Add support for lumina2 (#10642)

* Add support for lumina2


---------

Co-authored-by: csuhan <[email protected]>
Co-authored-by: YiYi Xu <[email protected]>
Co-authored-by: Aryan <[email protected]>
Co-authored-by: hlky <[email protected]>

docs/source/en/_toctree.yml

@@ -290,6 +290,8 @@
         title: LatteTransformer3DModel
       - local: api/models/lumina_nextdit2d
         title: LuminaNextDiT2DModel
+      - local: api/models/lumina2_transformer2d
+        title: Lumina2Transformer2DModel
       - local: api/models/ltx_video_transformer3d
         title: LTXVideoTransformer3DModel
       - local: api/models/mochi_transformer3d
@@ -442,6 +444,8 @@
       title: LEDITS++
     - local: api/pipelines/ltx_video
       title: LTXVideo
+    - local: api/pipelines/lumina2
+      title: Lumina 2.0
     - local: api/pipelines/lumina
       title: Lumina-T2X
     - local: api/pipelines/marigold

docs/source/en/api/models/lumina2_transformer2d.md

@@ -0,0 +1,30 @@
+<!-- Copyright 2024 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License. -->
+
+# Lumina2Transformer2DModel
+
+A Diffusion Transformer model for 3D video-like data was introduced in [Lumina Image 2.0](https://huggingface.co/Alpha-VLLM/Lumina-Image-2.0) by Alpha-VLLM.
+
+The model can be loaded with the following code snippet.
+
+```python
+from diffusers import Lumina2Transformer2DModel
+
+transformer = Lumina2Transformer2DModel.from_pretrained("Alpha-VLLM/Lumina-Image-2.0", subfolder="transformer", torch_dtype=torch.bfloat16)
+```
+
+## Lumina2Transformer2DModel
+
+[[autodoc]] Lumina2Transformer2DModel
+
+## Transformer2DModelOutput
+
+[[autodoc]] models.modeling_outputs.Transformer2DModelOutput

docs/source/en/api/pipelines/lumina2.md

@@ -0,0 +1,33 @@
+<!-- Copyright 2024 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License. -->
+
+# Lumina2
+
+[Lumina Image 2.0: A Unified and Efficient Image Generative Model](https://huggingface.co/Alpha-VLLM/Lumina-Image-2.0) is a 2 billion parameter flow-based diffusion transformer capable of generating diverse images from text descriptions.
+
+The abstract from the paper is:
+
+*We introduce Lumina-Image 2.0, an advanced text-to-image model that surpasses previous state-of-the-art methods across multiple benchmarks, while also shedding light on its potential to evolve into a generalist vision intelligence model. Lumina-Image 2.0 exhibits three key properties: (1) Unification – it adopts a unified architecture that treats text and image tokens as a joint sequence, enabling natural cross-modal interactions and facilitating task expansion. Besides, since high-quality captioners can provide semantically better-aligned text-image training pairs, we introduce a unified captioning system, UniCaptioner, which generates comprehensive and precise captions for the model. This not only accelerates model convergence but also enhances prompt adherence, variable-length prompt handling, and task generalization via prompt templates. (2) Efficiency – to improve the efficiency of the unified architecture, we develop a set of optimization techniques that improve semantic learning and fine-grained texture generation during training while incorporating inference-time acceleration strategies without compromising image quality. (3) Transparency – we open-source all training details, code, and models to ensure full reproducibility, aiming to bridge the gap between well-resourced closed-source research teams and independent developers.*
+
+<Tip>
+
+Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers) to learn how to explore the tradeoff between scheduler speed and quality, and see the [reuse components across pipelines](../../using-diffusers/loading#reuse-a-pipeline) section to learn how to efficiently load the same components into multiple pipelines.
+
+</Tip>
+
+## Lumina2Text2ImgPipeline
+
+[[autodoc]] Lumina2Text2ImgPipeline
+  - all
+  - __call__

src/diffusers/__init__.py

@@ -118,6 +118,7 @@
             "Kandinsky3UNet",
             "LatteTransformer3DModel",
             "LTXVideoTransformer3DModel",
+            "Lumina2Transformer2DModel",
             "LuminaNextDiT2DModel",
             "MochiTransformer3DModel",
             "ModelMixin",
@@ -338,6 +339,7 @@
             "LEditsPPPipelineStableDiffusionXL",
             "LTXImageToVideoPipeline",
             "LTXPipeline",
+            "Lumina2Text2ImgPipeline",
             "LuminaText2ImgPipeline",
             "MarigoldDepthPipeline",
             "MarigoldNormalsPipeline",
@@ -634,6 +636,7 @@
             Kandinsky3UNet,
             LatteTransformer3DModel,
             LTXVideoTransformer3DModel,
+            Lumina2Transformer2DModel,
             LuminaNextDiT2DModel,
             MochiTransformer3DModel,
             ModelMixin,
@@ -833,6 +836,7 @@
             LEditsPPPipelineStableDiffusionXL,
             LTXImageToVideoPipeline,
             LTXPipeline,
+            Lumina2Text2ImgPipeline,
             LuminaText2ImgPipeline,
             MarigoldDepthPipeline,
             MarigoldNormalsPipeline,

src/diffusers/models/__init__.py

@@ -72,6 +72,7 @@
     _import_structure["transformers.transformer_flux"] = ["FluxTransformer2DModel"]
     _import_structure["transformers.transformer_hunyuan_video"] = ["HunyuanVideoTransformer3DModel"]
     _import_structure["transformers.transformer_ltx"] = ["LTXVideoTransformer3DModel"]
+    _import_structure["transformers.transformer_lumina2"] = ["Lumina2Transformer2DModel"]
     _import_structure["transformers.transformer_mochi"] = ["MochiTransformer3DModel"]
     _import_structure["transformers.transformer_omnigen"] = ["OmniGenTransformer2DModel"]
     _import_structure["transformers.transformer_sd3"] = ["SD3Transformer2DModel"]
@@ -141,6 +142,7 @@
             HunyuanVideoTransformer3DModel,
             LatteTransformer3DModel,
             LTXVideoTransformer3DModel,
+            Lumina2Transformer2DModel,
             LuminaNextDiT2DModel,
             MochiTransformer3DModel,
             OmniGenTransformer2DModel,

src/diffusers/models/attention.py

@@ -612,7 +612,6 @@ def __init__(
         ffn_dim_multiplier: Optional[float] = None,
     ):
         super().__init__()
-        inner_dim = int(2 * inner_dim / 3)
         # custom hidden_size factor multiplier
         if ffn_dim_multiplier is not None:
             inner_dim = int(ffn_dim_multiplier * inner_dim)

src/diffusers/models/normalization.py

@@ -219,14 +219,13 @@ def __init__(self, embedding_dim: int, norm_eps: float, norm_elementwise_affine:
             4 * embedding_dim,
             bias=True,
         )
-        self.norm = RMSNorm(embedding_dim, eps=norm_eps, elementwise_affine=norm_elementwise_affine)
+        self.norm = RMSNorm(embedding_dim, eps=norm_eps)
 
     def forward(
         self,
         x: torch.Tensor,
         emb: Optional[torch.Tensor] = None,
     ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]:
-        # emb = self.emb(timestep, encoder_hidden_states, encoder_mask)
         emb = self.linear(self.silu(emb))
         scale_msa, gate_msa, scale_mlp, gate_mlp = emb.chunk(4, dim=1)
         x = self.norm(x) * (1 + scale_msa[:, None])
@@ -515,6 +514,16 @@ def forward(self, hidden_states):
             hidden_states = torch_npu.npu_rms_norm(hidden_states, self.weight, epsilon=self.eps)[0]
             if self.bias is not None:
                 hidden_states = hidden_states + self.bias
+        elif is_torch_version(">=", "2.4"):
+            if self.weight is not None:
+                # convert into half-precision if necessary
+                if self.weight.dtype in [torch.float16, torch.bfloat16]:
+                    hidden_states = hidden_states.to(self.weight.dtype)
+            hidden_states = nn.functional.rms_norm(
+                hidden_states, normalized_shape=(hidden_states.shape[-1],), weight=self.weight, eps=self.eps
+            )
+            if self.bias is not None:
+                hidden_states = hidden_states + self.bias
         else:
             input_dtype = hidden_states.dtype
             variance = hidden_states.to(torch.float32).pow(2).mean(-1, keepdim=True)

src/diffusers/models/transformers/__init__.py

@@ -21,6 +21,7 @@
     from .transformer_flux import FluxTransformer2DModel
     from .transformer_hunyuan_video import HunyuanVideoTransformer3DModel
     from .transformer_ltx import LTXVideoTransformer3DModel
+    from .transformer_lumina2 import Lumina2Transformer2DModel
     from .transformer_mochi import MochiTransformer3DModel
     from .transformer_omnigen import OmniGenTransformer2DModel
     from .transformer_sd3 import SD3Transformer2DModel

src/diffusers/models/transformers/lumina_nextdit2d.py

@@ -98,7 +98,7 @@ def __init__(
 
         self.feed_forward = LuminaFeedForward(
             dim=dim,
-            inner_dim=4 * dim,
+            inner_dim=int(4 * 2 * dim / 3),
             multiple_of=multiple_of,
             ffn_dim_multiplier=ffn_dim_multiplier,
         )