersAttnProcessor, diffusers.models.attention_processor.SD3IPAdapterJointAttnProcessor2_0, diffusers.models.attention_processor.PAGIdentitySelfAttnProcessor2_0, diffusers.models.attention_processor.PAGCFGIdentitySelfAttnProcessor2_0, diffusers.models.attention_processor.LoRAAttnProcessor, diffusers.models.attention_processor.LoRAAttnProcessor2_0, diffusers.models.attention_processor.LoRAXFormersAttnProcessor, diffusers.models.attention_processor.LoRAAttnAddedKVProcessor]]]"}],parametersDescription:[{anchor:"diffusers.HunyuanVideoTransformer3DModel.set_attn_processor.processor",description:`<strong>processor</strong> (<code>dict</code> of <code>AttentionProcessor</code> or only <code>AttentionProcessor</code>) &#x2014;
The instantiated processor class or a dictionary of processor classes that will be set as the processor
for <strong>all</strong> <code>Attention</code> layers.</p>
<p>If <code>processor</code> is a dict, the key needs to define the path to the corresponding cross attention
processor. This is strongly recommended when setting trainable attention processors.`,name:"processor"}],source:"https://github.com/huggingface/diffusers/blob/vr_12595/src/diffusers/models/transformers/transformer_hunyuan_video.py#L1016"}}),x=new ne({props:{title:"Transformer2DModelOutput",local:"diffusers.models.modeling_outputs.Transformer2DModelOutput",headingTag:"h2"}}),D=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_12595/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_12595/src/diffusers/models/modeling_outputs.py#L21"}}),H=new ye({props:{source:"https://github.com/huggingface/diffusers/blob/main/docs/source/en/api/models/hunyuan_video_transformer_3d.md"}}),{c(){c=i("meta"),X=r(),L=i("p"),z=r(),l(P.$$.fragment),R=r(),l(y.$$.fragment),Z=r(),T=i("p"),T.innerHTML=ie,j=r(),b=i("p"),b.textContent=ae,E=r(),l($.$$.fragment),N=r(),l(M.$$.fragment),U=r(),d=i("div"),l(V.$$.fragment),ee=r(),F=i("p"),F.innerHTML=ce,oe=r(),g=i("div"),l(v.$$.fragment),se=r(),w=i("p"),w.textContent=fe,W=r(),l(x.$$.fragment),K=r(),f=i("div"),l(D.$$.fragment),te=r(),C=i("p"),C.innerHTML=le,q=r(),l(H.$$.fragment),O=r(),S=i("p"),this.h()},l(e){const o=Ae("svelte-u9bgzb",document.head);c=a(o,"META",{name:!0,content:!0}),o.forEach(s),X=n(e),L=a(e,"P",{}),I(L).forEach(s),z=n(e),u(P.$$.fragment,e),R=n(e),u(y.$$.fragment,e),Z=n(e),T=a(e,"P",{"data-svelte-h":!0}),J(T)!=="svelte-fdngxv"&&(T.innerHTML=ie),j=n(e),b=a(e,"P",{"data-svelte-h":!0}),J(b)!=="svelte-1vuni30"&&(b.textContent=ae),E=n(e),u($.$$.fragment,e),N=n(e),u(M.$$.fragment,e),U=n(e),d=a(e,"DIV",{class:!0});var G=I(d);u(V.$$.fragment,G),ee=n(G),F=a(G,"P",{"data-svelte-h":!0}),J(F)!=="svelte-wu7xtw"&&(F.innerHTML=ce),oe=n(G),g=a(G,"DIV",{class:!0});var Q=I(g);u(v.$$.fragment,Q),se=n(Q),w=a(Q,"P",{"data-svelte-h":!0}),J(w)!=="svelte-1o77hl2"&&(w.textContent=fe),Q.forEach(s),G.forEach(s),W=n(e),u(x.$$.fragment,e),K=n(e),f=a(e,"DIV",{class:!0});var B=I(f);u(D.$$.fragment,B),te=n(B),C=a(B,"P",{"data-svelte-h":!0}),J(C)!=="svelte-1mn2kcc"&&(C.innerHTML=le),B.forEach(s),q=n(e),u(H.$$.fragment,e),O=n(e),S=a(e,"P",{}),I(S).forEach(s),this.h()},h(){k(c,"name","hf:doc:metadata"),k(c,"content",be),k(g,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),k(d,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),k(f,"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){m(document.head,c),t(e,X,o),t(e,L,o),t(e,z,o),_(P,e,o),t(e,R,o),_(y,e,o),t(e,Z,o),t(e,T,o),t(e,j,o),t(e,b,o),t(e,E,o),_($,e,o),t(e,N,o),_(M,e,o),t(e,U,o),t(e,d,o),_(V,d,null),m(d,ee),m(d,F),m(d,oe),m(d,g),_(v,g,null),m(g,se),m(g,w),t(e,W,o),_(x,e,o),t(e,K,o),t(e,f,o),_(D,f,null),m(f,te),m(f,C),t(e,q,o),_(H,e,o),t(e,O,o),t(e,S,o),Y=!0},p:me,i(e){Y||(p(P.$$.fragment,e),p(y.$$.fragment,e),p($.$$.fragment,e),p(M.$$.fragment,e),p(V.$$.fragment,e),p(v.$$.fragment,e),p(x.$$.fragment,e),p(D.$$.fragment,e),p(H.$$.fragment,e),Y=!0)},o(e){h(P.$$.fragment,e),h(y.$$.fragment,e),h($.$$.fragment,e),h(M.$$.fragment,e),h(V.$$.fragment,e),h(v.$$.fragment,e),h(x.$$.fragment,e),h(D.$$.fragment,e),h(H.$$.fragment,e),Y=!1},d(e){e&&(s(X),s(L),s(z),s(R),s(Z),s(T),s(j),s(b),s(E),s(N),s(U),s(d),s(W),s(K),s(f),s(q),s(O),s(S)),s(c),A(P,e),A(y,e),A($,e),A(M,e),A(V),A(v),A(x,e),A(D),A(H,e)}}}const be='{"title":"HunyuanVideoTransformer3DModel","local":"hunyuanvideotransformer3dmodel","sections":[{"title":"HunyuanVideoTransformer3DModel","local":"diffusers.HunyuanVideoTransformer3DModel","sections":[],"depth":2},{"title":"Transformer2DModelOutput","local":"diffusers.models.modeling_outputs.Transformer2DModelOutput","sections":[],"depth":2}],"depth":1}';function $e(de){return _e(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class Fe extends pe{constructor(c){super(),he(this,c,$e,Te,ue,{})}}export{Fe as component};