diffusers/pr_13966/en/_app/immutable/nodes/78.D6USbnGF.js

import"../chunks/DsnmJJEf.js";import{i as w,h as x,C as k,H as r,a as z,D as t,E as J,s as Z}from"../chunks/BtE7mKSK.js";import{p as E,o as I,s as e,f as U,a as T,b as N,c as a,d as b,n as s,r as i}from"../chunks/jDjavuwI.js";const A='{"title":"Lumina2Transformer2DModel","local":"lumina2transformer2dmodel","sections":[{"title":"Lumina2Transformer2DModel","local":"diffusers.Lumina2Transformer2DModel","sections":[],"depth":2},{"title":"Transformer2DModelOutput","local":"diffusers.models.modeling_outputs.Transformer2DModelOutput","sections":[],"depth":2}],"depth":1}';var j=b('<meta name="hf:doc:metadata"/>'),O=b('<p></p> <!> <!> <p>A Diffusion Transformer model for 3D video-like data was introduced in <a href="https://huggingface.co/Alpha-VLLM/Lumina-Image-2.0" rel="nofollow">Lumina Image 2.0</a> by Alpha-VLLM.</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>Lumina2NextDiT: Diffusion model with a Transformer backbone.</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/lumina2_transformer2d#diffusers.Lumina2Transformer2DModel">Lumina2Transformer2DModel</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 F(v,M){E(M,!1),I(()=>{new URLSearchParams(window.location.search).get("fw")}),w();var d=O();x("18t8sbp",_=>{var g=j();Z(g,"content",A),T(_,g)});var m=e(U(d),2);k(m,{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"});var l=e(m,2);r(l,{title:"Lumina2Transformer2DModel",local:"lumina2transformer2dmodel",headingTag:"h1"});var c=e(l,6);z(c,{code:"ZnJvbSUyMGRpZmZ1c2VycyUyMGltcG9ydCUyMEx1bWluYTJUcmFuc2Zvcm1lcjJETW9kZWwlMEElMEF0cmFuc2Zvcm1lciUyMCUzRCUyMEx1bWluYTJUcmFuc2Zvcm1lcjJETW9kZWwuZnJvbV9wcmV0cmFpbmVkKCUyMkFscGhhLVZMTE0lMkZMdW1pbmEtSW1hZ2UtMi4wJTIyJTJDJTIwc3ViZm9sZGVyJTNEJTIydHJhbnNmb3JtZXIlMjIlMkMlMjB0b3JjaF9kdHlwZSUzRHRvcmNoLmJmbG9hdDE2KQ==",highlighted:`<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> Lumina2Transformer2DModel

transformer = Lumina2Transformer2DModel.from_pretrained(<span class="hljs-string">&quot;Alpha-VLLM/Lumina-Image-2.0&quot;</span>, subfolder=<span class="hljs-string">&quot;transformer&quot;</span>, torch_dtype=torch.bfloat16)`,lang:"python",wrap:!1});var f=e(c,2);r(f,{title:"Lumina2Transformer2DModel",local:"diffusers.Lumina2Transformer2DModel",headingTag:"h2"});var n=e(f,2),u=a(n);t(u,{name:"class diffusers.Lumina2Transformer2DModel",anchor:"diffusers.Lumina2Transformer2DModel",source:"https://github.com/huggingface/diffusers/blob/vr_13966/src/diffusers/models/transformers/transformer_lumina2.py#L325",parameters:[{name:"sample_size",val:": int = 128"},{name:"patch_size",val:": int = 2"},{name:"in_channels",val:": int = 16"},{name:"out_channels",val:": int | None = None"},{name:"hidden_size",val:": int = 2304"},{name:"num_layers",val:": int = 26"},{name:"num_refiner_layers",val:": int = 2"},{name:"num_attention_heads",val:": int = 24"},{name:"num_kv_heads",val:": int = 8"},{name:"multiple_of",val:": int = 256"},{name:"ffn_dim_multiplier",val:": float | None = None"},{name:"norm_eps",val:": float = 1e-05"},{name:"scaling_factor",val:": float = 1.0"},{name:"axes_dim_rope",val:": tuple = (32, 32, 32)"},{name:"axes_lens",val:": tuple = (300, 512, 512)"},{name:"cap_feat_dim",val:": int = 1024"}],parametersDescription:[{anchor:"diffusers.Lumina2Transformer2DModel.sample_size",description:`<strong>sample_size</strong> (<code>int</code>) &#x2014; The width of the latent images. This is fixed during training since
it is used to learn a number of position embeddings.`,name:"sample_size"},{anchor:"diffusers.Lumina2Transformer2DModel.patch_size",description:`<strong>patch_size</strong> (<code>int</code>, <em>optional</em>, (<code>int</code>, <em>optional</em>, defaults to 2) &#x2014;
The size of each patch in the image. This parameter defines the resolution of patches fed into the model.`,name:"patch_size"},{anchor:"diffusers.Lumina2Transformer2DModel.in_channels",description:`<strong>in_channels</strong> (<code>int</code>, <em>optional</em>, defaults to 4) &#x2014;
The number of input channels for the model. Typically, this matches the number of channels in the input
images.`,name:"in_channels"},{anchor:"diffusers.Lumina2Transformer2DModel.hidden_size",description:`<strong>hidden_size</strong> (<code>int</code>, <em>optional</em>, defaults to 4096) &#x2014;
The dimensionality of the hidden layers in the model. This parameter determines the width of the model&#x2019;s
hidden representations.`,name:"hidden_size"},{anchor:"diffusers.Lumina2Transformer2DModel.num_layers",description:`<strong>num_layers</strong> (<code>int</code>, <em>optional</em>, default to 32) &#x2014;
The number of layers in the model. This defines the depth of the neural network.`,name:"num_layers"},{anchor:"diffusers.Lumina2Transformer2DModel.num_attention_heads",description:`<strong>num_attention_heads</strong> (<code>int</code>, <em>optional</em>, defaults to 32) &#x2014;
The number of attention heads in each attention layer. This parameter specifies how many separate attention
mechanisms are used.`,name:"num_attention_heads"},{anchor:"diffusers.Lumina2Transformer2DModel.num_kv_heads",description:`<strong>num_kv_heads</strong> (<code>int</code>, <em>optional</em>, defaults to 8) &#x2014;
The number of key-value heads in the attention mechanism, if different from the number of attention heads.
If None, it defaults to num_attention_heads.`,name:"num_kv_heads"},{anchor:"diffusers.Lumina2Transformer2DModel.multiple_of",description:`<strong>multiple_of</strong> (<code>int</code>, <em>optional</em>, defaults to 256) &#x2014;
A factor that the hidden size should be a multiple of. This can help optimize certain hardware
configurations.`,name:"multiple_of"},{anchor:"diffusers.Lumina2Transformer2DModel.ffn_dim_multiplier",description:`<strong>ffn_dim_multiplier</strong> (<code>float</code>, <em>optional</em>) &#x2014;
A multiplier for the dimensionality of the feed-forward network. If None, it uses a default value based on
the model configuration.`,name:"ffn_dim_multiplier"},{anchor:"diffusers.Lumina2Transformer2DModel.norm_eps",description:`<strong>norm_eps</strong> (<code>float</code>, <em>optional</em>, defaults to 1e-5) &#x2014;
A small value added to the denominator for numerical stability in normalization layers.`,name:"norm_eps"},{anchor:"diffusers.Lumina2Transformer2DModel.scaling_factor",description:`<strong>scaling_factor</strong> (<code>float</code>, <em>optional</em>, defaults to 1.0) &#x2014;
A scaling factor applied to certain parameters or layers in the model. This can be used for adjusting the
overall scale of the model&#x2019;s operations.`,name:"scaling_factor"}]});var p=e(u,4),D=a(p);t(D,{name:"forward",anchor:"diffusers.Lumina2Transformer2DModel.forward",source:"https://github.com/huggingface/diffusers/blob/vr_13966/src/diffusers/models/transformers/transformer_lumina2.py#L458",parameters:[{name:"hidden_states",val:": Tensor"},{name:"timestep",val:": Tensor"},{name:"encoder_hidden_states",val:": Tensor"},{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.Lumina2Transformer2DModel.forward.hidden_states",description:`<strong>hidden_states</strong> (<code>torch.Tensor</code> of shape <code>(batch_size, in_channels, height, width)</code>) &#x2014;
Input <code>hidden_states</code>.`,name:"hidden_states"},{anchor:"diffusers.Lumina2Transformer2DModel.forward.timestep",description:`<strong>timestep</strong> (<code>torch.LongTensor</code>) &#x2014;
Used to indicate denoising step.`,name:"timestep"},{anchor:"diffusers.Lumina2Transformer2DModel.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.Lumina2Transformer2DModel.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.Lumina2Transformer2DModel.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.Lumina2Transformer2DModel.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"}],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>
`}),s(2),i(p),i(n);var h=e(n,2);r(h,{title:"Transformer2DModelOutput",local:"diffusers.models.modeling_outputs.Transformer2DModelOutput",headingTag:"h2"});var o=e(h,2),L=a(o);t(L,{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"}]}),s(2),i(o);var y=e(o,2);J(y,{source:"https://github.com/huggingface/diffusers/blob/main/docs/source/en/api/models/lumina2_transformer2d.md"}),s(2),T(v,d),N()}export{F as component};