diffusers/pr_13966/en/_app/immutable/nodes/91.DccyCxrl.js

import"../chunks/DsnmJJEf.js";import{i as x,h as V,C as S,H as r,a as w,D as s,E as R,s as q}from"../chunks/BtE7mKSK.js";import{p as j,o as z,s as e,f as N,a as y,b as Z,c as t,d as T,n as d,r as a}from"../chunks/jDjavuwI.js";const J='{"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}';var F=T('<meta name="hf:doc:metadata"/>'),E=T('<p></p> <!> <!> <p>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.</p> <p>The model can be loaded with the following code snippet.</p> <!> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>A Transformer model for video-like data used in the Wan-based SkyReels-V2 model.</p> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The <a href="/docs/diffusers/pr_13966/en/api/models/skyreels_v2_transformer_3d#diffusers.SkyReelsV2Transformer3DModel">SkyReelsV2Transformer3DModel</a> forward method.</p></div></div> <!> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"><!> <p>The output of <a href="/docs/diffusers/pr_13966/en/api/models/transformer2d#diffusers.Transformer2DModel">Transformer2DModel</a>.</p></div> <!> <p></p>',1);function O(b,k){j(k,!1),z(()=>{new URLSearchParams(window.location.search).get("fw")}),x();var i=E();V("10gdnk8",h=>{var g=F();q(g,"content",J),y(h,g)});var l=e(N(i),2);S(l,{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"});var c=e(l,2);r(c,{title:"SkyReelsV2Transformer3DModel",local:"skyreelsv2transformer3dmodel",headingTag:"h1"});var m=e(c,6);w(m,{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});var f=e(m,2);r(f,{title:"SkyReelsV2Transformer3DModel",local:"diffusers.SkyReelsV2Transformer3DModel",headingTag:"h2"});var o=e(f,2),p=t(o);s(p,{name:"class diffusers.SkyReelsV2Transformer3DModel",anchor:"diffusers.SkyReelsV2Transformer3DModel",source:"https://github.com/huggingface/diffusers/blob/vr_13966/src/diffusers/models/transformers/transformer_skyreels_v2.py#L518",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"}]});var _=e(p,4),v=t(_);s(v,{name:"forward",anchor:"diffusers.SkyReelsV2Transformer3DModel.forward",source:"https://github.com/huggingface/diffusers/blob/vr_13966/src/diffusers/models/transformers/transformer_skyreels_v2.py#L633",parameters:[{name:"hidden_states",val:": Tensor"},{name:"timestep",val:": LongTensor"},{name:"encoder_hidden_states",val:": Tensor"},{name:"encoder_hidden_states_image",val:": typing.Optional[torch.Tensor] = None"},{name:"enable_diffusion_forcing",val:": bool = False"},{name:"fps",val:": typing.Optional[torch.Tensor] = 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"}],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>
`}),d(2),a(_),a(o);var u=e(o,2);r(u,{title:"Transformer2DModelOutput",local:"diffusers.models.modeling_outputs.Transformer2DModelOutput",headingTag:"h2"});var n=e(u,2),M=t(n);s(M,{name:"class diffusers.models.modeling_outputs.Transformer2DModelOutput",anchor:"diffusers.models.modeling_outputs.Transformer2DModelOutput",source:"https://github.com/huggingface/diffusers/blob/vr_13966/src/diffusers/models/modeling_outputs.py#L21",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_13966/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"}]}),d(2),a(n);var D=e(n,2);R(D,{source:"https://github.com/huggingface/diffusers/blob/main/docs/source/en/api/models/skyreels_v2_transformer_3d.md"}),d(2),y(b,i),Z()}export{O as component};