add moss to accept video wrapper, aimed towards SRT-H

Two related bugs in vivit.py:

1. Transformer.__init__ accepted use_flash_attn but never forwarded it to the
   Attention modules it creates. Since Attention defaults to use_flash_attn=True,
   setting use_flash_attn=False on ViViT had no effect on the factorized_encoder
   variant's spatial and temporal transformers.

2. Attention.forward reshaped the mask from 2D to 4D before the flash/non-flash
   branch (line 82), then attempted to reshape it again inside the non-flash
   branch (line 92). When the non-flash code path is actually reached with a
   mask, einops raises an error because the mask is already 4D.

   These bugs masked each other: bug #1 prevented bug #2 from triggering because
   the non-flash path was never taken even when requested.

Fix: pass use_flash_attn through to Attention in Transformer.__init__, and
remove the redundant second mask rearrange in the non-flash branch.

.github/FUNDING.yml

@@ -1,3 +0,0 @@
-# These are supported funding model platforms
-
-github: [lucidrains]

.github/workflows/python-publish.yml

@@ -1,36 +0,0 @@
-# This workflow will upload a Python Package using Twine when a release is created
-# For more information see: https://help.github.com/en/actions/language-and-framework-guides/using-python-with-github-actions#publishing-to-package-registries
-
-# This workflow uses actions that are not certified by GitHub.
-# They are provided by a third-party and are governed by
-# separate terms of service, privacy policy, and support
-# documentation.
-
-name: Upload Python Package
-
-on:
-  release:
-    types: [published]
-
-jobs:
-  deploy:
-
-    runs-on: ubuntu-latest
-
-    steps:
-    - uses: actions/checkout@v4
-    - name: Set up Python
-      uses: actions/setup-python@v5
-      with:
-        python-version: '3.x'
-    - name: Install dependencies
-      run: |
-        python -m pip install --upgrade pip
-        pip install build
-    - name: Build package
-      run: python -m build
-    - name: Publish package
-      uses: pypa/gh-action-pypi-publish@27b31702a0e7fc50959f5ad993c78deac1bdfc29
-      with:
-        user: __token__
-        password: ${{ secrets.PYPI_API_TOKEN }}

.github/workflows/python-test.yml

@@ -1,34 +0,0 @@
-# This workflow will install Python dependencies, run tests and lint with a variety of Python versions
-# For more information see: https://help.github.com/actions/language-and-framework-guides/using-python-with-github-actions
-
-name: Test
-
-on:
-  push:
-    branches: [ main ]
-  pull_request:
-    branches: [ main ]
-
-jobs:
-  build:
-
-    runs-on: ubuntu-latest
-    strategy:
-      matrix:
-        python-version: [3.8, 3.9]
-
-    steps:
-    - uses: actions/checkout@v4
-    - name: Set up Python ${{ matrix.python-version }}
-      uses: actions/setup-python@v5
-      with:
-        python-version: ${{ matrix.python-version }}
-    - name: Install dependencies
-      run: |
-        python -m pip install --upgrade pip
-        python -m pip install torch==2.4.0 torchvision==0.19.0 --index-url https://download.pytorch.org/whl/cpu
-        python -m pip install -e .
-        python -m pip install pytest
-    - name: Test with pytest
-      run: |
-        pytest -q

.gitignore

@@ -127,3 +127,6 @@ dmypy.json
 
 # Pyre type checker
 .pyre/
+
+# scripts
+*.sh

README.md

@@ -49,7 +49,7 @@
 
 ## Vision Transformer - Pytorch
 
-Implementation of <a href="https://openreview.net/pdf?id=YicbFdNTTy">Vision Transformer</a>, a simple way to achieve SOTA in vision classification with only a single transformer encoder, in Pytorch. Significance is further explained in <a href="https://www.youtube.com/watch?v=TrdevFK_am4">Yannic Kilcher's</a> video. There's really not much to code here, but may as well lay it out for everyone so we expedite the attention revolution.
+Implementation of <a href="https://openreview.net/pdf?id=YicbFdNTTy">Vision Transformer</a>, a simple way to achieve SOTA in vision classification with only a single transformer encoder, in Pytorch. Significance is further explained in <a href="https://www.youtube.com/watch?v=TrdevFK_am4">Yannic Kilcher's</a> video. There's really not much to code here, but may as well lay it out for everyone so we expedite the [attention](https://www.youtube.com/watch?v=eMlx5fFNoYc) revolution.
 
 For a Pytorch implementation with pretrained models, please see Ross Wightman's repository <a href="https://github.com/rwightman/pytorch-image-models">here</a>.
 
@@ -90,26 +90,26 @@ preds = v(img) # (1, 1000)
 
 ## Parameters
 
-- `image_size`: int.  
+- `image_size`: int.
 Image size. If you have rectangular images, make sure your image size is the maximum of the width and height
-- `patch_size`: int.  
-Size of patches. `image_size` must be divisible by `patch_size`.  
+- `patch_size`: int.
+Size of patches. `image_size` must be divisible by `patch_size`.
 The number of patches is: ` n = (image_size // patch_size) ** 2` and `n` **must be greater than 16**.
-- `num_classes`: int.  
+- `num_classes`: int.
 Number of classes to classify.
-- `dim`: int.  
+- `dim`: int.
 Last dimension of output tensor after linear transformation `nn.Linear(..., dim)`.
-- `depth`: int.  
+- `depth`: int.
 Number of Transformer blocks.
-- `heads`: int.  
-Number of heads in Multi-head Attention layer. 
-- `mlp_dim`: int.  
-Dimension of the MLP (FeedForward) layer. 
-- `channels`: int, default `3`.  
-Number of image's channels. 
-- `dropout`: float between `[0, 1]`, default `0.`.  
-Dropout rate. 
-- `emb_dropout`: float between `[0, 1]`, default `0`.  
+- `heads`: int.
+Number of heads in Multi-head Attention layer.
+- `mlp_dim`: int.
+Dimension of the MLP (FeedForward) layer.
+- `channels`: int, default `3`.
+Number of image's channels.
+- `dropout`: float between `[0, 1]`, default `0.`.
+Dropout rate.
+- `emb_dropout`: float between `[0, 1]`, default `0`.
 Embedding dropout rate.
 - `pool`: string, either `cls` token pooling or `mean` pooling
 
@@ -972,7 +972,7 @@ torch.save(model.state_dict(), './pretrained-net.pt')
 
 <img src="./images/mp3.png" width="400px"></img>
 
-New <a href="https://arxiv.org/abs/2207.07611">paper</a> that introduces masked position prediction pre-training criteria. This strategy is more efficient than the Masked Autoencoder strategy and has comparable performance.  
+New <a href="https://arxiv.org/abs/2207.07611">paper</a> that introduces masked position prediction pre-training criteria. This strategy is more efficient than the Masked Autoencoder strategy and has comparable performance.
 
 ```python
 import torch
@@ -1358,10 +1358,10 @@ learner = Dino(
     hidden_layer = 'to_latent',        # hidden layer name or index, from which to extract the embedding
     projection_hidden_size = 256,      # projector network hidden dimension
     projection_layers = 4,             # number of layers in projection network
-    num_classes_K = 65336,             # output logits dimensions (referenced as K in paper)
+    num_classes_K = 65536,             # output logits dimensions (referenced as K in paper)
     student_temp = 0.9,                # student temperature
     teacher_temp = 0.04,               # teacher temperature, needs to be annealed from 0.04 to 0.07 over 30 epochs
-    local_upper_crop_scale = 0.4,      # upper bound for local crop - 0.4 was recommended in the paper 
+    local_upper_crop_scale = 0.4,      # upper bound for local crop - 0.4 was recommended in the paper
     global_lower_crop_scale = 0.5,     # lower bound for global crop - 0.5 was recommended in the paper
     moving_average_decay = 0.9,        # moving average of encoder - paper showed anywhere from 0.9 to 0.999 was ok
     center_moving_average_decay = 0.9, # moving average of teacher centers - paper showed anywhere from 0.9 to 0.999 was ok
@@ -1735,7 +1735,7 @@ Coming from computer vision and new to transformers? Here are some resources tha
 
 ```bibtex
 @misc{touvron2020training,
-    title   = {Training data-efficient image transformers & distillation through attention}, 
+    title   = {Training data-efficient image transformers & distillation through attention},
     author  = {Hugo Touvron and Matthieu Cord and Matthijs Douze and Francisco Massa and Alexandre Sablayrolles and Hervé Jégou},
     year    = {2020},
     eprint  = {2012.12877},
@@ -1768,7 +1768,7 @@ Coming from computer vision and new to transformers? Here are some resources tha
 
 ```bibtex
 @misc{touvron2021going,
-    title   = {Going deeper with Image Transformers}, 
+    title   = {Going deeper with Image Transformers},
     author  = {Hugo Touvron and Matthieu Cord and Alexandre Sablayrolles and Gabriel Synnaeve and Hervé Jégou},
     year    = {2021},
     eprint  = {2103.17239},
@@ -1801,7 +1801,7 @@ Coming from computer vision and new to transformers? Here are some resources tha
 
 ```bibtex
 @misc{heo2021rethinking,
-    title   = {Rethinking Spatial Dimensions of Vision Transformers}, 
+    title   = {Rethinking Spatial Dimensions of Vision Transformers},
     author  = {Byeongho Heo and Sangdoo Yun and Dongyoon Han and Sanghyuk Chun and Junsuk Choe and Seong Joon Oh},
     year    = {2021},
     eprint  = {2103.16302},
@@ -1845,7 +1845,7 @@ Coming from computer vision and new to transformers? Here are some resources tha
 
 ```bibtex
 @misc{su2021roformer,
-    title   = {RoFormer: Enhanced Transformer with Rotary Position Embedding}, 
+    title   = {RoFormer: Enhanced Transformer with Rotary Position Embedding},
     author  = {Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu},
     year    = {2021},
     eprint  = {2104.09864},
@@ -1867,7 +1867,7 @@ Coming from computer vision and new to transformers? Here are some resources tha
 
 ```bibtex
 @misc{chen2021regionvit,
-    title   = {RegionViT: Regional-to-Local Attention for Vision Transformers}, 
+    title   = {RegionViT: Regional-to-Local Attention for Vision Transformers},
     author  = {Chun-Fu Chen and Rameswar Panda and Quanfu Fan},
     year    = {2021},
     eprint  = {2106.02689},
@@ -1878,7 +1878,7 @@ Coming from computer vision and new to transformers? Here are some resources tha
 
 ```bibtex
 @misc{wang2021crossformer,
-    title   = {CrossFormer: A Versatile Vision Transformer Hinging on Cross-scale Attention}, 
+    title   = {CrossFormer: A Versatile Vision Transformer Hinging on Cross-scale Attention},
     author  = {Wenxiao Wang and Lu Yao and Long Chen and Binbin Lin and Deng Cai and Xiaofei He and Wei Liu},
     year    = {2021},
     eprint  = {2108.00154},
@@ -1900,7 +1900,7 @@ Coming from computer vision and new to transformers? Here are some resources tha
 
 ```bibtex
 @misc{he2021masked,
-    title   = {Masked Autoencoders Are Scalable Vision Learners}, 
+    title   = {Masked Autoencoders Are Scalable Vision Learners},
     author  = {Kaiming He and Xinlei Chen and Saining Xie and Yanghao Li and Piotr Dollár and Ross Girshick},
     year    = {2021},
     eprint  = {2111.06377},
@@ -1911,7 +1911,7 @@ Coming from computer vision and new to transformers? Here are some resources tha
 
 ```bibtex
 @misc{xie2021simmim,
-    title   = {SimMIM: A Simple Framework for Masked Image Modeling}, 
+    title   = {SimMIM: A Simple Framework for Masked Image Modeling},
     author  = {Zhenda Xie and Zheng Zhang and Yue Cao and Yutong Lin and Jianmin Bao and Zhuliang Yao and Qi Dai and Han Hu},
     year    = {2021},
     eprint  = {2111.09886},
@@ -1944,7 +1944,7 @@ Coming from computer vision and new to transformers? Here are some resources tha
 
 ```bibtex
 @misc{lee2021vision,
-    title   = {Vision Transformer for Small-Size Datasets}, 
+    title   = {Vision Transformer for Small-Size Datasets},
     author  = {Seung Hoon Lee and Seunghyun Lee and Byung Cheol Song},
     year    = {2021},
     eprint  = {2112.13492},
@@ -1966,7 +1966,7 @@ Coming from computer vision and new to transformers? Here are some resources tha
 
 ```bibtex
 @misc{yang2022scalablevit,
-    title   = {ScalableViT: Rethinking the Context-oriented Generalization of Vision Transformer}, 
+    title   = {ScalableViT: Rethinking the Context-oriented Generalization of Vision Transformer},
     author  = {Rui Yang and Hailong Ma and Jie Wu and Yansong Tang and Xuefeng Xiao and Min Zheng and Xiu Li},
     year    = {2022},
     eprint  = {2203.10790},
@@ -2203,13 +2203,128 @@ Coming from computer vision and new to transformers? Here are some resources tha
 
 ```bibtex
 @misc{carrigg2025decorrelationspeedsvisiontransformers,
-    title   = {Decorrelation Speeds Up Vision Transformers}, 
+    title   = {Decorrelation Speeds Up Vision Transformers},
     author  = {Kieran Carrigg and Rob van Gastel and Melda Yeghaian and Sander Dalm and Faysal Boughorbel and Marcel van Gerven},
     year    = {2025},
     eprint  = {2510.14657},
     archivePrefix = {arXiv},
     primaryClass = {cs.CV},
-    url     = {https://arxiv.org/abs/2510.14657}, 
+    url     = {https://arxiv.org/abs/2510.14657},
+}
+```
+
+```bibtex
+@misc{gopalakrishnan2025decouplingwhatwherepolar,
+    title   = {Decoupling the "What" and "Where" With Polar Coordinate Positional Embeddings},
+    author  = {Anand Gopalakrishnan and Robert Csordás and Jürgen Schmidhuber and Michael C. Mozer},
+    year    = {2025},
+    eprint  = {2509.10534},
+    archivePrefix = {arXiv},
+    primaryClass = {cs.LG},
+    url     = {https://arxiv.org/abs/2509.10534},
+}
+```
+
+```bibtex
+@misc{qiu2025gatedattentionlargelanguage,
+    title   = {Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free},
+    author  = {Zihan Qiu and Zekun Wang and Bo Zheng and Zeyu Huang and Kaiyue Wen and Songlin Yang and Rui Men and Le Yu and Fei Huang and Suozhi Huang and Dayiheng Liu and Jingren Zhou and Junyang Lin},
+    year    = {2025},
+    eprint  = {2505.06708},
+    archivePrefix = {arXiv},
+    primaryClass = {cs.CL},
+    url     = {https://arxiv.org/abs/2505.06708}
+}
+```
+
+```bibtex
+@misc{chen2026postlayernormbackstableexpressive,
+    title   = {Post-LayerNorm Is Back: Stable, ExpressivE, and Deep},
+    author  = {Chen Chen and Lai Wei},
+    year    = {2026},
+    eprint  = {2601.19895},
+    archivePrefix = {arXiv},
+    primaryClass = {cs.LG},
+    url     = {https://arxiv.org/abs/2601.19895},
+}
+```
+
+```bibtex
+@misc{intelligence2025pi06vlalearnsexperience,
+      title     = {$\pi^{*}_{0.6}$: a VLA That Learns From Experience},
+      author    = {Physical Intelligence and Ali Amin and Raichelle Aniceto and Ashwin Balakrishna and Kevin Black and Ken Conley and Grace Connors and James Darpinian and Karan Dhabalia and Jared DiCarlo and Danny Driess and Michael Equi and Adnan Esmail and Yunhao Fang and Chelsea Finn and Catherine Glossop and Thomas Godden and Ivan Goryachev and Lachy Groom and Hunter Hancock and Karol Hausman and Gashon Hussein and Brian Ichter and Szymon Jakubczak and Rowan Jen and Tim Jones and Ben Katz and Liyiming Ke and Chandra Kuchi and Marinda Lamb and Devin LeBlanc and Sergey Levine and Adrian Li-Bell and Yao Lu and Vishnu Mano and Mohith Mothukuri and Suraj Nair and Karl Pertsch and Allen Z. Ren and Charvi Sharma and Lucy Xiaoyang Shi and Laura Smith and Jost Tobias Springenberg and Kyle Stachowicz and Will Stoeckle and Alex Swerdlow and James Tanner and Marcel Torne and Quan Vuong and Anna Walling and Haohuan Wang and Blake Williams and Sukwon Yoo and Lili Yu and Ury Zhilinsky and Zhiyuan Zhou},
+      year      = {2025},
+      eprint    = {2511.14759},
+      archivePrefix = {arXiv},
+      primaryClass = {cs.LG},
+      url       = {https://arxiv.org/abs/2511.14759},
+}
+```
+
+```bibtex
+@misc{kimiteam2026attentionresiduals,
+    title   = {Attention Residuals},
+    author  = {Kimi Team and Guangyu Chen and Yu Zhang and Jianlin Su and Weixin Xu and Siyuan Pan and Yaoyu Wang and Yucheng Wang and Guanduo Chen and Bohong Yin and Yutian Chen and Junjie Yan and Ming Wei and Y. Zhang and Fanqing Meng and Chao Hong and Xiaotong Xie and Shaowei Liu and Enzhe Lu and Yunpeng Tai and Yanru Chen and Xin Men and Haiqing Guo and Y. Charles and Haoyu Lu and Lin Sui and Jinguo Zhu and Zaida Zhou and Weiran He and Weixiao Huang and Xinran Xu and Yuzhi Wang and Guokun Lai and Yulun Du and Yuxin Wu and Zhilin Yang and Xinyu Zhou},
+    year    = {2026},
+    eprint  = {2603.15031},
+    archivePrefix = {arXiv},
+    primaryClass = {cs.CL},
+    url     = {https://arxiv.org/abs/2603.15031},
+}
+```
+
+```bibtex
+@misc{balestriero2025lejepa,
+    title   = {LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics},
+    author  = {Randall Balestriero and Yann LeCun},
+    year    = {2025},
+    eprint  = {2511.08544},
+    archivePrefix = {arXiv},
+    primaryClass = {cs.LG},
+    url     = {https://arxiv.org/abs/2511.08544},
+}
+```
+
+```bibtex
+@misc{oh2026revisitingresidualconnectionsorthogonal,
+    title   = {Revisiting Residual Connections: Orthogonal Updates for Stable and Efficient Deep Networks},
+    author  = {Giyeong Oh and Woohyun Cho and Siyeol Kim and Suhwan Choi and Youngjae Yu},
+    year    = {2026},
+    eprint  = {2505.11881},
+    archivePrefix = {arXiv},
+    primaryClass = {cs.CV},
+    url     = {https://arxiv.org/abs/2505.11881},
+}
+```
+
+```bibtex
+@inproceedings{niu2026learning,
+    title   = {Learning to Grasp Anything By Playing with Random Toys},
+    author  = {Dantong Niu and Yuvan Sharma and Baifeng Shi and Rachel Ding and Matteo Gioia and Haoru Xue and Henry Tsai and Konstantinos Kallidromitis and Anirudh Pai and S. Shankar Sastry and Trevor Darrell and Jitendra Malik and Roei Herzig},
+    booktitle = {The Fourteenth International Conference on Learning Representations},
+    year    = {2026},
+    url     = {https://openreview.net/forum?id=NZDaMcpXZm}
+}
+```
+
+```bibtex
+@misc{marouani2026revisitingclspatchtoken,
+    title   = {Revisiting [CLS] and Patch Token Interaction in Vision Transformers},
+    author  = {Alexis Marouani and Oriane Siméoni and Hervé Jégou and Piotr Bojanowski and Huy V. Vo},
+    year    = {2026},
+    eprint  = {2602.08626},
+    archivePrefix = {arXiv},
+    primaryClass = {cs.CV},
+    url     = {https://arxiv.org/abs/2602.08626},
+}
+```
+
+```bibtex
+@misc{kim2026exploring,
+    title   = {Exploring High-Order Self-Similarity for Video Understanding},
+    author  = {Manjin Kim and Heeseung Kwon and Karteek Alahari and Minsu Cho},
+    year    = {2026},
+    url     = {https://openreview.net/forum?id=Co6SCyBIjo}
 }
 ```
 

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "vit-pytorch"
-version = "1.16.4"
+version = "1.21.0"
 description = "Vision Transformer (ViT) - Pytorch"
 readme = { file = "README.md", content-type = "text/markdown" }
 license = { file = "LICENSE" }
@@ -31,8 +31,8 @@ classifiers = [
   "Programming Language :: Python :: 3.12",
 ]
 dependencies = [
-  "einops>=0.7.0",
-  "torch>=1.10",
+  "einops>=0.8.2",
+  "torch>=2.4",
   "torchvision",
 ]
 
@@ -44,8 +44,8 @@ test = [
 ]
 
 [project.urls]
-Homepage = "https://github.com/lucidrains/vit-pytorch"
-Repository = "https://github.com/lucidrains/vit-pytorch"
+Homepage = "https://codeberg.org/lucidrains/vit-pytorch"
+Repository = "https://codeberg.org/lucidrains/vit-pytorch"
 
 [tool.setuptools]
 include-package-data = true

vit_pytorch/accept_video_wrapper.py

@@ -1,3 +1,5 @@
+from __future__ import annotations
+
 from contextlib import nullcontext
 
 import torch
@@ -5,6 +7,8 @@ from torch import is_tensor, randn
 from torch.nn import Module, Linear, Parameter
 from torch.utils._pytree import tree_flatten, tree_unflatten
 
+from vit_pytorch.vivit_with_moss import MOSS
+
 from einops import rearrange, repeat
 
 # helper functions
@@ -15,6 +19,9 @@ def exists(v):
 def default(v, d):
     return v if exists(v) else d
 
+def pair(t):
+    return t if isinstance(t, tuple) else (t, t)
+
 # classes
 
 class AcceptVideoWrapper(Module):
@@ -26,8 +33,10 @@ class AcceptVideoWrapper(Module):
         dim_emb = None,
         time_seq_len = None,
         embed_is_channel_first = False,
-        output_pos_add_pos_emb = 0, # defaults to first output position to add embedding 
-        proj_embed_to_dim = None
+        output_pos_add_pos_emb = 0, # defaults to first output position to add embedding
+        proj_embed_to_dim = None,
+        patch_size = None,
+        moss: Module | dict | None = None
     ):
         super().__init__()
         self.image_net = image_net
@@ -56,6 +65,24 @@ class AcceptVideoWrapper(Module):
 
         self.embed_is_channel_first = embed_is_channel_first
 
+        # patch size and moss
+
+        if not exists(patch_size):
+            if hasattr(image_net, 'patch_size'):
+                patch_size = image_net.patch_size
+            elif hasattr(image_net, 'vit') and hasattr(image_net.vit, 'patch_size'):
+                patch_size = image_net.vit.patch_size
+
+        self.patch_size = patch_size
+
+        if isinstance(moss, dict):
+            moss = MOSS(**moss)
+
+        self.moss = moss
+
+        if exists(self.moss):
+            assert exists(self.patch_size), '`patch_size` must be provided either on the `image_net` or passed in explicitly if using MOSS'
+
     def forward(
         self,
         video, # (b c t h w)
@@ -70,6 +97,8 @@ class AcceptVideoWrapper(Module):
         if add_time_pos_emb:
             assert time <= self.time_seq_len, f'received video with {time} frames but `time_seq_len` ({self.time_seq_len}) is too low'
 
+        video_height, video_width = video.shape[-2:]
+
         video = rearrange(video, 'b c t h w -> b t c h w')
 
         video = rearrange(video, 'b t ... -> (b t) ...')
@@ -127,6 +156,27 @@ class AcceptVideoWrapper(Module):
 
             outputs[self.output_pos_add_pos_emb] = embed
 
+        # moss - stack of ssts
+        # https://openreview.net/forum?id=Co6SCyBIjo
+
+        if exists(self.moss):
+            outputs = list(outputs)
+            embed = outputs[self.output_pos_add_pos_emb]
+
+            patch_h, patch_w = pair(self.patch_size)
+            num_h, num_w = video_height // patch_h, video_width // patch_w
+            num_patches = num_h * num_w
+
+            num_cls_tokens = embed.shape[-2] - num_patches
+            cls_tokens, patch_tokens = embed[:, :, :num_cls_tokens], embed[:, :, num_cls_tokens:]
+
+            patch_tokens = rearrange(patch_tokens, 'b t (h w) d -> b t h w d', h = num_h, w = num_w)
+            patch_tokens = self.moss(patch_tokens)
+            patch_tokens = rearrange(patch_tokens, 'b t h w d -> b t (h w) d')
+
+            embed = torch.cat((cls_tokens, patch_tokens), dim = -2)
+            outputs[self.output_pos_add_pos_emb] = embed
+
         return tree_unflatten(outputs, tree_spec)
 
 # main
@@ -151,9 +201,28 @@ if __name__ == '__main__':
     # step up the difficulty and return embeddings for robotics
 
     from vit_pytorch.extractor import Extractor
+
     v = Extractor(v)
 
-    video_acceptor = AcceptVideoWrapper(v, add_time_pos_emb = True, output_pos_add_pos_emb = 1, time_seq_len = 12, dim_emb = 1024, proj_embed_to_dim = 512)
+    moss_kwargs = dict(
+        dim = 512,
+        local_time = 3,
+        local_height = 3,
+        local_width = 3,
+        hidden_dim = 64,
+        orders = 2,
+        causal = True
+    )
+
+    video_acceptor = AcceptVideoWrapper(
+        v,
+        add_time_pos_emb = True,
+        output_pos_add_pos_emb = 1,
+        time_seq_len = 12,
+        dim_emb = 1024,
+        proj_embed_to_dim = 512,
+        moss = moss_kwargs
+    )
 
     logits, embeddings = video_acceptor(videos, eval_with_no_grad = True) # always (batch, channels, time, height, width) - time is always dimension 2
 

vit_pytorch/ats_vit.py

@@ -216,7 +216,7 @@ class ViT(nn.Module):
     def __init__(self, *, image_size, patch_size, num_classes, dim, depth, max_tokens_per_depth, heads, mlp_dim, channels = 3, dim_head = 64, dropout = 0., emb_dropout = 0.):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
 

vit_pytorch/distill.py

@@ -25,12 +25,12 @@ class DistillMixin:
         x = self.to_patch_embedding(img)
         b, n, _ = x.shape
 
-        cls_tokens = repeat(self.cls_token, '1 n d -> b n d', b = b)
+        cls_tokens = repeat(self.cls_token, 'n d -> b n d', b = b)
         x = torch.cat((cls_tokens, x), dim = 1)
-        x += self.pos_embedding[:, :(n + 1)]
+        x += self.pos_embedding[:(n + 1)]
 
         if distilling:
-            distill_tokens = repeat(distill_token, '1 n d -> b n d', b = b)
+            distill_tokens = repeat(distill_token, 'n d -> b n d', b = b)
             x = torch.cat((x, distill_tokens), dim = 1)
 
         x = self._attend(x)
@@ -125,7 +125,7 @@ class DistillWrapper(Module):
         self.alpha = alpha
         self.hard = hard
 
-        self.distillation_token = nn.Parameter(torch.randn(1, 1, dim))
+        self.distillation_token = nn.Parameter(torch.randn(1, dim))
 
         self.distill_mlp = nn.Sequential(
             nn.LayerNorm(dim) if mlp_layernorm else nn.Identity(),

vit_pytorch/jumbo_vit.py

@@ -86,7 +86,7 @@ class JumboViT(Module):
     ):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert divisible_by(image_height, patch_height) and divisible_by(image_width, patch_width), 'Image dimensions must be divisible by the patch size.'
 
@@ -103,7 +103,7 @@ class JumboViT(Module):
             h = image_height // patch_height,
             w = image_width // patch_width,
             dim = dim,
-        ) 
+        )
 
         jumbo_cls_dim = dim * jumbo_cls_k
 

vit_pytorch/learnable_memory_vit.py

@@ -108,7 +108,7 @@ class ViT(nn.Module):
     def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, pool = 'cls', channels = 3, dim_head = 64, dropout = 0., emb_dropout = 0.):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
 
@@ -145,7 +145,7 @@ class ViT(nn.Module):
         return x
 
     def forward(self, img):
-        x = self.img_to_tokens(img)        
+        x = self.img_to_tokens(img)
 
         x = self.transformer(x)
 
@@ -160,7 +160,7 @@ class Adapter(nn.Module):
         *,
         vit,
         num_memories_per_layer = 10,
-        num_classes = 2,   
+        num_classes = 2,
     ):
         super().__init__()
         assert isinstance(vit, ViT)
@@ -188,7 +188,7 @@ class Adapter(nn.Module):
         )
 
         # specialized attention mask to preserve the output of the original ViT
-        # it allows the memory CLS token to attend to all other tokens (and the learnable memory layer tokens), but not vice versa        
+        # it allows the memory CLS token to attend to all other tokens (and the learnable memory layer tokens), but not vice versa
 
         attn_mask = torch.ones((num_patches, num_patches), dtype = torch.bool)
         attn_mask = F.pad(attn_mask, (1, num_memories_per_layer), value = False)  # main tokens cannot attend to learnable memories per layer
@@ -203,7 +203,7 @@ class Adapter(nn.Module):
         # add task specific memory tokens
 
         memory_cls_tokens = repeat(self.memory_cls_token, 'd -> b 1 d', b = b)
-        tokens = torch.cat((memory_cls_tokens, tokens), dim = 1)        
+        tokens = torch.cat((memory_cls_tokens, tokens), dim = 1)
 
         # pass memories along with image tokens through transformer for attending
 

vit_pytorch/lejepa.py

@@ -0,0 +1,319 @@
+import random
+from functools import wraps
+
+import torch
+from torch import nn
+from torch.nn import Module
+import torch.nn.functional as F
+
+from torchvision import transforms as T
+from einops import rearrange
+
+# helper functions
+
+def exists(val):
+    return val is not None
+
+def default(val, d):
+    return val if exists(val) else d
+
+def singleton(cache_key):
+    def inner_fn(fn):
+        @wraps(fn)
+        def wrapper(self, *args, **kwargs):
+            instance = getattr(self, cache_key)
+            if instance is not None:
+                return instance
+
+            instance = fn(self, *args, **kwargs)
+            setattr(self, cache_key, instance)
+            return instance
+        return wrapper
+    return inner_fn
+
+def get_module_device(module):
+    return next(module.parameters()).device
+
+def l2norm(t, eps = 1e-6):
+    return F.normalize(t, dim = -1, eps = eps)
+
+# loss function
+
+def sigreg_loss(
+    x,
+    num_slices = 1024,
+    domain = (-5, 5),
+    num_knots = 17
+):
+    # Randall Balestriero - https://arxiv.org/abs/2511.08544
+
+    dim, device = x.shape[-1], x.device
+
+    # slice sampling
+
+    rand_projs = torch.randn((num_slices, dim), device = device)
+    rand_projs = l2norm(rand_projs)
+
+    # integration points
+
+    t = torch.linspace(*domain, num_knots, device = device)
+
+    # theoretical CF for N(0, 1) and Gauss. window
+
+    exp_f = (-0.5 * t.square()).exp()
+
+    # empirical CF
+
+    x_t = torch.einsum('... d, m d -> ... m', x, rand_projs)
+    x_t = rearrange(x_t, '... m -> (...) m')
+
+    x_t = rearrange(x_t, 'n m -> n m 1') * t
+    ecf = (1j * x_t).exp().mean(dim = 0)
+
+    # weighted L2 distance
+
+    err = ecf.sub(exp_f).abs().square().mul(exp_f)
+
+    return torch.trapezoid(err, t, dim = -1).mean()
+
+# augmentation utils
+
+class RandomApply(Module):
+    def __init__(self, fn, p):
+        super().__init__()
+        self.fn = fn
+        self.p = p
+
+    def forward(self, x):
+        if random.random() > self.p:
+            return x
+        return self.fn(x)
+
+# MLP class for projector
+
+class L2Norm(Module):
+    def forward(self, x, eps = 1e-6):
+        return l2norm(x, eps)
+
+class MLP(Module):
+    def __init__(self, dim, dim_out, num_layers, hidden_size = 256):
+        super().__init__()
+
+        layers = []
+        dims = (dim, *((hidden_size,) * (num_layers - 1)))
+
+        for ind, (layer_dim_in, layer_dim_out) in enumerate(zip(dims[:-1], dims[1:])):
+            is_last = ind == (len(dims) - 1)
+
+            layers.extend([
+                nn.Linear(layer_dim_in, layer_dim_out),
+                nn.GELU() if not is_last else nn.Identity()
+            ])
+
+        self.net = nn.Sequential(
+            *layers,
+            L2Norm(),
+            nn.Linear(hidden_size, dim_out)
+        )
+
+    def forward(self, x):
+        return self.net(x)
+
+# wrapper
+
+class NetWrapper(Module):
+    def __init__(self, net, output_dim, projection_hidden_size, projection_num_layers, layer = -2):
+        super().__init__()
+        self.net = net
+        self.layer = layer
+
+        self.projector = None
+        self.projection_hidden_size = projection_hidden_size
+        self.projection_num_layers = projection_num_layers
+        self.output_dim = output_dim
+
+        self.hidden = {}
+        self.hook_registered = False
+
+    def _find_layer(self):
+        if type(self.layer) == str:
+            modules = dict([*self.net.named_modules()])
+            return modules.get(self.layer, None)
+        elif type(self.layer) == int:
+            children = [*self.net.children()]
+            return children[self.layer]
+        return None
+
+    def _hook(self, _, input, output):
+        device = input[0].device
+        self.hidden[device] = output.flatten(1)
+
+    def _register_hook(self):
+        layer = self._find_layer()
+        assert layer is not None, f'hidden layer ({self.layer}) not found'
+        handle = layer.register_forward_hook(self._hook)
+        self.hook_registered = True
+
+    @singleton('projector')
+    def _get_projector(self, hidden):
+        _, dim = hidden.shape
+        projector = MLP(dim, self.output_dim, self.projection_num_layers, self.projection_hidden_size)
+        return projector.to(hidden)
+
+    def get_embedding(self, x):
+        if self.layer == -1:
+            return self.net(x)
+
+        if not self.hook_registered:
+            self._register_hook()
+
+        self.hidden.clear()
+        _ = self.net(x)
+        hidden = self.hidden[x.device]
+        self.hidden.clear()
+
+        assert hidden is not None, f'hidden layer {self.layer} never emitted an output'
+        return hidden
+
+    def forward(self, x, return_projection = True):
+        embed = self.get_embedding(x)
+        if not return_projection:
+            return embed
+
+        projector = self._get_projector(embed)
+        return projector(embed), embed
+
+# main class
+
+class LeJEPA(nn.Module):
+    def __init__(
+        self,
+        net,
+        image_size,
+        hidden_layer = -2,
+        projection_hidden_size = 256,
+        num_classes_K = 65336,
+        projection_layers = 4,
+        local_upper_crop_scale = 0.4,
+        global_lower_crop_scale = 0.5,
+        target_loss_weight = 1.,
+        sigreg_loss_weight = 1.,
+        sigreg_loss_kwargs = dict(
+            num_slices = 1024,
+            domain = (-5, 5),
+            num_knots = 17
+        ),
+        augment_fn = None,
+        augment_fn2 = None
+    ):
+        super().__init__()
+        self.net = net
+
+        # default BYOL augmentation
+
+        DEFAULT_AUG = torch.nn.Sequential(
+            RandomApply(
+                T.ColorJitter(0.8, 0.8, 0.8, 0.2),
+                p = 0.3
+            ),
+            T.RandomGrayscale(p=0.2),
+            T.RandomHorizontalFlip(),
+            RandomApply(
+                T.GaussianBlur((3, 3), (1.0, 2.0)),
+                p = 0.2
+            ),
+            T.Normalize(
+                mean=torch.tensor([0.485, 0.456, 0.406]),
+                std=torch.tensor([0.229, 0.224, 0.225])),
+        )
+
+        self.augment1 = default(augment_fn, DEFAULT_AUG)
+        self.augment2 = default(augment_fn2, DEFAULT_AUG)
+
+        # local and global crops
+
+        self.local_crop = T.RandomResizedCrop((image_size, image_size), scale = (0.05, local_upper_crop_scale))
+        self.global_crop = T.RandomResizedCrop((image_size, image_size), scale = (global_lower_crop_scale, 1.))
+
+        self.encoder = NetWrapper(net, num_classes_K, projection_hidden_size, projection_layers, layer = hidden_layer)
+
+        self.target_loss_weight = target_loss_weight
+        self.sigreg_loss_weight = sigreg_loss_weight
+        self.sigreg_loss_kwargs = sigreg_loss_kwargs
+
+        # get device of network and make wrapper same device
+        device = get_module_device(net)
+        self.to(device)
+
+        # send a mock image tensor to instantiate singleton parameters
+        self.forward(torch.randn(2, 3, image_size, image_size, device=device))
+
+    def forward(
+        self,
+        x,
+        return_embedding = False,
+        return_projection = True
+    ):
+        if return_embedding:
+            return self.encoder(x, return_projection = return_projection)
+
+        image_one, image_two = self.augment1(x), self.augment2(x)
+
+        local_image_one, local_image_two   = self.local_crop(image_one),  self.local_crop(image_two)
+        global_image_one, global_image_two = self.global_crop(image_one), self.global_crop(image_two)
+
+        local_images = torch.cat((local_image_one, local_image_two), dim = 0)
+        proj_locals, _ = self.encoder(local_images)
+        proj_local_one, proj_local_two = proj_locals.chunk(2, dim = 0)
+
+        with torch.no_grad():
+            global_images = torch.cat((global_image_one, global_image_two), dim = 0)
+            proj_globals, _ = self.encoder(global_images)
+            proj_global_one, proj_global_two = proj_globals.chunk(2, dim = 0)
+
+        # invariance loss
+
+        mse_loss = F.mse_loss(proj_local_one, proj_global_two) + F.mse_loss(proj_local_two, proj_global_one)
+
+        # sigreg loss
+
+        sreg_loss = sigreg_loss(proj_locals, **self.sigreg_loss_kwargs)
+
+        return mse_loss * self.target_loss_weight + sreg_loss * self.sigreg_loss_weight
+
+# quick run
+
+if __name__ == '__main__':
+    from vit_pytorch import ViT
+
+    model = ViT(
+        image_size = 256,
+        patch_size = 32,
+        num_classes = 1000,
+        dim = 1024,
+        depth = 6,
+        heads = 8,
+        mlp_dim = 2048
+    )
+
+    learner = LeJEPA(
+        model,
+        image_size = 256,
+        hidden_layer = 'to_latent',          # layer name where output is hidden dimension
+        projection_hidden_size = 256,        # projector network hidden dimension
+        projection_layers = 4,               # number of layers in projection network
+        num_classes_K = 65336,               # output dimension
+        target_loss_weight = 1.0,
+        sigreg_loss_weight = 1.0
+    )
+
+    opt = torch.optim.Adam(learner.parameters(), lr = 3e-4)
+
+    images = torch.randn(8, 3, 256, 256)
+
+    loss = learner(images)
+    opt.zero_grad()
+    loss.backward()
+    opt.step()
+
+    print('loss:', loss.item())

vit_pytorch/levit.py

@@ -182,7 +182,7 @@ class LeViT(nn.Module):
     def forward(self, img):
         x = self.conv_embedding(img)
 
-        x = self.backbone(x)        
+        x = self.backbone(x)
 
         x = self.pool(x)
 

vit_pytorch/mae.py

@@ -52,7 +52,7 @@ class MAE(nn.Module):
         if self.encoder.pool == "cls":
             tokens += self.encoder.pos_embedding[:, 1:(num_patches + 1)]
         elif self.encoder.pool == "mean":
-            tokens += self.encoder.pos_embedding.to(device, dtype=tokens.dtype) 
+            tokens += self.encoder.pos_embedding.to(device, dtype=tokens.dtype)
 
         # calculate of patches needed to be masked, and get random indices, dividing it up for mask vs unmasked
 
@@ -87,7 +87,7 @@ class MAE(nn.Module):
         mask_tokens = mask_tokens + self.decoder_pos_emb(masked_indices)
 
         # concat the masked tokens to the decoder tokens and attend with decoder
-        
+
         decoder_tokens = torch.zeros(batch, num_patches, self.decoder_dim, device=device)
         decoder_tokens[batch_range, unmasked_indices] = unmasked_decoder_tokens
         decoder_tokens[batch_range, masked_indices] = mask_tokens

vit_pytorch/max_vit.py

@@ -77,7 +77,7 @@ class Dropsample(nn.Module):
     def __init__(self, prob = 0):
         super().__init__()
         self.prob = prob
-  
+
     def forward(self, x):
         device = x.device
 

vit_pytorch/max_vit_with_registers.py

@@ -72,7 +72,7 @@ class Dropsample(Module):
     def __init__(self, prob = 0):
         super().__init__()
         self.prob = prob
-  
+
     def forward(self, x):
         device = x.device
 

vit_pytorch/mobile_vit.py

@@ -1,243 +1,243 @@
-import torch
-import torch.nn as nn
-
-from einops import rearrange
-from einops.layers.torch import Reduce
-
-# helpers
-
-def conv_1x1_bn(inp, oup):
-    return nn.Sequential(
-        nn.Conv2d(inp, oup, 1, 1, 0, bias=False),
-        nn.BatchNorm2d(oup),
-        nn.SiLU()
-    )
-
-def conv_nxn_bn(inp, oup, kernel_size=3, stride=1):
-    return nn.Sequential(
-        nn.Conv2d(inp, oup, kernel_size, stride, 1, bias=False),
-        nn.BatchNorm2d(oup),
-        nn.SiLU()
-    )
-
-# classes
-
-class FeedForward(nn.Module):
-    def __init__(self, dim, hidden_dim, dropout=0.):
-        super().__init__()
-        self.net = nn.Sequential(
-            nn.LayerNorm(dim),
-            nn.Linear(dim, hidden_dim),
-            nn.SiLU(),
-            nn.Dropout(dropout),
-            nn.Linear(hidden_dim, dim),
-            nn.Dropout(dropout)
-        )
-
-    def forward(self, x):
-        return self.net(x)
-
-class Attention(nn.Module):
-    def __init__(self, dim, heads=8, dim_head=64, dropout=0.):
-        super().__init__()
-        inner_dim = dim_head * heads
-        self.heads = heads
-        self.scale = dim_head ** -0.5
-
-        self.norm = nn.LayerNorm(dim)
-        self.attend = nn.Softmax(dim=-1)
-        self.dropout = nn.Dropout(dropout)
-
-        self.to_qkv = nn.Linear(dim, inner_dim * 3, bias=False)
-
-        self.to_out = nn.Sequential(
-            nn.Linear(inner_dim, dim),
-            nn.Dropout(dropout)
-        )
-
-    def forward(self, x):
-        x = self.norm(x)
-        qkv = self.to_qkv(x).chunk(3, dim=-1)
-
-        q, k, v = map(lambda t: rearrange(t, 'b p n (h d) -> b p h n d', h=self.heads), qkv)
-
-        dots = torch.matmul(q, k.transpose(-1, -2)) * self.scale
-
-        attn = self.attend(dots)
-        attn = self.dropout(attn)
-
-        out = torch.matmul(attn, v)
-        out = rearrange(out, 'b p h n d -> b p n (h d)')
-        return self.to_out(out)
-
-class Transformer(nn.Module):
-    """Transformer block described in ViT.
-    Paper: https://arxiv.org/abs/2010.11929
-    Based on: https://github.com/lucidrains/vit-pytorch
-    """
-
-    def __init__(self, dim, depth, heads, dim_head, mlp_dim, dropout=0.):
-        super().__init__()
-        self.layers = nn.ModuleList([])
-        for _ in range(depth):
-            self.layers.append(nn.ModuleList([
-                Attention(dim, heads, dim_head, dropout),
-                FeedForward(dim, mlp_dim, dropout)
-            ]))
-
-    def forward(self, x):
-        for attn, ff in self.layers:
-            x = attn(x) + x
-            x = ff(x) + x
-        return x
-
-class MV2Block(nn.Module):
-    """MV2 block described in MobileNetV2.
-    Paper: https://arxiv.org/pdf/1801.04381
-    Based on: https://github.com/tonylins/pytorch-mobilenet-v2
-    """
-
-    def __init__(self, inp, oup, stride=1, expansion=4):
-        super().__init__()
-        self.stride = stride
-        assert stride in [1, 2]
-
-        hidden_dim = int(inp * expansion)
-        self.use_res_connect = self.stride == 1 and inp == oup
-
-        if expansion == 1:
-            self.conv = nn.Sequential(
-                # dw
-                nn.Conv2d(hidden_dim, hidden_dim, 3, stride,
-                          1, groups=hidden_dim, bias=False),
-                nn.BatchNorm2d(hidden_dim),
-                nn.SiLU(),
-                # pw-linear
-                nn.Conv2d(hidden_dim, oup, 1, 1, 0, bias=False),
-                nn.BatchNorm2d(oup),
-            )
-        else:
-            self.conv = nn.Sequential(
-                # pw
-                nn.Conv2d(inp, hidden_dim, 1, 1, 0, bias=False),
-                nn.BatchNorm2d(hidden_dim),
-                nn.SiLU(),
-                # dw
-                nn.Conv2d(hidden_dim, hidden_dim, 3, stride,
-                          1, groups=hidden_dim, bias=False),
-                nn.BatchNorm2d(hidden_dim),
-                nn.SiLU(),
-                # pw-linear
-                nn.Conv2d(hidden_dim, oup, 1, 1, 0, bias=False),
-                nn.BatchNorm2d(oup),
-            )
-
-    def forward(self, x):
-        out = self.conv(x)
-        if self.use_res_connect:
-            out = out + x
-        return out
-
-class MobileViTBlock(nn.Module):
-    def __init__(self, dim, depth, channel, kernel_size, patch_size, mlp_dim, dropout=0.):
-        super().__init__()
-        self.ph, self.pw = patch_size
-
-        self.conv1 = conv_nxn_bn(channel, channel, kernel_size)
-        self.conv2 = conv_1x1_bn(channel, dim)
-
-        self.transformer = Transformer(dim, depth, 4, 8, mlp_dim, dropout)
-
-        self.conv3 = conv_1x1_bn(dim, channel)
-        self.conv4 = conv_nxn_bn(2 * channel, channel, kernel_size)
-
-    def forward(self, x):
-        y = x.clone()
-
-        # Local representations
-        x = self.conv1(x)
-        x = self.conv2(x)
-
-        # Global representations
-        _, _, h, w = x.shape
-        x = rearrange(x, 'b d (h ph) (w pw) -> b (ph pw) (h w) d', ph=self.ph, pw=self.pw)
-        x = self.transformer(x)        
-        x = rearrange(x, 'b (ph pw) (h w) d -> b d (h ph) (w pw)', h=h//self.ph, w=w//self.pw, ph=self.ph, pw=self.pw)
-
-        # Fusion
-        x = self.conv3(x)
-        x = torch.cat((x, y), 1)
-        x = self.conv4(x)
-        return x
-
-class MobileViT(nn.Module):
-    """MobileViT.
-    Paper: https://arxiv.org/abs/2110.02178
-    Based on: https://github.com/chinhsuanwu/mobilevit-pytorch
-    """
-
-    def __init__(
-        self,
-        image_size,
-        dims,
-        channels,
-        num_classes,
-        expansion=4,
-        kernel_size=3,
-        patch_size=(2, 2),
-        depths=(2, 4, 3)
-    ):
-        super().__init__()
-        assert len(dims) == 3, 'dims must be a tuple of 3'
-        assert len(depths) == 3, 'depths must be a tuple of 3'
-
-        ih, iw = image_size
-        ph, pw = patch_size
-        assert ih % ph == 0 and iw % pw == 0
-
-        init_dim, *_, last_dim = channels
-
-        self.conv1 = conv_nxn_bn(3, init_dim, stride=2)
-
-        self.stem = nn.ModuleList([])
-        self.stem.append(MV2Block(channels[0], channels[1], 1, expansion))
-        self.stem.append(MV2Block(channels[1], channels[2], 2, expansion))
-        self.stem.append(MV2Block(channels[2], channels[3], 1, expansion))
-        self.stem.append(MV2Block(channels[2], channels[3], 1, expansion))
-
-        self.trunk = nn.ModuleList([])
-        self.trunk.append(nn.ModuleList([
-            MV2Block(channels[3], channels[4], 2, expansion),
-            MobileViTBlock(dims[0], depths[0], channels[5],
-                           kernel_size, patch_size, int(dims[0] * 2))
-        ]))
-
-        self.trunk.append(nn.ModuleList([
-            MV2Block(channels[5], channels[6], 2, expansion),
-            MobileViTBlock(dims[1], depths[1], channels[7],
-                           kernel_size, patch_size, int(dims[1] * 4))
-        ]))
-
-        self.trunk.append(nn.ModuleList([
-            MV2Block(channels[7], channels[8], 2, expansion),
-            MobileViTBlock(dims[2], depths[2], channels[9],
-                           kernel_size, patch_size, int(dims[2] * 4))
-        ]))
-
-        self.to_logits = nn.Sequential(
-            conv_1x1_bn(channels[-2], last_dim),
-            Reduce('b c h w -> b c', 'mean'),
-            nn.Linear(channels[-1], num_classes, bias=False)
-        )
-
-    def forward(self, x):
-        x = self.conv1(x)
-
-        for conv in self.stem:
-            x = conv(x)
-
-        for conv, attn in self.trunk:
-            x = conv(x)
-            x = attn(x)
-
-        return self.to_logits(x)
+import torch
+import torch.nn as nn
+
+from einops import rearrange
+from einops.layers.torch import Reduce
+
+# helpers
+
+def conv_1x1_bn(inp, oup):
+    return nn.Sequential(
+        nn.Conv2d(inp, oup, 1, 1, 0, bias=False),
+        nn.BatchNorm2d(oup),
+        nn.SiLU()
+    )
+
+def conv_nxn_bn(inp, oup, kernel_size=3, stride=1):
+    return nn.Sequential(
+        nn.Conv2d(inp, oup, kernel_size, stride, 1, bias=False),
+        nn.BatchNorm2d(oup),
+        nn.SiLU()
+    )
+
+# classes
+
+class FeedForward(nn.Module):
+    def __init__(self, dim, hidden_dim, dropout=0.):
+        super().__init__()
+        self.net = nn.Sequential(
+            nn.LayerNorm(dim),
+            nn.Linear(dim, hidden_dim),
+            nn.SiLU(),
+            nn.Dropout(dropout),
+            nn.Linear(hidden_dim, dim),
+            nn.Dropout(dropout)
+        )
+
+    def forward(self, x):
+        return self.net(x)
+
+class Attention(nn.Module):
+    def __init__(self, dim, heads=8, dim_head=64, dropout=0.):
+        super().__init__()
+        inner_dim = dim_head * heads
+        self.heads = heads
+        self.scale = dim_head ** -0.5
+
+        self.norm = nn.LayerNorm(dim)
+        self.attend = nn.Softmax(dim=-1)
+        self.dropout = nn.Dropout(dropout)
+
+        self.to_qkv = nn.Linear(dim, inner_dim * 3, bias=False)
+
+        self.to_out = nn.Sequential(
+            nn.Linear(inner_dim, dim),
+            nn.Dropout(dropout)
+        )
+
+    def forward(self, x):
+        x = self.norm(x)
+        qkv = self.to_qkv(x).chunk(3, dim=-1)
+
+        q, k, v = map(lambda t: rearrange(t, 'b p n (h d) -> b p h n d', h=self.heads), qkv)
+
+        dots = torch.matmul(q, k.transpose(-1, -2)) * self.scale
+
+        attn = self.attend(dots)
+        attn = self.dropout(attn)
+
+        out = torch.matmul(attn, v)
+        out = rearrange(out, 'b p h n d -> b p n (h d)')
+        return self.to_out(out)
+
+class Transformer(nn.Module):
+    """Transformer block described in ViT.
+    Paper: https://arxiv.org/abs/2010.11929
+    Based on: https://github.com/lucidrains/vit-pytorch
+    """
+
+    def __init__(self, dim, depth, heads, dim_head, mlp_dim, dropout=0.):
+        super().__init__()
+        self.layers = nn.ModuleList([])
+        for _ in range(depth):
+            self.layers.append(nn.ModuleList([
+                Attention(dim, heads, dim_head, dropout),
+                FeedForward(dim, mlp_dim, dropout)
+            ]))
+
+    def forward(self, x):
+        for attn, ff in self.layers:
+            x = attn(x) + x
+            x = ff(x) + x
+        return x
+
+class MV2Block(nn.Module):
+    """MV2 block described in MobileNetV2.
+    Paper: https://arxiv.org/pdf/1801.04381
+    Based on: https://github.com/tonylins/pytorch-mobilenet-v2
+    """
+
+    def __init__(self, inp, oup, stride=1, expansion=4):
+        super().__init__()
+        self.stride = stride
+        assert stride in [1, 2]
+
+        hidden_dim = int(inp * expansion)
+        self.use_res_connect = self.stride == 1 and inp == oup
+
+        if expansion == 1:
+            self.conv = nn.Sequential(
+                # dw
+                nn.Conv2d(hidden_dim, hidden_dim, 3, stride,
+                          1, groups=hidden_dim, bias=False),
+                nn.BatchNorm2d(hidden_dim),
+                nn.SiLU(),
+                # pw-linear
+                nn.Conv2d(hidden_dim, oup, 1, 1, 0, bias=False),
+                nn.BatchNorm2d(oup),
+            )
+        else:
+            self.conv = nn.Sequential(
+                # pw
+                nn.Conv2d(inp, hidden_dim, 1, 1, 0, bias=False),
+                nn.BatchNorm2d(hidden_dim),
+                nn.SiLU(),
+                # dw
+                nn.Conv2d(hidden_dim, hidden_dim, 3, stride,
+                          1, groups=hidden_dim, bias=False),
+                nn.BatchNorm2d(hidden_dim),
+                nn.SiLU(),
+                # pw-linear
+                nn.Conv2d(hidden_dim, oup, 1, 1, 0, bias=False),
+                nn.BatchNorm2d(oup),
+            )
+
+    def forward(self, x):
+        out = self.conv(x)
+        if self.use_res_connect:
+            out = out + x
+        return out
+
+class MobileViTBlock(nn.Module):
+    def __init__(self, dim, depth, channel, kernel_size, patch_size, mlp_dim, dropout=0.):
+        super().__init__()
+        self.ph, self.pw = patch_size
+
+        self.conv1 = conv_nxn_bn(channel, channel, kernel_size)
+        self.conv2 = conv_1x1_bn(channel, dim)
+
+        self.transformer = Transformer(dim, depth, 4, 8, mlp_dim, dropout)
+
+        self.conv3 = conv_1x1_bn(dim, channel)
+        self.conv4 = conv_nxn_bn(2 * channel, channel, kernel_size)
+
+    def forward(self, x):
+        y = x.clone()
+
+        # Local representations
+        x = self.conv1(x)
+        x = self.conv2(x)
+
+        # Global representations
+        _, _, h, w = x.shape
+        x = rearrange(x, 'b d (h ph) (w pw) -> b (ph pw) (h w) d', ph=self.ph, pw=self.pw)
+        x = self.transformer(x)
+        x = rearrange(x, 'b (ph pw) (h w) d -> b d (h ph) (w pw)', h=h//self.ph, w=w//self.pw, ph=self.ph, pw=self.pw)
+
+        # Fusion
+        x = self.conv3(x)
+        x = torch.cat((x, y), 1)
+        x = self.conv4(x)
+        return x
+
+class MobileViT(nn.Module):
+    """MobileViT.
+    Paper: https://arxiv.org/abs/2110.02178
+    Based on: https://github.com/chinhsuanwu/mobilevit-pytorch
+    """
+
+    def __init__(
+        self,
+        image_size,
+        dims,
+        channels,
+        num_classes,
+        expansion=4,
+        kernel_size=3,
+        patch_size=(2, 2),
+        depths=(2, 4, 3)
+    ):
+        super().__init__()
+        assert len(dims) == 3, 'dims must be a tuple of 3'
+        assert len(depths) == 3, 'depths must be a tuple of 3'
+
+        ih, iw = image_size
+        ph, pw = patch_size
+        assert ih % ph == 0 and iw % pw == 0
+
+        init_dim, *_, last_dim = channels
+
+        self.conv1 = conv_nxn_bn(3, init_dim, stride=2)
+
+        self.stem = nn.ModuleList([])
+        self.stem.append(MV2Block(channels[0], channels[1], 1, expansion))
+        self.stem.append(MV2Block(channels[1], channels[2], 2, expansion))
+        self.stem.append(MV2Block(channels[2], channels[3], 1, expansion))
+        self.stem.append(MV2Block(channels[2], channels[3], 1, expansion))
+
+        self.trunk = nn.ModuleList([])
+        self.trunk.append(nn.ModuleList([
+            MV2Block(channels[3], channels[4], 2, expansion),
+            MobileViTBlock(dims[0], depths[0], channels[5],
+                           kernel_size, patch_size, int(dims[0] * 2))
+        ]))
+
+        self.trunk.append(nn.ModuleList([
+            MV2Block(channels[5], channels[6], 2, expansion),
+            MobileViTBlock(dims[1], depths[1], channels[7],
+                           kernel_size, patch_size, int(dims[1] * 4))
+        ]))
+
+        self.trunk.append(nn.ModuleList([
+            MV2Block(channels[7], channels[8], 2, expansion),
+            MobileViTBlock(dims[2], depths[2], channels[9],
+                           kernel_size, patch_size, int(dims[2] * 4))
+        ]))
+
+        self.to_logits = nn.Sequential(
+            conv_1x1_bn(channels[-2], last_dim),
+            Reduce('b c h w -> b c', 'mean'),
+            nn.Linear(channels[-1], num_classes, bias=False)
+        )
+
+    def forward(self, x):
+        x = self.conv1(x)
+
+        for conv in self.stem:
+            x = conv(x)
+
+        for conv, attn in self.trunk:
+            x = conv(x)
+            x = attn(x)
+
+        return self.to_logits(x)

vit_pytorch/mp3.py

@@ -107,10 +107,10 @@ class ViT(nn.Module):
     def __init__(self, *, num_classes, image_size, patch_size, dim, depth, heads, mlp_dim, channels = 3, dim_head = 64, dropout = 0.):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
-        
+
         num_patches = (image_height // patch_height) * (image_width // patch_width)
         patch_dim = channels * patch_height * patch_width
 
@@ -178,7 +178,7 @@ class MP3(nn.Module):
 
         attended_tokens = self.vit.transformer(tokens, tokens_unmasked)
         logits = rearrange(self.mlp_head(attended_tokens), 'b n d -> (b n) d')
-        
+
         # Define labels
         labels = repeat(torch.arange(num_patches, device = device), 'n -> (b n)', b = batch)
         loss = F.cross_entropy(logits, labels)

vit_pytorch/na_vit.py

@@ -343,11 +343,11 @@ class NaViT(nn.Module):
 
         # need to know how many images for final attention pooling
 
-        num_images = torch.tensor(num_images, device = device, dtype = torch.long)        
+        num_images = torch.tensor(num_images, device = device, dtype = torch.long)
 
         # to patches
 
-        x = self.to_patch_embedding(patches)        
+        x = self.to_patch_embedding(patches)
 
         # factorized 2d absolute positional embedding
 

vit_pytorch/na_vit_nested_tensor.py

@@ -64,7 +64,7 @@ class Attention(Module):
         self.to_out = nn.Linear(dim_inner, dim, bias = False)
 
     def forward(
-        self, 
+        self,
         x,
         context: Tensor | None = None
     ):

vit_pytorch/na_vit_nested_tensor_3d.py

@@ -64,7 +64,7 @@ class Attention(Module):
         self.to_out = nn.Linear(dim_inner, dim, bias = False)
 
     def forward(
-        self, 
+        self,
         x,
         context: Tensor | None = None
     ):

vit_pytorch/parallel_vit.py

@@ -91,7 +91,7 @@ class ViT(nn.Module):
     def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, pool = 'cls', num_parallel_branches = 2, channels = 3, dim_head = 64, dropout = 0., emb_dropout = 0.):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
 

vit_pytorch/pit.py

@@ -154,7 +154,7 @@ class PiT(nn.Module):
 
         for ind, (layer_depth, layer_heads) in enumerate(zip(depth, heads)):
             not_last = ind < (len(depth) - 1)
-            
+
             layers.append(Transformer(dim, layer_depth, layer_heads, dim_head, mlp_dim, dropout))
 
             if not_last:

vit_pytorch/regionvit.py

@@ -146,7 +146,7 @@ class R2LTransformer(nn.Module):
         region_tokens = rearrange(region_tokens, 'b c h w -> b (h w) c')
 
         # calculate local relative positional bias
-        
+
         h_range = torch.arange(window_size_h, device = device)
         w_range = torch.arange(window_size_w, device = device)
 

vit_pytorch/scalable_vit.py

@@ -187,7 +187,7 @@ class InteractiveWindowedSelfAttention(nn.Module):
 
         out = rearrange(out, '(b x y) h (w1 w2) d -> b (h d) (x w1) (y w2)', x = height // wsz_h, y = width // wsz_w, w1 = wsz_h, w2 = wsz_w)
 
-        # add LIM output 
+        # add LIM output
 
         out = out + local_out
 

vit_pytorch/simple_flash_attn_vit.py

@@ -26,7 +26,7 @@ def posemb_sincos_2d(patches, temperature = 10000, dtype = torch.float32):
     omega = 1. / (temperature ** omega)
 
     y = y.flatten()[:, None] * omega[None, :]
-    x = x.flatten()[:, None] * omega[None, :] 
+    x = x.flatten()[:, None] * omega[None, :]
     pe = torch.cat((x.sin(), x.cos(), y.sin(), y.cos()), dim = 1)
     return pe.type(dtype)
 
@@ -57,7 +57,7 @@ class Attend(nn.Module):
         config = self.cuda_config if q.is_cuda else self.cpu_config
 
         # flash attention - https://arxiv.org/abs/2205.14135
-        
+
         with torch.backends.cuda.sdp_kernel(**config._asdict()):
             out = F.scaled_dot_product_attention(q, k, v)
 
@@ -140,7 +140,7 @@ class SimpleViT(nn.Module):
     def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, channels = 3, dim_head = 64, use_flash = True):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
 

vit_pytorch/simple_flash_attn_vit_3d.py

@@ -34,7 +34,7 @@ def posemb_sincos_3d(patches, temperature = 10000, dtype = torch.float32):
 
     z = z.flatten()[:, None] * omega[None, :]
     y = y.flatten()[:, None] * omega[None, :]
-    x = x.flatten()[:, None] * omega[None, :] 
+    x = x.flatten()[:, None] * omega[None, :]
 
     pe = torch.cat((x.sin(), x.cos(), y.sin(), y.cos(), z.sin(), z.cos()), dim = 1)
 
@@ -52,7 +52,7 @@ class Attend(Module):
 
     def flash_attn(self, q, k, v):
         # flash attention - https://arxiv.org/abs/2205.14135
-        
+
         with torch.backends.cuda.sdp_kernel(**self.config._asdict()):
             out = F.scaled_dot_product_attention(q, k, v)
 
@@ -137,7 +137,7 @@ class SimpleViT(Module):
     def __init__(self, *, image_size, image_patch_size, frames, frame_patch_size, num_classes, dim, depth, heads, mlp_dim, channels = 3, dim_head = 64, use_flash_attn = True):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(image_patch_size)
+        self.patch_size = patch_height, patch_width = pair(image_patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
         assert frames % frame_patch_size == 0, 'Frames must be divisible by the frame patch size'

vit_pytorch/simple_uvit.py

@@ -35,7 +35,7 @@ def FeedForward(dim, hidden_dim):
         nn.Linear(dim, hidden_dim),
         nn.GELU(),
         nn.Linear(hidden_dim, dim),
-    )    
+    )
 
 class Attention(Module):
     def __init__(self, dim, heads = 8, dim_head = 64):
@@ -107,7 +107,7 @@ class SimpleUViT(Module):
     def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, num_register_tokens = 4, channels = 3, dim_head = 64):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert divisible_by(image_height, patch_height) and divisible_by(image_width, patch_width), 'Image dimensions must be divisible by the patch size.'
 

vit_pytorch/simple_vit.py

@@ -81,7 +81,7 @@ class SimpleViT(nn.Module):
     def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, channels = 3, dim_head = 64):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
 
@@ -98,7 +98,7 @@ class SimpleViT(nn.Module):
             h = image_height // patch_height,
             w = image_width // patch_width,
             dim = dim,
-        ) 
+        )
 
         self.transformer = Transformer(dim, depth, heads, dim_head, mlp_dim)
 

vit_pytorch/simple_vit_3d.py

@@ -26,7 +26,7 @@ def posemb_sincos_3d(patches, temperature = 10000, dtype = torch.float32):
 
     z = z.flatten()[:, None] * omega[None, :]
     y = y.flatten()[:, None] * omega[None, :]
-    x = x.flatten()[:, None] * omega[None, :] 
+    x = x.flatten()[:, None] * omega[None, :]
 
     pe = torch.cat((x.sin(), x.cos(), y.sin(), y.cos(), z.sin(), z.cos()), dim = 1)
 
@@ -94,7 +94,7 @@ class SimpleViT(nn.Module):
     def __init__(self, *, image_size, image_patch_size, frames, frame_patch_size, num_classes, dim, depth, heads, mlp_dim, channels = 3, dim_head = 64):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(image_patch_size)
+        self.patch_size = patch_height, patch_width = pair(image_patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
         assert frames % frame_patch_size == 0, 'Frames must be divisible by the frame patch size'

vit_pytorch/simple_vit_attn_residual.py

@@ -0,0 +1,243 @@
+# @bycloud explanation https://www.youtube.com/watch?v=iw1VF8HOCrk
+
+from __future__ import annotations
+
+import torch
+from torch import nn, Tensor
+from torch.nn import Module, ModuleList
+
+from einops import rearrange, repeat
+from einops.layers.torch import Rearrange
+
+# helpers
+
+def exists(v):
+    return v is not None
+
+def default(v, d):
+    return v if exists(v) else d
+
+def last(arr):
+    return arr[-1]
+
+def pair(t):
+    return t if isinstance(t, tuple) else (t, t)
+
+def divisible_by(num, den):
+    return (num % den) == 0
+
+def posemb_sincos_2d(h, w, dim, temperature = 10000, dtype = torch.float32):
+    y, x = torch.meshgrid(torch.arange(h), torch.arange(w), indexing = 'ij')
+    assert divisible_by(dim, 4), 'feature dimension must be multiple of 4 for sincos emb'
+    omega = torch.arange(dim // 4) / (dim // 4 - 1)
+    omega = 1.0 / (temperature ** omega)
+
+    y = y.flatten()[:, None] * omega[None, :]
+    x = x.flatten()[:, None] * omega[None, :]
+    pe = torch.cat((x.sin(), x.cos(), y.sin(), y.cos()), dim = 1)
+    return pe.type(dtype)
+
+# classes
+
+class FeedForward(Module):
+    def __init__(self, dim, hidden_dim):
+        super().__init__()
+        self.net = nn.Sequential(
+            nn.LayerNorm(dim),
+            nn.Linear(dim, hidden_dim),
+            nn.GELU(),
+            nn.Linear(hidden_dim, dim),
+        )
+
+    def forward(self, x):
+        return self.net(x)
+
+class Attention(Module):
+    def __init__(self, dim, heads = 8, dim_head = 64, cross_attend = False):
+        super().__init__()
+        inner_dim = dim_head * heads
+        self.heads = heads
+        self.scale = dim_head ** -0.5
+
+        self.norm = nn.LayerNorm(dim)
+        self.norm_context = nn.LayerNorm(dim) if cross_attend else None
+
+        self.to_q = nn.Linear(dim, inner_dim, bias = False)
+        self.to_kv = nn.Linear(dim, inner_dim * 2, bias = False)
+        self.to_out = nn.Linear(inner_dim, dim, bias = False)
+
+    def forward(self, x, context = None):
+        x = self.norm(x)
+
+        if exists(context):
+            context = self.norm_context(context)
+        else:
+            context = x
+
+        q = self.to_q(x)
+        k, v = self.to_kv(context).chunk(2, dim = -1)
+
+        q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b h n d', h = self.heads), (q, k, v))
+
+        dots = torch.matmul(q, k.transpose(-1, -2)) * self.scale
+        attn = dots.softmax(dim = -1)
+
+        out = torch.matmul(attn, v)
+        out = rearrange(out, 'b h n d -> b n (h d)')
+        return self.to_out(out)
+
+class AttentionResidual(Module):
+    def __init__(self, fn, dim, heads = 8, dim_head = 64, learned_query = True, disable = False):
+        super().__init__()
+        self.fn = fn
+        self.disable = disable
+
+        if disable:
+            return
+
+        self.attn = Attention(dim, heads = heads, dim_head = dim_head, cross_attend = True)
+        self.learned_query = nn.Parameter(torch.randn(dim)) if learned_query else None
+
+    def forward(self, history: list[Tensor]) -> Tensor:
+        if self.disable:
+            return self.fn(last(history))
+
+        batch, seq_len = history[0].shape[:2]
+
+        context = torch.stack(history, dim = 2)
+        context = rearrange(context, 'b n l d -> (b n) l d')
+
+        if exists(self.learned_query):
+            q = repeat(self.learned_query, 'd -> (b n) 1 d', b = batch, n = seq_len)
+        else:
+            q = rearrange(last(history), 'b n d -> (b n) 1 d')
+
+        pooled = self.attn(q, context = context)
+        pooled = rearrange(pooled, '(b n) 1 d -> b n d', b = batch, n = seq_len)
+
+        return self.fn(pooled)
+
+class Transformer(Module):
+    def __init__(self, dim, depth, heads, dim_head, mlp_dim, learned_query = True):
+        super().__init__()
+
+        self.layers = ModuleList([])
+        for ind in range(depth):
+            is_first = ind == 0
+
+            self.layers.append(ModuleList([
+                AttentionResidual(Attention(dim, heads = heads, dim_head = dim_head), dim, heads = heads, dim_head = dim_head, learned_query = learned_query, disable = is_first),
+                AttentionResidual(FeedForward(dim, mlp_dim), dim, heads = heads, dim_head = dim_head, learned_query = learned_query),
+            ]))
+
+        self.final_pool = AttentionResidual(nn.LayerNorm(dim), dim, heads = heads, dim_head = dim_head, learned_query = learned_query)
+
+    def forward(
+        self,
+        x,
+        history: list[Tensor] | None = None,
+        return_history = False
+    ):
+        history = [*default(history, [])]
+
+        history.append(x)
+
+        for attn_residual, ff_residual in self.layers:
+            history.append(attn_residual(history))
+            history.append(ff_residual(history))
+
+        out = self.final_pool(history)
+
+        if return_history:
+            return out, history
+
+        return out
+
+class SimpleViTAttnResidual(Module):
+    def __init__(
+        self,
+        *,
+        image_size,
+        patch_size,
+        num_classes,
+        dim,
+        depth,
+        heads,
+        mlp_dim,
+        channels = 3,
+        dim_head = 64,
+        learned_query = True
+    ):
+        super().__init__()
+        image_height, image_width = pair(image_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
+
+        assert divisible_by(image_height, patch_height) and divisible_by(image_width, patch_width), 'Image dimensions must be divisible by the patch size.'
+
+        patch_dim = channels * patch_height * patch_width
+
+        self.to_patch_embedding = nn.Sequential(
+            Rearrange('b c (h p1) (w p2) -> b (h w) (p1 p2 c)', p1 = patch_height, p2 = patch_width),
+            nn.LayerNorm(patch_dim),
+            nn.Linear(patch_dim, dim),
+            nn.LayerNorm(dim),
+        )
+
+        self.pos_embedding = posemb_sincos_2d(
+            h = image_height // patch_height,
+            w = image_width // patch_width,
+            dim = dim,
+        )
+
+        self.transformer = Transformer(dim, depth, heads, dim_head, mlp_dim, learned_query = learned_query)
+
+        self.to_latent = nn.Identity()
+        self.linear_head = nn.Linear(dim, num_classes)
+
+    def forward(
+        self,
+        img,
+        history: list[Tensor] | None = None,
+        return_history = False
+    ):
+        device, dtype = img.device, img.dtype
+
+        x = self.to_patch_embedding(img)
+        x += self.pos_embedding.to(device, dtype = dtype)
+
+        x = self.transformer(x, history = history, return_history = return_history)
+
+        if return_history:
+            x, history = x
+
+        x = x.mean(dim = 1)
+
+        x = self.to_latent(x)
+        out = self.linear_head(x)
+
+        if return_history:
+            return out, history
+
+        return out
+
+if __name__ == '__main__':
+    for learned_query in (True, False):
+        v = SimpleViTAttnResidual(
+            image_size = 256,
+            patch_size = 32,
+            num_classes = 1000,
+            dim = 1024,
+            depth = 6,
+            heads = 16,
+            mlp_dim = 2048,
+            learned_query = learned_query
+        )
+
+        img = torch.randn(2, 3, 256, 256)
+        preds, history = v(img, return_history = True)
+
+        assert preds.shape == (2, 1000)
+
+        preds, _ = v(img, history = history, return_history = True)
+
+        assert preds.shape == (2, 1000)

vit_pytorch/simple_vit_orthog_residual_update.py

@@ -0,0 +1,206 @@
+# Revisiting Residual Connections: Orthogonal Updates for Stable and Efficient Deep Networks
+# Giyeong Oh et al. https://arxiv.org/abs/2505.11881
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+from torch.nn import Module, ModuleList
+
+from einops import rearrange
+from einops.layers.torch import Rearrange
+
+# helpers
+
+def exists(v):
+    return v is not None
+
+def pair(t):
+    return t if isinstance(t, tuple) else (t, t)
+
+def posemb_sincos_2d(h, w, dim, temperature: int = 10000, dtype = torch.float32):
+    y, x = torch.meshgrid(torch.arange(h), torch.arange(w), indexing="ij")
+    assert (dim % 4) == 0, "feature dimension must be multiple of 4 for sincos emb"
+    omega = torch.arange(dim // 4) / (dim // 4 - 1)
+    omega = 1.0 / (temperature ** omega)
+
+    y = y.flatten()[:, None] * omega[None, :]
+    x = x.flatten()[:, None] * omega[None, :]
+    pe = torch.cat((x.sin(), x.cos(), y.sin(), y.cos()), dim=1)
+    return pe.type(dtype)
+
+# classes
+
+class FeedForward(Module):
+    def __init__(self, dim, hidden_dim):
+        super().__init__()
+        self.net = nn.Sequential(
+            nn.LayerNorm(dim),
+            nn.Linear(dim, hidden_dim),
+            nn.GELU(),
+            nn.Linear(hidden_dim, dim),
+        )
+    def forward(self, x):
+        return self.net(x)
+
+class Attention(Module):
+    def __init__(self, dim, heads = 8, dim_head = 64):
+        super().__init__()
+        inner_dim = dim_head *  heads
+        self.heads = heads
+        self.scale = dim_head ** -0.5
+        self.norm = nn.LayerNorm(dim)
+
+        self.attend = nn.Softmax(dim = -1)
+
+        self.to_qkv = nn.Linear(dim, inner_dim * 3, bias = False)
+        self.to_out = nn.Linear(inner_dim, dim, bias = False)
+
+    def forward(self, x):
+        x = self.norm(x)
+
+        qkv = self.to_qkv(x).chunk(3, dim = -1)
+        q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b h n d', h = self.heads), qkv)
+
+        dots = torch.matmul(q, k.transpose(-1, -2)) * self.scale
+
+        attn = self.attend(dots)
+
+        out = torch.matmul(attn, v)
+        out = rearrange(out, 'b h n d -> b n (h d)')
+        return self.to_out(out)
+
+class OrthogonalResidualUpdate(Module):
+    def __init__(
+        self,
+        block: Module,
+        dim = None,
+        double_precision = True,
+        learned = False
+    ):
+        super().__init__()
+        self.block = block
+        self.double_precision = double_precision
+
+        self.learned = learned
+
+        if learned:
+            assert exists(dim)
+            self.to_modulation = nn.Linear(dim, 2)
+
+    def orthog_proj(self, block_out, residual):
+        use_double, dtype = self.double_precision, residual.dtype
+
+        if use_double:
+            residual, block_out = residual.double(), block_out.double()
+
+        # get orthogonal projection of the attention or feedforward output respect to residual
+
+        unit = F.normalize(residual, dim = -1)
+        parallel = (block_out * unit).sum(dim = -1, keepdim = True) * unit
+        orthogonal = block_out - parallel
+
+        # back to original dtype if double precision
+
+        if use_double:
+            parallel, orthogonal = parallel.to(dtype), orthogonal.to(dtype)
+
+        return parallel, orthogonal
+
+    def forward(self, residual):
+        block_out = self.block(residual)
+
+        parallel_update, orthog_update = self.orthog_proj(block_out, residual)
+
+        if self.learned:
+            parallel_mod, orthog_mod = self.to_modulation(block_out).sigmoid().split(1, dim = -1)
+            parallel_update = parallel_update * parallel_mod
+            orthog_update = orthog_update * orthog_mod
+        else:
+            parallel_update = 0
+
+        return residual + parallel_update + orthog_update
+
+class Transformer(Module):
+    def __init__(self, dim, depth, heads, dim_head, mlp_dim, orthog_residual_update_kwargs: dict = dict()):
+        super().__init__()
+        self.norm = nn.LayerNorm(dim)
+        self.layers = ModuleList([])
+
+        for _ in range(depth):
+            attn = Attention(dim, heads = heads, dim_head = dim_head)
+            ff = FeedForward(dim, mlp_dim)
+
+            self.layers.append(ModuleList([
+                OrthogonalResidualUpdate(attn, dim = dim, **orthog_residual_update_kwargs),
+                OrthogonalResidualUpdate(ff, dim = dim, **orthog_residual_update_kwargs)
+            ]))
+
+    def forward(self, x):
+
+        for attn, ff in self.layers:
+            x = attn(x)
+            x = ff(x)
+
+        return self.norm(x)
+
+class SimpleViT(Module):
+    def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, channels = 3, dim_head = 64, orthog_residual_update_kwargs: dict = dict()):
+        super().__init__()
+        image_height, image_width = pair(image_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
+
+        assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
+
+        patch_dim = channels * patch_height * patch_width
+
+        self.to_patch_embedding = nn.Sequential(
+            Rearrange("b c (h p1) (w p2) -> b (h w) (p1 p2 c)", p1 = patch_height, p2 = patch_width),
+            nn.LayerNorm(patch_dim),
+            nn.Linear(patch_dim, dim),
+            nn.LayerNorm(dim),
+        )
+
+        self.pos_embedding = posemb_sincos_2d(
+            h = image_height // patch_height,
+            w = image_width // patch_width,
+            dim = dim,
+        )
+
+        self.transformer = Transformer(dim, depth, heads, dim_head, mlp_dim, orthog_residual_update_kwargs)
+
+        self.pool = "mean"
+        self.to_latent = nn.Identity()
+
+        self.linear_head = nn.Linear(dim, num_classes)
+
+    def forward(self, img):
+        device = img.device
+
+        x = self.to_patch_embedding(img)
+        x += self.pos_embedding.to(device, dtype=x.dtype)
+
+        x = self.transformer(x)
+        x = x.mean(dim = 1)
+
+        x = self.to_latent(x)
+        return self.linear_head(x)
+
+# quick test
+
+if __name__ == '__main__':
+    vit = SimpleViT(
+        image_size = 256,
+        patch_size = 16,
+        num_classes = 10,
+        dim = 512,
+        depth = 2,
+        heads = 4,
+        mlp_dim = 2048,
+        orthog_residual_update_kwargs = dict(
+            learned = True
+        )
+    )
+
+    images = torch.randn(2, 3, 256, 256)
+
+    assert vit(images).shape == (2, 10)

vit_pytorch/simple_vit_with_fft.py

@@ -82,7 +82,7 @@ class SimpleViT(nn.Module):
     def __init__(self, *, image_size, patch_size, freq_patch_size, num_classes, dim, depth, heads, mlp_dim, channels = 3, dim_head = 64):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
         freq_patch_height, freq_patch_width = pair(freq_patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'

vit_pytorch/simple_vit_with_hyper_connections.py

@@ -167,7 +167,7 @@ class SimpleViT(nn.Module):
     def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, num_residual_streams, num_register_tokens = 4, channels = 3, dim_head = 64):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
 
@@ -186,7 +186,7 @@ class SimpleViT(nn.Module):
             h = image_height // patch_height,
             w = image_width // patch_width,
             dim = dim,
-        ) 
+        )
 
         self.transformer = Transformer(dim, depth, heads, dim_head, mlp_dim, num_residual_streams)
 

vit_pytorch/simple_vit_with_patch_dropout.py

@@ -18,7 +18,7 @@ def posemb_sincos_2d(patches, temperature = 10000, dtype = torch.float32):
     omega = 1. / (temperature ** omega)
 
     y = y.flatten()[:, None] * omega[None, :]
-    x = x.flatten()[:, None] * omega[None, :] 
+    x = x.flatten()[:, None] * omega[None, :]
     pe = torch.cat((x.sin(), x.cos(), y.sin(), y.cos()), dim = 1)
     return pe.type(dtype)
 
@@ -104,7 +104,7 @@ class SimpleViT(nn.Module):
     def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, channels = 3, dim_head = 64, patch_dropout = 0.5):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
 

vit_pytorch/simple_vit_with_qk_norm.py

@@ -102,7 +102,7 @@ class SimpleViT(nn.Module):
     def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, channels = 3, dim_head = 64):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
 
@@ -119,7 +119,7 @@ class SimpleViT(nn.Module):
             h = image_height // patch_height,
             w = image_width // patch_width,
             dim = dim,
-        ) 
+        )
 
         self.transformer = Transformer(dim, depth, heads, dim_head, mlp_dim)
 

vit_pytorch/simple_vit_with_register_tokens.py

@@ -86,7 +86,7 @@ class SimpleViT(nn.Module):
     def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, num_register_tokens = 4, channels = 3, dim_head = 64):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
 
@@ -105,7 +105,7 @@ class SimpleViT(nn.Module):
             h = image_height // patch_height,
             w = image_width // patch_width,
             dim = dim,
-        ) 
+        )
 
         self.transformer = Transformer(dim, depth, heads, dim_head, mlp_dim)
 

vit_pytorch/simple_vit_with_specialized_cls.py

@@ -0,0 +1,205 @@
+from __future__ import annotations
+
+# Alexis Marouani et al. https://arxiv.org/abs/2602.08626
+
+import torch
+from torch import nn, cat, Tensor, is_tensor
+from torch.nn import Module, ModuleList
+
+from einops import rearrange, repeat
+from einops.layers.torch import Rearrange
+
+# helpers
+
+def exists(v):
+    return v is not None
+
+def default(v, d):
+    return v if exists(v) else d
+
+def pair(t):
+    return t if isinstance(t, tuple) else (t, t)
+
+def posemb_sincos_2d(h, w, dim, temperature: int = 10000, dtype = torch.float32):
+    y, x = torch.meshgrid(torch.arange(h), torch.arange(w), indexing="ij")
+    assert (dim % 4) == 0, "feature dimension must be multiple of 4 for sincos emb"
+    omega = torch.arange(dim // 4) / (dim // 4 - 1)
+    omega = 1.0 / (temperature ** omega)
+
+    y = y.flatten()[:, None] * omega[None, :]
+    x = x.flatten()[:, None] * omega[None, :]
+    pe = cat((x.sin(), x.cos(), y.sin(), y.cos()), dim=1)
+    return pe.type(dtype)
+
+# classes
+
+class Specialized(Module):
+    def __init__(self, modules: list[Module]):
+        super().__init__()
+        self.fns = ModuleList(modules)
+
+    def forward(
+        self,
+        x: Tensor | list[Tensor],
+        token_lens: tuple[int, ...] = None
+    ):
+        if is_tensor(x):
+            assert exists(token_lens)
+            x = x.split(token_lens, dim = 1)
+
+        assert len(self.fns) == len(x)
+
+        out = tuple(fn(t) for fn, t in zip(self.fns, x))
+
+        if is_tensor:
+            out = cat(out, dim = 1)
+
+        return out
+
+class FeedForward(Module):
+    def __init__(self, dim, hidden_dim):
+        super().__init__()
+        self.norm = Specialized([
+            nn.LayerNorm(dim),
+            nn.LayerNorm(dim)
+        ])
+
+        self.net = nn.Sequential(
+            nn.Linear(dim, hidden_dim),
+            nn.GELU(),
+            nn.Linear(hidden_dim, dim),
+        )
+    def forward(self, x, token_lens = None):
+        x = self.norm(x, token_lens = token_lens)
+        return self.net(x)
+
+class Attention(Module):
+    def __init__(self, dim, heads = 8, dim_head = 64, specialize_qkv = False):
+        super().__init__()
+        inner_dim = dim_head *  heads
+        self.heads = heads
+        self.scale = dim_head ** -0.5
+
+        self.norm = Specialized([
+            nn.LayerNorm(dim),
+            nn.LayerNorm(dim)
+        ])
+
+        self.attend = nn.Softmax(dim = -1)
+        self.specialize_qkv = specialize_qkv
+
+        if specialize_qkv:
+            self.to_qkv = Specialized([
+                nn.Linear(dim, inner_dim * 3, bias = False),
+                nn.Linear(dim, inner_dim * 3, bias = False)
+            ])
+        else:
+            self.to_qkv = nn.Linear(dim, inner_dim * 3, bias = False)
+
+        self.to_out = nn.Linear(inner_dim, dim, bias = False)
+
+    def forward(self, x, token_lens = None):
+        x = self.norm(x, token_lens = token_lens)
+
+        if self.specialize_qkv:
+            qkv = self.to_qkv(x, token_lens = token_lens).chunk(3, dim = -1)
+        else:
+            qkv = self.to_qkv(x).chunk(3, dim = -1)
+
+        q, k, v = (rearrange(t, 'b n (h d) -> b h n d', h = self.heads) for t in qkv)
+
+        dots = torch.matmul(q, k.transpose(-1, -2)) * self.scale
+
+        attn = self.attend(dots)
+
+        out = torch.matmul(attn, v)
+        out = rearrange(out, 'b h n d -> b n (h d)')
+        return self.to_out(out)
+
+class Transformer(Module):
+    def __init__(self, dim, depth, heads, dim_head, mlp_dim, specialize_qkv_depth):
+        super().__init__()
+        self.norm = Specialized([nn.LayerNorm(dim), nn.LayerNorm(dim)])
+
+        self.layers = ModuleList([])
+
+        for ind in range(depth):
+            specialize_qkv = ind < specialize_qkv_depth
+            self.layers.append(ModuleList([
+                Attention(dim, heads = heads, dim_head = dim_head, specialize_qkv = specialize_qkv),
+                FeedForward(dim, mlp_dim)
+            ]))
+
+    def forward(self, x, token_lens = None):
+        for attn, ff in self.layers:
+            x = attn(x, token_lens = token_lens) + x
+            x = ff(x, token_lens = token_lens) + x
+
+        return self.norm(x, token_lens = token_lens)
+
+class SimpleViT(nn.Module):
+    def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, channels = 3, dim_head = 64, specialize_qkv_depth = None):
+        super().__init__()
+        image_height, image_width = pair(image_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
+
+        assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
+
+        patch_dim = channels * patch_height * patch_width
+
+        self.to_patch_embedding = nn.Sequential(
+            Rearrange("b c (h p1) (w p2) -> b (h w) (p1 p2 c)", p1 = patch_height, p2 = patch_width),
+            nn.LayerNorm(patch_dim),
+            nn.Linear(patch_dim, dim),
+            nn.LayerNorm(dim),
+        )
+
+        self.pos_embedding = posemb_sincos_2d(
+            h = image_height // patch_height,
+            w = image_width // patch_width,
+            dim = dim,
+        )
+
+        self.cls_token = nn.Parameter(torch.randn(dim) * 1e-2)
+
+        specialize_qkv_depth = default(specialize_qkv_depth, depth // 3) # author found just first third of transformer having specialized qkv projection for cls token is enough
+
+        self.transformer = Transformer(dim, depth, heads, dim_head, mlp_dim, specialize_qkv_depth)
+
+        self.pool = 'cls'
+        self.to_latent = nn.Identity()
+
+        self.linear_head = nn.Linear(dim, num_classes)
+
+    def forward(self, img):
+        device = img.device
+
+        x = self.to_patch_embedding(img)
+        x += self.pos_embedding.to(device, dtype=x.dtype)
+
+        b, n, _ = x.shape
+        cls_tokens = repeat(self.cls_token, 'd -> b 1 d', b = b)
+        x = cat((cls_tokens, x), dim = 1)
+
+        x = self.transformer(x, token_lens = (1, n))
+
+        x = x[:, 0]
+
+        x = self.to_latent(x)
+        return self.linear_head(x)
+
+if __name__ == '__main__':
+    v = SimpleViT(
+        image_size = 256,
+        patch_size = 32,
+        num_classes = 1000,
+        dim = 1024,
+        depth = 6,
+        heads = 16,
+        mlp_dim = 2048
+    )
+
+    img = torch.randn(1, 3, 256, 256)
+    out = v(img)
+
+    assert out.shape == (1, 1000)

vit_pytorch/simple_vit_with_value_residual.py

@@ -103,7 +103,7 @@ class SimpleViT(Module):
     def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, channels = 3, dim_head = 64):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
 
@@ -120,7 +120,7 @@ class SimpleViT(Module):
             h = image_height // patch_height,
             w = image_width // patch_width,
             dim = dim,
-        ) 
+        )
 
         self.transformer = Transformer(dim, depth, heads, dim_head, mlp_dim)
 

vit_pytorch/vaat.py

@@ -41,7 +41,7 @@ def posemb_sincos_2d(
     omega = temperature ** -omega
 
     y = y.flatten()[:, None] * omega[None, :]
-    x = x.flatten()[:, None] * omega[None, :] 
+    x = x.flatten()[:, None] * omega[None, :]
 
     pe = cat((x.sin(), x.cos(), y.sin(), y.cos()), dim = 1)
     pe = pe.type(dtype)
@@ -120,6 +120,12 @@ class Attention(Module):
         self.to_q = nn.Linear(dim, inner_dim, bias = False)
         self.to_kv = nn.Linear(dim_context, inner_dim * 2, bias = False)
 
+        self.to_out_gates = nn.Sequential(
+            nn.Linear(dim, heads),
+            Rearrange('b ... h -> b h ... 1'),
+            nn.Sigmoid()
+        )
+
         self.to_out = nn.Sequential(
             nn.Linear(inner_dim, dim),
             nn.Dropout(dropout)
@@ -150,6 +156,9 @@ class Attention(Module):
         attn = self.dropout(attn)
 
         out = torch.matmul(attn, v)
+
+        out = out * self.to_out_gates(x)
+
         out = rearrange(out, 'b h n d -> b n (h d)')
         return self.to_out(out)
 
@@ -222,7 +231,7 @@ class AST(Module):
         self.dim = dim
         self.depth = depth
 
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
         patch_input_dim = patch_height * patch_width
 
         self.patch_size = (patch_height, patch_width)
@@ -348,7 +357,7 @@ class ViT(Module):
         self.depth = depth
 
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
 
@@ -433,7 +442,8 @@ class VAAT(Module):
         self_attn_heads = 4,
         self_attn_dim_head = 32,
         ast_layer_indices: tuple[int, ...] | None = None,
-        vit_layer_indices: tuple[int, ...] | None = None
+        vit_layer_indices: tuple[int, ...] | None = None,
+        num_advantage_bins = 0
     ):
         super().__init__()
 
@@ -471,7 +481,7 @@ class VAAT(Module):
 
         assert len(ast_layer_indices) == depth, f'number of ast layer indices {len(ast_layer_indices)} does not much the VAAT depth {depth}'
 
-        self.register_buffer('ast_layer_indices', tensor(vit_layer_indices), persistent = False)
+        self.register_buffer('ast_layer_indices', tensor(ast_layer_indices), persistent = False)
 
         # handle maybe multiple frames
 
@@ -502,6 +512,14 @@ class VAAT(Module):
 
         self.action_pos_emb = nn.Parameter(torch.randn(action_chunk_len, dim) * 1e-2)
 
+        # handle maybe advantage conditioning
+
+        self.has_advantages = num_advantage_bins > 0
+        self.num_advantage_bins = num_advantage_bins
+
+        if self.has_advantages:
+            self.advantage_emb = nn.Embedding(num_advantage_bins + 1, dim)
+
         self.layers = ModuleList([])
 
         for _ in range(depth):
@@ -531,14 +549,15 @@ class VAAT(Module):
         video_or_image,   # (b v? c t? h w)      - batch, views [wrist + third person or more], channels, maybe time, height, width
         audio_or_spec,    # (b v? t) | (b v?f t) - batch, audio len | batch, spec freq, time
         *,
-        extra = None,     # (b d)                - batch, dim extra     
+        extra = None,     # (b d)                - batch, dim extra
         tasks = None,     # (b)
+        advantages = None,# (b)
         actions = None,   # (b k d)              - batch, action chunk length, action dimension
         return_hiddens = False,
         freeze_vit = False,
         freeze_ast = False
     ):
-        batch = video_or_image.shape[0]
+        batch, device = video_or_image.shape[0], video_or_image.device
         return_loss = exists(actions)
 
         # handle some various input dimensions
@@ -646,53 +665,66 @@ class VAAT(Module):
 
         audio_context = rearrange(audio_hiddens, 'l b v n d -> l b (v n) d')
 
-        # get main action tokens and maybe append extra
+        # main action tokens
 
-        action_tokens = repeat(self.action_pos_emb, 'k d -> b k d', b = batch)
+        action_tokens = repeat(self.action_pos_emb, 'n d -> b n d', b = batch)
 
-        has_extra = exists(extra)
+        # maybe advantage tokens
 
-        if has_extra:
-            assert self.accept_extra_token
+        empty_token = action_tokens[:, 0:0]
 
-            extra_token = self.to_extra_token(extra)
+        maybe_advantage_embed = empty_token
 
-            action_tokens, packed_extra = pack([action_tokens, extra_token], 'b * d')
+        if self.has_advantages and exists(advantages):
+            if isinstance(advantages, int):
+                advantages = torch.full((batch,), advantages, device = device, dtype = torch.long)
+
+            maybe_advantage_embed = self.advantage_emb(advantages + 1)
 
         # register tokens
 
-        register_tokens = repeat(self.register_tokens, 'n d -> b n d', b = batch)
+        register_tokens = empty_token
 
-        action_tokens, registers_packed_shape = pack((register_tokens, action_tokens), 'b * d')
+        if exists(self.register_tokens):
+            register_tokens = repeat(self.register_tokens, 'n d -> b n d', b = batch)
 
-        # cross attention
+        # extra
 
-        hiddens = [action_tokens]
+        maybe_extra_embed = empty_token
+
+        has_extra = exists(extra)
+        if has_extra:
+            assert self.accept_extra_token
+
+            maybe_extra_embed = self.to_extra_token(extra)
+
+        # pack all tokens for attention
+
+        tokens, ps = pack((register_tokens, maybe_advantage_embed, action_tokens, maybe_extra_embed), 'b * d')
+
+        # transformer
+
+        hiddens = [tokens]
 
         for (maybe_film, maybe_self_attn, image_cross_attn, audio_cross_attn, ff), image_layer_context, audio_layer_context in zip(self.layers, image_context, audio_context):
 
-            if exists(tasks):
-                action_tokens = maybe_film(action_tokens, task_emb)
+            if exists(maybe_film) and exists(tasks):
+                tokens = maybe_film(tokens, task_emb)
 
-            action_tokens = image_cross_attn(action_tokens, image_layer_context) + action_tokens
+            tokens = image_cross_attn(tokens, image_layer_context) + tokens
 
-            action_tokens = audio_cross_attn(action_tokens, audio_layer_context) + action_tokens
+            tokens = audio_cross_attn(tokens, audio_layer_context) + tokens
 
             if exists(maybe_self_attn):
-                action_tokens = maybe_self_attn(action_tokens) + action_tokens
+                tokens = maybe_self_attn(tokens) + tokens
 
-            action_tokens = ff(action_tokens) + action_tokens
+            tokens = ff(tokens) + tokens
 
-            hiddens.append(action_tokens)
+            hiddens.append(tokens)
 
-        # unpack registers
+        # unpack register, advantage, action, and extra tokens
 
-        _, action_tokens = unpack(action_tokens, registers_packed_shape, 'b * d')
-
-        # maybe unpack extra
-
-        if has_extra:
-            action_tokens, _ = unpack(action_tokens, packed_extra, 'b * d')
+        maybe_register_embed, maybe_advantage_embed, action_tokens, maybe_extra_embed = unpack(tokens, ps, 'b * d')
 
         # norm and prediction
 
@@ -735,43 +767,51 @@ if __name__ == '__main__':
         mlp_dim = 384 * 4
     )
 
-    vaat = VAAT(
-        vit,
-        ast,
-        dim = 512,
-        depth = 9,
-        heads = 8,
-        dim_head = 64,
-        mlp_dim = 2048,
-        dim_action = 20,
-        action_chunk_len = 7,
-        time_seq_len = 4,
-        num_image_views = 2,
-        num_audio_views = 2,
-        num_tasks = 4,
-        add_self_attn = True,
-        dim_extra_token = 33,               # extra token with some variable dimension
-        vit_layer_indices = (               # extending on the paper, allow for any order of hiddens, and also allow for depth index (which equates to the final embedding output from the vit)
-            0, 0, 1, 1, 2, 2, 3, 3, 4
-        ),
-        ast_layer_indices = (
-            1, 1, 1, 2, 2, 2, 3, 3, 3
+    for num_adv_bins in (0, 2, 10):
+        vaat = VAAT(
+            vit,
+            ast,
+            dim = 512,
+            depth = 9,
+            heads = 8,
+            dim_head = 64,
+            mlp_dim = 2048,
+            dim_action = 20,
+            action_chunk_len = 7,
+            time_seq_len = 4,
+            num_image_views = 2,
+            num_audio_views = 2,
+            num_tasks = 4,
+            num_advantage_bins = num_adv_bins,
+            add_self_attn = True,
+            dim_extra_token = 33,               # extra token with some variable dimension
+            vit_layer_indices = (               # extending on the paper, allow for any order of hiddens, and also allow for depth index (which equates to the final embedding output from the vit)
+                0, 0, 1, 1, 2, 2, 3, 3, 4
+            ),
+            ast_layer_indices = (
+                1, 1, 1, 2, 2, 2, 3, 3, 3
+            )
         )
-    )
 
-    images = torch.randn(2, 2, 3, 4, 256, 256) # (2 views with 4 frames)
-    audio = torch.randn(2, 2, 14_100 * 5)
+        images = torch.randn(2, 2, 3, 4, 256, 256) # (2 views with 4 frames)
+        audio = torch.randn(2, 2, 14_100 * 5)
 
-    tasks = torch.randint(0, 4, (2,))
-    extra = torch.randn(2, 33)                 # extra internal state
+        tasks = torch.randint(0, 4, (2,))
+        extra = torch.randn(2, 33)                 # extra internal state
 
-    actions = torch.randn(2, 7, 20)         # actions for learning
+        # advantage conditioning
 
-    loss = vaat(images, audio, actions = actions, tasks = tasks, extra = extra, freeze_vit = True)
-    loss.backward()
+        advantages = None
+        if num_adv_bins > 0:
+            advantages = torch.randint(-1, num_adv_bins, (2,))
 
-    # after much training
+        actions = torch.randn(2, 7, 20)         # actions for learning
 
-    pred_actions, hiddens = vaat(images, audio, tasks = tasks, extra = extra, return_hiddens = True)
+        loss = vaat(images, audio, actions = actions, advantages = advantages, tasks = tasks, extra = extra, freeze_vit = True)
+        loss.backward()
 
-    assert pred_actions.shape == (2, 7, 20)
+        # after much training
+
+        pred_actions, hiddens = vaat(images, audio, advantages = advantages, tasks = tasks, extra = extra, return_hiddens = True)
+
+        assert pred_actions.shape == (2, 7, 20)

vit_pytorch/vat.py

@@ -92,6 +92,12 @@ class Attention(Module):
         self.to_q = nn.Linear(dim, inner_dim, bias = False)
         self.to_kv = nn.Linear(dim_context, inner_dim * 2, bias = False)
 
+        self.to_out_gates = nn.Sequential(
+            nn.Linear(dim, heads),
+            Rearrange('b ... h -> b h ... 1'),
+            nn.Sigmoid()
+        )
+
         self.to_out = nn.Sequential(
             nn.Linear(inner_dim, dim),
             nn.Dropout(dropout)
@@ -122,6 +128,8 @@ class Attention(Module):
         attn = self.dropout(attn)
 
         out = torch.matmul(attn, v)
+        out = out * self.to_out_gates(x) # https://arxiv.org/abs/2505.06708
+
         out = rearrange(out, 'b h n d -> b n (h d)')
         return self.to_out(out)
 
@@ -189,7 +197,7 @@ class ViT(Module):
         self.depth = depth
 
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
 
@@ -270,7 +278,8 @@ class VAT(Module):
         add_self_attn = True,  # in the paper, they didn't have any ways for the action token to exchange information with the extra token, so we'll just add it as an option
         self_attn_heads = 4,
         self_attn_dim_head = 32,
-        vit_layer_indices: tuple[int, ...] | None = None
+        vit_layer_indices: tuple[int, ...] | None = None,
+        num_advantage_bins = 0
     ):
         super().__init__()
 
@@ -316,6 +325,14 @@ class VAT(Module):
 
         self.action_pos_emb = nn.Parameter(torch.randn(action_chunk_len, dim) * 1e-2)
 
+        # handle maybe advantage conditioning
+
+        self.has_advantages = num_advantage_bins > 0
+        self.num_advantage_bins = num_advantage_bins
+
+        if self.has_advantages:
+            self.advantage_emb = nn.Embedding(num_advantage_bins + 1, dim)
+
         self.layers = ModuleList([])
 
         for _ in range(depth):
@@ -343,13 +360,14 @@ class VAT(Module):
         self,
         video_or_image,   # (b v? c t? h w) - batch, views [wrist + third person or more], channels, maybe time, height, width
         *,
-        extra = None,     # (b d)           - batch, dim extra     
+        extra = None,     # (b d)           - batch, dim extra
         tasks = None,     # (b)
+        advantages = None,# (b)
         actions = None,   # (b k d)         - batch, action chunk length, action dimension
         return_hiddens = False,
         freeze_vit = False
     ):
-        batch = video_or_image.shape[0]
+        batch, device = video_or_image.shape[0], video_or_image.device
         return_loss = exists(actions)
 
         # handle some various input dimensions
@@ -415,51 +433,64 @@ class VAT(Module):
 
         context = rearrange(hiddens, 'l b v t n d -> l b (v t n) d')
 
-        # get main action tokens and maybe append extra
+        # main action tokens
 
-        action_tokens = repeat(self.action_pos_emb, 'k d -> b k d', b = batch)
+        action_tokens = repeat(self.action_pos_emb, 'n d -> b n d', b = batch)
 
-        has_extra = exists(extra)
+        # maybe advantage tokens
 
-        if has_extra:
-            assert self.accept_extra_token
+        empty_token = action_tokens[:, 0:0]
 
-            extra_token = self.to_extra_token(extra)
+        maybe_advantage_embed = empty_token
 
-            action_tokens, packed_extra = pack([action_tokens, extra_token], 'b * d')
+        if self.has_advantages and exists(advantages):
+            if isinstance(advantages, int):
+                advantages = torch.full((batch,), advantages, device = device, dtype = torch.long)
+
+            maybe_advantage_embed = self.advantage_emb(advantages + 1)
 
         # register tokens
 
-        register_tokens = repeat(self.register_tokens, 'n d -> b n d', b = batch)
+        register_tokens = empty_token
 
-        action_tokens, registers_packed_shape = pack((register_tokens, action_tokens), 'b * d')
+        if exists(self.register_tokens):
+            register_tokens = repeat(self.register_tokens, 'n d -> b n d', b = batch)
 
-        # cross attention
+        # extra
 
-        hiddens = [action_tokens]
+        maybe_extra_embed = empty_token
+
+        has_extra = exists(extra)
+        if has_extra:
+            assert self.accept_extra_token
+
+            maybe_extra_embed = self.to_extra_token(extra)
+
+        # pack all tokens for attention
+
+        tokens, ps = pack((register_tokens, maybe_advantage_embed, action_tokens, maybe_extra_embed), 'b * d')
+
+        # transformer
+
+        hiddens = [tokens]
 
         for (maybe_film, maybe_self_attn, cross_attn, ff), layer_context in zip(self.layers, context):
 
-            if exists(tasks):
-                action_tokens = maybe_film(action_tokens, task_emb)
+            if exists(maybe_film) and exists(tasks):
+                tokens = maybe_film(tokens, task_emb)
 
-            action_tokens = cross_attn(action_tokens, layer_context) + action_tokens
+            tokens = cross_attn(tokens, layer_context) + tokens
 
             if exists(maybe_self_attn):
-                action_tokens = maybe_self_attn(action_tokens) + action_tokens
+                tokens = maybe_self_attn(tokens) + tokens
 
-            action_tokens = ff(action_tokens) + action_tokens
+            tokens = ff(tokens) + tokens
 
-            hiddens.append(action_tokens)
+            hiddens.append(tokens)
 
-        # unpack registers
+        # unpack register, advantage, action, and extra tokens
 
-        _, action_tokens = unpack(action_tokens, registers_packed_shape, 'b * d')
-
-        # maybe unpack extra
-
-        if has_extra:
-            action_tokens, _ = unpack(action_tokens, packed_extra, 'b * d')
+        maybe_register_embed, maybe_advantage_embed, action_tokens, maybe_extra_embed = unpack(tokens, ps, 'b * d')
 
         # norm and prediction
 
@@ -493,36 +524,44 @@ if __name__ == '__main__':
         mlp_dim = 1024
     )
 
-    vat = VAT(
-        vit,
-        dim = 512,
-        depth = 9,
-        heads = 8,
-        dim_head = 64,
-        mlp_dim = 2048,
-        dim_action = 20,
-        action_chunk_len = 7,
-        time_seq_len = 4,
-        num_views = 2,
-        num_tasks = 4,
-        add_self_attn = True,
-        dim_extra_token = 33,               # extra token with some variable dimension
-        vit_layer_indices = (               # extending on the paper, allow for any order of hiddens, and also allow for depth index (which equates to the final embedding output from the vit)
-            0, 0, 1, 1, 2, 2, 3, 3, 4
+    for num_adv_bins in (0, 2, 10):
+        vat = VAT(
+            vit,
+            dim = 512,
+            depth = 9,
+            heads = 8,
+            dim_head = 64,
+            mlp_dim = 2048,
+            dim_action = 20,
+            action_chunk_len = 7,
+            time_seq_len = 4,
+            num_views = 2,
+            num_tasks = 4,
+            num_advantage_bins = num_adv_bins,
+            add_self_attn = True,
+            dim_extra_token = 33,               # extra token with some variable dimension
+            vit_layer_indices = (               # extending on the paper, allow for any order of hiddens, and also allow for depth index (which equates to the final embedding output from the vit)
+                0, 0, 1, 1, 2, 2, 3, 3, 4
+            )
         )
-    )
 
-    images = torch.randn(2, 2, 3, 4, 256, 256) # (2 views with 4 frames)
-    tasks = torch.randint(0, 4, (2,))
-    extra = torch.randn(2, 33)                 # extra internal state
+        images = torch.randn(2, 2, 3, 4, 256, 256) # (2 views with 4 frames)
+        tasks = torch.randint(0, 4, (2,))
+        extra = torch.randn(2, 33)                 # extra internal state
 
-    actions = torch.randn(2, 7, 20)         # actions for learning
+        # advantage conditioning
 
-    loss = vat(images, actions = actions, tasks = tasks, extra = extra, freeze_vit = True)
-    loss.backward()
+        advantages = None
+        if num_adv_bins > 0:
+            advantages = torch.randint(-1, num_adv_bins, (2,))
 
-    # after much training
+        actions = torch.randn(2, 7, 20)         # actions for learning
 
-    pred_actions, hiddens = vat(images, tasks = tasks, extra = extra, return_hiddens = True)
+        loss = vat(images, actions = actions, advantages = advantages, tasks = tasks, extra = extra, freeze_vit = True)
+        loss.backward()
 
-    assert pred_actions.shape == (2, 7, 20)
+        # after much training
+
+        pred_actions, hiddens = vat(images, advantages = advantages, tasks = tasks, extra = extra, return_hiddens = True)
+
+        assert pred_actions.shape == (2, 7, 20)

vit_pytorch/vat_siglip.py

@@ -0,0 +1,521 @@
+from __future__ import annotations
+from contextlib import nullcontext
+from pathlib import Path
+
+import torch
+import torch.nn.functional as F
+from torch import nn, cat, stack, tensor, einsum
+from torch.nn import Module, ModuleList
+
+from einops import rearrange, repeat, pack, unpack
+from einops.layers.torch import Rearrange
+
+# helpers
+
+def exists(v):
+    return v is not None
+
+def default(v, d):
+    return v if exists(v) else d
+
+def pair(t):
+    return t if isinstance(t, tuple) else (t, t)
+
+# attention
+
+class Attention(Module):
+    def __init__(
+        self,
+        dim,
+        dim_context = None,
+        heads = 8,
+        dim_head = 64,
+        dropout = 0.,
+        norm_eps = 1e-6,
+        gate_attn = False
+    ):
+        super().__init__()
+        self.heads = heads
+        self.scale = dim_head ** -0.5
+        inner_dim = dim_head * heads
+
+        self.norm = nn.LayerNorm(dim, eps = norm_eps)
+
+        self.is_cross_attn = exists(dim_context)
+        dim_context = default(dim_context, dim)
+        self.norm_context = nn.LayerNorm(dim_context, eps = norm_eps) if self.is_cross_attn else None
+
+        self.to_q = nn.Linear(dim, inner_dim)
+        self.to_kv = nn.Linear(dim_context, inner_dim * 2)
+
+        self.to_out_gates = nn.Sequential(
+            nn.Linear(dim, heads),
+            Rearrange('b ... h -> b h ... 1'),
+            nn.Sigmoid()
+        ) if gate_attn else None
+
+        self.to_out = nn.Sequential(
+            nn.Linear(inner_dim, dim),
+            nn.Dropout(dropout)
+        )
+
+    def forward(self, x, context = None):
+        x = self.norm(x)
+
+        if self.is_cross_attn:
+            assert exists(context)
+            context = self.norm_context(context)
+        else:
+            context = x
+
+        q = self.to_q(x)
+        k, v = self.to_kv(context).chunk(2, dim = -1)
+
+        q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b h n d', h = self.heads), (q, k, v))
+
+        sim = einsum('b h i d, b h j d -> b h i j', q, k) * self.scale
+        attn = sim.softmax(dim = -1)
+
+        out = einsum('b h i j, b h j d -> b h i d', attn, v)
+
+        if exists(self.to_out_gates):
+            out = out * self.to_out_gates(x) # https://arxiv.org/abs/2505.06708
+
+        out = rearrange(out, 'b h n d -> b n (h d)')
+        return self.to_out(out)
+
+def FeedForward(
+    dim,
+    dim_inner,
+    norm_eps = 1e-6
+):
+    return nn.Sequential(
+        nn.LayerNorm(dim, eps = norm_eps),
+        nn.Linear(dim, dim_inner),
+        nn.GELU(approximate = 'tanh'),
+        nn.Linear(dim_inner, dim)
+    )
+
+class SigLIP(Module):
+    def __init__(
+        self,
+        image_size = 224,
+        patch_size = 14,
+        dim = 1152,
+        depth = 27,
+        heads = 16,
+        mlp_dim = 4304,
+        norm_eps = 1e-6
+    ):
+        super().__init__()
+        self.dim = dim
+        self.depth = depth
+        num_patches = (image_size // patch_size) ** 2
+        dim_head = dim // heads
+
+        self.to_patch_embed = nn.Sequential(
+            Rearrange('b c (h p1) (w p2) -> b (h w) (p1 p2 c)', p1 = patch_size, p2 = patch_size),
+            nn.Linear(patch_size * patch_size * 3, dim)
+        )
+
+        self.pos_embed = nn.Parameter(torch.randn(num_patches, dim))
+
+        self.layers = ModuleList([])
+        for _ in range(depth):
+            self.layers.append(ModuleList([
+                Attention(dim, heads = heads, dim_head = dim_head, norm_eps = norm_eps),
+                FeedForward(dim = dim, dim_inner = mlp_dim, norm_eps = norm_eps)
+            ]))
+
+        self.norm = nn.LayerNorm(dim, eps = norm_eps)
+
+    def forward(self, x, return_hiddens = False):
+        x = self.to_patch_embed(x)
+        num_patches = x.shape[1]
+
+        x = x + self.pos_embed[:num_patches]
+
+        hiddens = []
+
+        for attn, ff in self.layers:
+            hiddens.append(x)
+
+            x = attn(x) + x
+            x = ff(x) + x
+
+        out = self.norm(x)
+
+        if not return_hiddens:
+            return out
+
+        return out, stack(hiddens)
+
+class FiLM(Module):
+    def __init__(self, dim):
+        super().__init__()
+        proj = nn.Linear(dim, dim * 2)
+
+        self.to_gamma_beta = nn.Sequential(
+            proj,
+            Rearrange('b (two d) -> two b 1 d', two = 2)
+        )
+
+        nn.init.zeros_(proj.weight)
+        nn.init.zeros_(proj.bias)
+
+    def forward(self, tokens, cond):
+        gamma, beta = self.to_gamma_beta(cond)
+        return tokens * gamma + beta
+
+class SigLIPVAT(Module):
+    def __init__(
+        self,
+        *,
+        dim = 512,
+        depth = 27,
+        heads = 8,
+        dim_head = 64,
+        dim_action = 32,
+        mlp_dim = 2048,
+        num_views = 1,
+        num_tasks = None,
+        dim_extra_token = None,
+        num_register_tokens = 4,
+        action_chunk_len = 50,
+        time_seq_len = 1,
+        dropout = 0.,
+        add_self_attn = True,
+        self_attn_heads = 4,
+        self_attn_dim_head = 32,
+        vit_layer_indices: tuple[int, ...] | None = None,
+        num_advantage_bins = 0,
+        siglip_image_size = 224,
+        siglip_patch_size = 14,
+        siglip_dim = 1152,
+        siglip_depth = 27,
+        siglip_heads = 16,
+        siglip_mlp_dim = 4304,
+        siglip_norm_eps = 1e-6,
+    ):
+        super().__init__()
+
+        self.vit = SigLIP(
+            image_size = siglip_image_size,
+            patch_size = siglip_patch_size,
+            dim = siglip_dim,
+            depth = siglip_depth,
+            heads = siglip_heads,
+            mlp_dim = siglip_mlp_dim,
+            norm_eps = siglip_norm_eps
+        )
+
+        vit_dim = siglip_dim
+        self.vit_dim = vit_dim
+
+        vit_layer_indices = default(vit_layer_indices, tuple(range(depth)))
+        assert len(vit_layer_indices) == depth, f'number of vit layer indices {len(vit_layer_indices)} does not much the VAT depth {depth}'
+        self.register_buffer('layer_indices', tensor(vit_layer_indices), persistent = False)
+
+        # handle maybe multiple frames
+
+        is_video = time_seq_len > 1
+        self.is_video = is_video
+        self.time_seq_len = time_seq_len
+        self.time_pos_emb = nn.Parameter(torch.randn(time_seq_len, vit_dim) * 1e-2) if is_video else None
+
+        # maybe view embeddings
+
+        self.view_emb = nn.Parameter(torch.randn(num_views, vit_dim) * 1e-2) if exists(num_views) and num_views > 1 else None
+
+        # handle maybe task conditioning
+
+        self.has_tasks = exists(num_tasks)
+        if self.has_tasks:
+            self.task_emb = nn.Parameter(torch.randn(num_tasks, dim) * 1e-2)
+
+        # register tokens
+
+        self.register_tokens = nn.Parameter(torch.randn(num_register_tokens, dim) * 1e-2)
+
+        # to action tokens
+
+        self.action_pos_emb = nn.Parameter(torch.randn(action_chunk_len, dim) * 1e-2)
+
+        # handle maybe advantage conditioning
+
+        self.has_advantages = num_advantage_bins > 0
+        self.num_advantage_bins = num_advantage_bins
+
+        if self.has_advantages:
+            self.advantage_emb = nn.Embedding(num_advantage_bins + 1, dim)
+
+        self.layers = ModuleList([])
+        for _ in range(depth):
+            maybe_film = FiLM(dim = dim) if self.has_tasks else None
+            maybe_self_attn = Attention(dim = dim, heads = self_attn_heads, dim_head = self_attn_dim_head, dropout = dropout) if add_self_attn else None
+
+            self.layers.append(ModuleList([
+                maybe_film,
+                maybe_self_attn,
+                Attention(dim = dim, dim_context = vit_dim, heads = heads, dim_head = dim_head, dropout = dropout, gate_attn = True),
+                FeedForward(dim = dim, dim_inner = mlp_dim)
+            ]))
+
+        self.final_norm = nn.LayerNorm(dim)
+        self.to_pred_action = nn.Linear(dim, dim_action, bias = False)
+
+        # handle the extra token
+
+        self.accept_extra_token = exists(dim_extra_token)
+        if exists(dim_extra_token):
+            self.to_extra_token = nn.Linear(dim_extra_token, dim)
+
+    def load_siglip(
+        self,
+        repo_id = 'google/siglip-so400m-patch14-224',
+        folder = 'checkpoints/siglip'
+    ):
+        folder = Path(folder)
+        if not folder.exists():
+            from huggingface_hub import snapshot_download
+            snapshot_download(
+                repo_id = repo_id,
+                local_dir = folder,
+                allow_patterns = ['config.json', 'model.safetensors']
+            )
+
+        from safetensors import safe_open
+        weights_path = folder / 'model.safetensors'
+
+        # Auto-detect prefix based on keys
+        with safe_open(weights_path, framework = 'pt') as f:
+            keys = f.keys()
+
+            vi_p = ''
+            if any(k.startswith('paligemma_with_expert.paligemma.model.vision_tower.vision_model') for k in keys):
+                vi_p = 'paligemma_with_expert.paligemma.model.vision_tower.vision_model.'
+            elif any(k.startswith('vision_model') for k in keys):
+                vi_p = 'vision_model.'
+
+            pz_state = self.vit.state_dict()
+
+            def copy_weight_bias(pz_prefix, vi_prefix):
+                pz_state[f'{pz_prefix}.weight'].copy_(f.get_tensor(f'{vi_prefix}.weight'))
+                pz_state[f'{pz_prefix}.bias'].copy_(f.get_tensor(f'{vi_prefix}.bias'))
+
+            # patch embedding
+            patch_weight = rearrange(f.get_tensor(f'{vi_p}embeddings.patch_embedding.weight'), 'd c h w -> d (h w c)')
+            pz_state['to_patch_embed.1.weight'].copy_(patch_weight)
+            pz_state['to_patch_embed.1.bias'].copy_(f.get_tensor(f'{vi_p}embeddings.patch_embedding.bias'))
+
+            # position embedding
+            pz_state['pos_embed'].copy_(f.get_tensor(f'{vi_p}embeddings.position_embedding.weight'))
+
+            # transformer layers
+            for i in range(self.vit.depth):
+                v_pi = f'{vi_p}encoder.layers.{i}'
+                v_pz = f'layers.{i}'
+
+                # attention
+                copy_weight_bias(f'{v_pz}.0.norm', f'{v_pi}.layer_norm1')
+                copy_weight_bias(f'{v_pz}.0.to_q', f'{v_pi}.self_attn.q_proj')
+
+                vk, vv = [f.get_tensor(f'{v_pi}.self_attn.{x}_proj.weight') for x in ('k', 'v')]
+                bk, bv = [f.get_tensor(f'{v_pi}.self_attn.{x}_proj.bias') for x in ('k', 'v')]
+
+                pz_state[f'{v_pz}.0.to_kv.weight'].copy_(cat((vk, vv), dim = 0))
+                pz_state[f'{v_pz}.0.to_kv.bias'].copy_(cat((bk, bv), dim = 0))
+
+                copy_weight_bias(f'{v_pz}.0.to_out.0', f'{v_pi}.self_attn.out_proj')
+
+                # feedforward
+                copy_weight_bias(f'{v_pz}.1.0', f'{v_pi}.layer_norm2')
+                copy_weight_bias(f'{v_pz}.1.1', f'{v_pi}.mlp.fc1')
+                copy_weight_bias(f'{v_pz}.1.3', f'{v_pi}.mlp.fc2')
+
+            # post-layernorm
+            copy_weight_bias('norm', f'{vi_p}post_layernorm')
+
+            self.vit.load_state_dict(pz_state)
+
+        print(f'Successfully loaded SigLIP weights from {repo_id}')
+
+    def forward(
+        self,
+        video_or_image,   # (b v? c t? h w) - batch, views [wrist + third person or more], channels, maybe time, height, width
+        *,
+        extra = None,     # (b d)           - batch, dim extra
+        tasks = None,     # (b)
+        advantages = None,# (b)
+        actions = None,   # (b k d)         - batch, action chunk length, action dimension
+        return_hiddens = False,
+        freeze_vit = False
+    ):
+        batch, device = video_or_image.shape[0], video_or_image.device
+        return_loss = exists(actions)
+
+        # handle some various input dimensions
+
+        if video_or_image.ndim == 4:
+            video_or_image = rearrange(video_or_image, 'b 1 c h w')
+
+        if video_or_image.ndim == 5:
+            video_or_image = rearrange(video_or_image, 'b v c h w -> b v c 1 h w')
+
+        assert video_or_image.shape[3] == self.time_seq_len
+
+        # to images
+
+        images = rearrange(video_or_image, 'b v c t h w -> b v t c h w')
+        images, packed_shape = pack([images], '* c h w')
+
+        # get representation trajectory from vit
+
+        vit_forward_context = torch.no_grad if freeze_vit else nullcontext
+
+        with vit_forward_context():
+            embed, hiddens = self.vit(images, return_hiddens = True)
+
+        hiddens = cat((hiddens, embed[None, ...]))
+
+        # extract the hiddens needed for the action cross attention
+
+        hiddens = hiddens[self.layer_indices]
+
+        # pack temporarily for embedding
+
+        hiddens, = unpack(hiddens, packed_shape, 'l * n d') # l for layers
+
+        # maybe add time embeddings
+
+        if self.is_video:
+            time_pos_emb = rearrange(self.time_pos_emb, 't d -> t 1 d')
+            hiddens = hiddens + time_pos_emb
+
+        # maybe view embeddings
+
+        if exists(self.view_emb):
+            view_emb = rearrange(self.view_emb, 'v d -> v 1 1 d')
+            hiddens = hiddens + view_emb
+
+        # maybe tasks
+
+        if exists(tasks):
+            task_emb = self.task_emb[tasks]
+
+        # cross from actions to representation trajectory
+
+        context = rearrange(hiddens, 'l b v t n d -> l b (v t n) d')
+
+        # main action tokens
+
+        action_tokens = repeat(self.action_pos_emb, 'n d -> b n d', b = batch)
+
+        # maybe advantage tokens
+
+        empty_token = action_tokens[:, 0:0]
+
+        maybe_advantage_embed = empty_token
+
+        if self.has_advantages and exists(advantages):
+            if isinstance(advantages, int):
+                advantages = torch.full((batch,), advantages, device = device, dtype = torch.long)
+
+            maybe_advantage_embed = self.advantage_emb(advantages + 1)
+
+        # register tokens
+
+        register_tokens = empty_token
+
+        if exists(self.register_tokens):
+            register_tokens = repeat(self.register_tokens, 'n d -> b n d', b = batch)
+
+        # extra
+
+        maybe_extra_embed = empty_token
+
+        has_extra = exists(extra)
+        if has_extra:
+            maybe_extra_embed = self.to_extra_token(extra)
+
+        # pack all tokens for attention
+
+        tokens, ps = pack((register_tokens, maybe_advantage_embed, action_tokens, maybe_extra_embed), 'b * d')
+
+        # transformer
+
+        vat_hiddens = [tokens]
+
+        for (maybe_film, maybe_self_attn, cross_attn, ff), layer_context in zip(self.layers, context):
+
+            if exists(maybe_film) and exists(tasks):
+                tokens = maybe_film(tokens, task_emb)
+
+            tokens = cross_attn(tokens, layer_context) + tokens
+
+            if exists(maybe_self_attn):
+                tokens = maybe_self_attn(tokens) + tokens
+
+            tokens = ff(tokens) + tokens
+
+            vat_hiddens.append(tokens)
+
+        # unpack register, advantage, action, and extra tokens
+
+        maybe_register_embed, maybe_advantage_embed, action_tokens, maybe_extra_embed = unpack(tokens, ps, 'b * d')
+
+        # norm and prediction
+
+        action_tokens = self.final_norm(action_tokens)
+        pred_action = self.to_pred_action(action_tokens)
+
+        if not return_loss:
+            if not return_hiddens:
+                return pred_action
+
+            return pred_action, stack(vat_hiddens)
+
+        assert pred_action.shape[1] == actions.shape[1]
+
+        return F.l1_loss(pred_action, actions)
+
+# quick test
+
+if __name__ == '__main__':
+    for num_adv_bins in (0, 2, 10):
+        vat = SigLIPVAT(
+            num_tasks = 4,
+            dim_extra_token = 32,
+            time_seq_len = 2,
+            num_views = 2,
+            depth = 4,
+            num_advantage_bins = num_adv_bins,
+            vit_layer_indices = (               # extending on the paper, allow for any order of hiddens, and also allow for depth index (which equates to the final embedding output from the vit)
+                0, 1, 26, 27
+            )
+        )
+
+        vat.load_siglip() # load siglip weights from hf
+
+        # inputs
+
+        images = torch.randn(1, 2, 3, 2, 224, 224) # (b, v, c, t, h, w)
+        tasks = torch.randint(0, 4, (1,))
+        extra = torch.randn(1, 32)
+
+        # advantage conditioning
+
+        advantages = None
+        if num_adv_bins > 0:
+            advantages = torch.randint(-1, num_adv_bins, (1,))
+
+        actions = torch.randn(1, 50, 32) # actions for learning
+
+        loss = vat(images, actions = actions, advantages = advantages, tasks = tasks, extra = extra, freeze_vit = True)
+        loss.backward()
+
+        # after much training
+
+        pred_actions = vat(images, advantages = advantages, tasks = tasks, extra = extra)
+
+        assert pred_actions.shape == (1, 50, 32)

vit_pytorch/vit.py

@@ -86,7 +86,7 @@ class ViT(Module):
     def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, pool = 'cls', channels = 3, dim_head = 64, dropout = 0., emb_dropout = 0.):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
 
@@ -113,7 +113,7 @@ class ViT(Module):
         self.pool = pool
         self.to_latent = nn.Identity()
 
-        self.mlp_head = nn.Linear(dim, num_classes)
+        self.mlp_head = nn.Linear(dim, num_classes) if num_classes > 0 else None
 
     def forward(self, img):
         batch = img.shape[0]
@@ -129,6 +129,9 @@ class ViT(Module):
 
         x = self.transformer(x)
 
+        if self.mlp_head is None:
+            return x
+
         x = x.mean(dim = 1) if self.pool == 'mean' else x[:, 0]
 
         x = self.to_latent(x)

vit_pytorch/vit_3d.py

@@ -78,7 +78,7 @@ class ViT(nn.Module):
     def __init__(self, *, image_size, image_patch_size, frames, frame_patch_size, num_classes, dim, depth, heads, mlp_dim, pool = 'cls', channels = 3, dim_head = 64, dropout = 0., emb_dropout = 0.):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(image_patch_size)
+        self.patch_size = patch_height, patch_width = pair(image_patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
         assert frames % frame_patch_size == 0, 'Frames must be divisible by frame patch size'

vit_pytorch/vit_detpool.py

@@ -0,0 +1,254 @@
+from __future__ import annotations
+
+# DetPool ViT - a vit that accepts an object mask and attends and pools only using that mask - table 1
+# Dantong Niu et al. - https://openreview.net/forum?id=NZDaMcpXZm
+
+import torch
+import torch.nn.functional as F
+from torch import nn
+from torch.nn import Module, ModuleList
+
+from einops import rearrange, repeat, pack, unpack
+from einops.layers.torch import Rearrange, Reduce
+
+# helpers
+
+def exists(val):
+    return val is not None
+
+def pair(t):
+    return t if isinstance(t, tuple) else (t, t)
+
+def masked_mean(t, mask, dim = 1, eps = 1e-5):
+    if not exists(mask):
+        return t.mean(dim = dim)
+
+    mask = rearrange(mask.bool(), '... -> ... 1')
+    t = t.masked_fill(~mask, 0.)
+    return t.sum(dim = dim) / mask.sum(dim = dim).clamp(min = eps)
+
+# classes
+
+class FeedForward(Module):
+    def __init__(self, dim, hidden_dim, dropout = 0.):
+        super().__init__()
+        self.net = nn.Sequential(
+            nn.LayerNorm(dim),
+            nn.Linear(dim, hidden_dim),
+            nn.GELU(),
+            nn.Dropout(dropout),
+            nn.Linear(hidden_dim, dim),
+            nn.Dropout(dropout)
+        )
+
+    def forward(self, x):
+        return self.net(x)
+
+class Attention(Module):
+    def __init__(self, dim, heads = 8, dim_head = 64, dropout = 0.):
+        super().__init__()
+        inner_dim = dim_head *  heads
+        project_out = not (heads == 1 and dim_head == dim)
+
+        self.heads = heads
+        self.scale = dim_head ** -0.5
+
+        self.norm = nn.LayerNorm(dim)
+
+        self.attend = nn.Softmax(dim = -1)
+        self.dropout = nn.Dropout(dropout)
+
+        self.to_qkv = nn.Linear(dim, inner_dim * 3, bias = False)
+
+        self.to_out = nn.Sequential(
+            nn.Linear(inner_dim, dim),
+            nn.Dropout(dropout)
+        ) if project_out else nn.Identity()
+
+    def forward(self, x, mask = None):
+        x = self.norm(x)
+
+        qkv = self.to_qkv(x).chunk(3, dim = -1)
+        q, k, v = (rearrange(t, 'b n (h d) -> b h n d', h = self.heads) for t in qkv)
+
+        dots = torch.matmul(q, k.transpose(-1, -2)) * self.scale
+
+        if exists(mask):
+            mask = rearrange(mask, 'b j -> b 1 1 j')
+            mask_value = -torch.finfo(dots.dtype).max
+            dots = dots.masked_fill(~mask, mask_value)
+
+        attn = self.attend(dots)
+        attn = self.dropout(attn)
+
+        out = torch.matmul(attn, v)
+        out = rearrange(out, 'b h n d -> b n (h d)')
+        return self.to_out(out)
+
+class Transformer(Module):
+    def __init__(self, dim, depth, heads, dim_head, mlp_dim, dropout = 0.):
+        super().__init__()
+        self.norm = nn.LayerNorm(dim)
+        self.layers = ModuleList([])
+
+        for _ in range(depth):
+            self.layers.append(ModuleList([
+                Attention(dim, heads = heads, dim_head = dim_head, dropout = dropout),
+                FeedForward(dim, mlp_dim, dropout = dropout)
+            ]))
+
+    def forward(self, x, mask = None):
+        for attn, ff in self.layers:
+            x = attn(x, mask = mask) + x
+            x = ff(x) + x
+
+        return self.norm(x)
+
+class ViTDetPool(Module):
+    def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, use_cls_token = True, channels = 3, dim_head = 64, dropout = 0., emb_dropout = 0., mask_generator: Module | None = None):
+        super().__init__()
+        image_height, image_width = pair(image_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
+
+        assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
+
+        num_patches = (image_height // patch_height) * (image_width // patch_width)
+        patch_dim = channels * patch_height * patch_width
+
+        self.patch_height = patch_height
+        self.patch_width = patch_width
+
+        self.downsample_mask = Reduce('b (h p1) (w p2) -> b (h w)', 'max', p1 = patch_height, p2 = patch_width)
+
+        self.to_patch_embedding = nn.Sequential(
+            Rearrange('b c (h p1) (w p2) -> b (h w) (p1 p2 c)', p1 = patch_height, p2 = patch_width),
+            nn.LayerNorm(patch_dim),
+            nn.Linear(patch_dim, dim),
+            nn.LayerNorm(dim),
+        )
+
+        # maybe cls
+
+        self.use_cls_token = use_cls_token
+
+        if use_cls_token:
+            self.cls_token = nn.Parameter(torch.randn(dim) * 1e-2)
+
+        self.pos_embedding = nn.Parameter(torch.randn(num_patches, dim) * 1e-2)
+
+        self.dropout = nn.Dropout(emb_dropout)
+
+        self.transformer = Transformer(dim, depth, heads, dim_head, mlp_dim, dropout)
+
+        self.to_latent = nn.Identity()
+
+        self.mlp_head = nn.Linear(dim, num_classes) if num_classes > 0 else None
+
+        self.mask_generator = mask_generator
+
+    def forward(self, img, object_mask = None):
+
+        if not exists(object_mask) and exists(self.mask_generator):
+            with torch.no_grad():
+                self.mask_generator.eval()
+                object_mask = self.mask_generator(img)
+
+        has_cls = self.use_cls_token
+
+        batch, _, height, width = img.shape
+        tokens = self.to_patch_embedding(img)
+
+        seq = tokens.shape[1]
+        tokens = tokens + self.pos_embedding[:seq]
+
+        if has_cls:
+            cls_token = repeat(self.cls_token, 'd -> b d', b = batch)
+            tokens, packed_shape = pack((cls_token, tokens), 'b * d')
+
+        tokens = self.dropout(tokens)
+
+        # handle the attention mask, and for final pooling
+
+        mask = None
+
+        if exists(object_mask):
+            assert object_mask.ndim in {3, 2}
+
+            if object_mask.shape == (batch, height, width):
+                mask = self.downsample_mask(object_mask)
+            else:
+                mask = object_mask
+
+            mask = rearrange(mask, 'b ... -> b (...)')
+
+            assert mask.shape == (batch, seq)
+
+            if has_cls:
+                mask = F.pad(mask, (1, 0), value = True)
+
+        # attend with maybe mask
+
+        tokens = self.transformer(tokens, mask = mask)
+
+        if not exists(self.mlp_head):
+            return tokens
+
+        # splice out cls
+
+        if has_cls:
+            _, tokens = unpack(tokens, packed_shape, 'b * d')
+
+            if exists(mask):
+                mask = mask[..., 1:]
+
+        # pooling with the mask
+
+        pooled = masked_mean(tokens, mask, dim = 1)
+
+        pooled = self.to_latent(pooled)
+        return self.mlp_head(pooled)
+
+# quick test
+
+if __name__ == '__main__':
+    vit = ViTDetPool(
+        image_size = 256,
+        patch_size = 32,
+        num_classes = 1000,
+        dim = 1024,
+        depth = 6,
+        heads = 16,
+        mlp_dim = 2048,
+        dropout = 0.1,
+        emb_dropout = 0.1
+    )
+
+    img = torch.randn(1, 3, 256, 256)
+    object_mask = torch.randint(0, 2, (1, 256, 256)).bool()
+
+    preds = vit(img, object_mask = object_mask)
+    assert preds.shape == (1, 1000)
+
+    preds_no_mask = vit(img)
+    assert preds_no_mask.shape == (1, 1000)
+
+    # test with module included
+
+    class MockMasker(Module):
+        def forward(self, img):
+            batch, _, height, width = img.shape
+            return torch.ones(batch, height, width).bool()
+
+    vit = ViTDetPool(
+        image_size = 256,
+        patch_size = 32,
+        num_classes = 1000,
+        dim = 1024,
+        depth = 1,
+        heads = 16,
+        mlp_dim = 2048,
+        mask_generator = MockMasker()
+    )
+
+    preds = vit(img)
+    assert preds.shape == (1, 1000)

vit_pytorch/vit_for_small_dataset.py

@@ -99,7 +99,7 @@ class ViT(nn.Module):
     def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, pool = 'cls', channels = 3, dim_head = 64, dropout = 0., emb_dropout = 0.):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
 

vit_pytorch/vit_nd.py

@@ -31,7 +31,7 @@ class FeedForward(Module):
             nn.Linear(hidden_dim, dim),
             nn.Dropout(dropout)
         )
-    
+
     def forward(self, x):
         return self.net(x)
 
@@ -40,31 +40,31 @@ class Attention(Module):
         super().__init__()
         inner_dim = dim_head * heads
         project_out = not (heads == 1 and dim_head == dim)
-        
+
         self.heads = heads
         self.scale = dim_head ** -0.5
-        
+
         self.norm = nn.LayerNorm(dim)
         self.attend = nn.Softmax(dim = -1)
         self.dropout = nn.Dropout(dropout)
-        
+
         self.to_qkv = nn.Linear(dim, inner_dim * 3, bias = False)
-        
+
         self.to_out = nn.Sequential(
             nn.Linear(inner_dim, dim),
             nn.Dropout(dropout)
         ) if project_out else nn.Identity()
-    
+
     def forward(self, x):
         x = self.norm(x)
         qkv = self.to_qkv(x).chunk(3, dim = -1)
         q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b h n d', h = self.heads), qkv)
-        
+
         dots = torch.matmul(q, k.transpose(-1, -2)) * self.scale
-        
+
         attn = self.attend(dots)
         attn = self.dropout(attn)
-        
+
         out = torch.matmul(attn, v)
         out = rearrange(out, 'b h n d -> b n (h d)')
         return self.to_out(out)
@@ -79,7 +79,7 @@ class Transformer(Module):
                 Attention(dim, heads = heads, dim_head = dim_head, dropout = dropout),
                 FeedForward(dim, mlp_dim, dropout = dropout)
             ]))
-    
+
     def forward(self, x):
         for attn, ff in self.layers:
             x = attn(x) + x
@@ -105,73 +105,73 @@ class ViTND(Module):
         emb_dropout: float = 0.
     ):
         super().__init__()
-        
+
         assert 1 <= ndim <= 7, 'ndim must be between 1 and 7'
         assert pool in {'cls', 'mean'}, 'pool type must be either cls (cls token) or mean (mean pooling)'
-        
+
         self.ndim = ndim
         self.pool = pool
-        
+
         input_shape = ensure_tuple(input_shape, ndim)
         patch_size = ensure_tuple(patch_size, ndim)
-        
+
         for i, (inp_dim, patch_dim) in enumerate(zip(input_shape, patch_size)):
             assert inp_dim % patch_dim == 0, f'Input dimension {i} ({inp_dim}) must be divisible by patch size ({patch_dim})'
-        
+
         num_patches_per_dim = [inp_dim // patch_dim for inp_dim, patch_dim in zip(input_shape, patch_size)]
         num_patches = 1
         for n in num_patches_per_dim:
             num_patches *= n
-        
+
         patch_dim = channels
         for p in patch_size:
             patch_dim *= p
-        
+
         dim_names = 'fghijkl'[:ndim]
-        
+
         input_dims = [f'({d} p{i})' for i, d in enumerate(dim_names)]
         patch_dims = [f'p{i}' for i in range(ndim)]
-        
+
         input_pattern = f'b c {join(input_dims)}'
         output_pattern = f'b ({join(dim_names)}) ({join(patch_dims)} c)'
         rearrange_str = f'{input_pattern} -> {output_pattern}'
-        
+
         rearrange_kwargs = {f'p{i}': p for i, p in enumerate(patch_size)}
-        
+
         self.to_patch_embedding = nn.Sequential(
             Rearrange(rearrange_str, **rearrange_kwargs),
             nn.Linear(patch_dim, dim),
             nn.LayerNorm(dim),
         )
-        
+
         self.pos_embedding = nn.Parameter(torch.randn(1, num_patches + 1, dim))
         self.cls_token = nn.Parameter(torch.randn(1, 1, dim))
         self.dropout = nn.Dropout(emb_dropout)
-        
+
         self.transformer = Transformer(dim, depth, heads, dim_head, mlp_dim, dropout)
-        
+
         self.to_latent = nn.Identity()
         self.mlp_head = nn.Linear(dim, num_classes)
-    
+
     def forward(self, x):
         x = self.to_patch_embedding(x)
         b, n, _ = x.shape
-        
+
         cls_tokens = repeat(self.cls_token, '1 1 d -> b 1 d', b = b)
         x = torch.cat((cls_tokens, x), dim = 1)
         x += self.pos_embedding[:, :(n + 1)]
         x = self.dropout(x)
-        
+
         x = self.transformer(x)
-        
+
         x = x[:, 1:].mean(dim = 1) if self.pool == 'mean' else x[:, 0]
-        
+
         x = self.to_latent(x)
         return self.mlp_head(x)
 
 
 if __name__ == '__main__':
-    
+
     model = ViTND(
         ndim = 4,
         input_shape = (8, 16, 32, 64),
@@ -185,7 +185,7 @@ if __name__ == '__main__':
         dropout = 0.1,
         emb_dropout = 0.1
     )
-    
+
     occupancy_time = torch.randn(2, 3, 8, 16, 32, 64)
-    
+
     logits = model(occupancy_time)

vit_pytorch/vit_nd_pope.py

@@ -0,0 +1,353 @@
+from __future__ import annotations
+
+import torch
+import torch.nn.functional as F
+from torch import pi, nn, arange, cat, stack, Tensor
+from torch.nn import Module, ModuleList
+from torch.amp import autocast
+
+from einops import rearrange, repeat, reduce, pack, unpack
+from einops.layers.torch import Rearrange
+
+# helpers
+
+def exists(val):
+    return val is not None
+
+def l2norm(t):
+    return F.normalize(t, dim = -1, p = 2)
+
+def join(arr, delimiter = ' '):
+    return delimiter.join(arr)
+
+def ensure_tuple(t, length):
+    if isinstance(t, (tuple, list)):
+        assert len(t) == length, f'Expected tuple of length {length}, got {len(t)}'
+        return tuple(t)
+
+    return (t,) * length
+
+# golden gate rotary - Jerry Xiong, PhD student at UIUC
+# https://jerryxio.ng/posts/nd-rope/
+
+# but using polar version instead
+# Gopalakrishnan et al. https://arxiv.org/abs/2509.10534
+
+def _phi(m: int) -> float:
+    x = 2.0
+    for _ in range(10):
+        x = (1 + x) ** (1.0 / (m + 1.0))
+    return x
+
+def make_directions(n: int, d: int) -> Tensor:
+    g = _phi(d)
+    alpha = (1.0 / g) ** arange(1, d + 1, dtype = torch.float64)
+    i = arange(1, n + 1, dtype = torch.float64).unsqueeze(1)
+    z = torch.fmod(i * alpha, 1.0)
+    directions = torch.erfinv(2.0 * z - 1.0)
+    directions = l2norm(directions)
+    return directions.float()
+
+class GoldenGatePoPENd(Module):
+    def __init__(
+        self,
+        dim_pos: int,
+        heads: int,
+        dim_head: int,
+        min_freq: float = 1.0,
+        max_freq: float = 10000.0,
+        p_zero_freqs: float = 0.0, # proportion of frequencies set to 0
+        init_learned_bias_uniform = False
+    ):
+        super().__init__()
+        n_freqs = dim_head
+        n_zero_freqs = round(p_zero_freqs * n_freqs)
+
+        omega = cat((
+            torch.zeros(n_zero_freqs),
+            min_freq * (max_freq / min_freq) ** torch.linspace(0, 1, n_freqs - n_zero_freqs),
+        ))
+
+        directions = rearrange(
+            make_directions(heads * n_freqs, dim_pos),
+            '(h f) p -> h f p',
+            h = heads
+        )
+
+        omega_expanded = rearrange(omega, 'f -> f 1')
+        self.register_buffer('freqs', directions * omega_expanded)  # shape: (h, f, p)
+
+        self.learned_bias = nn.Parameter(torch.zeros(heads, dim_head))
+
+        if init_learned_bias_uniform:
+            self.learned_bias.uniform_(-2. * pi, 0.)
+
+    @autocast('cuda', enabled = False)
+    def forward(self, pos):
+
+        freqs = rearrange(self.freqs, 'h f p -> 1 h 1 f p')
+        positions = rearrange(pos.float(), 'b n p -> b 1 n 1 p')
+
+        # compute theta for each (batch, head, seq, freq)
+
+        theta = reduce(freqs * positions, 'b h n f p -> b h n f', 'sum')
+
+        bias = self.learned_bias.clamp(-2. * pi, 0.)
+        bias = rearrange(bias, 'h d -> h 1 d')
+
+        return theta, bias
+
+@autocast('cuda', enabled = False)
+def apply_polar_pos_emb(t, freqs):
+    orig_dtype = t.dtype
+
+    t = t.float()
+    t = F.softplus(t)
+
+    out = cat((t * freqs.cos(), t * freqs.sin()), dim = -1)
+
+    return out.type(orig_dtype)
+
+# classes
+
+class FeedForward(Module):
+    def __init__(self, dim, hidden_dim, dropout = 0.):
+        super().__init__()
+        self.net = nn.Sequential(
+            nn.LayerNorm(dim),
+            nn.Linear(dim, hidden_dim),
+            nn.GELU(),
+            nn.Dropout(dropout),
+            nn.Linear(hidden_dim, dim),
+            nn.Dropout(dropout)
+        )
+
+    def forward(self, x):
+        return self.net(x)
+
+class Attention(Module):
+    def __init__(self, dim, heads = 8, dim_head = 64, dropout = 0.):
+        super().__init__()
+        inner_dim = dim_head * heads
+        project_out = not (heads == 1 and dim_head == dim)
+
+        self.heads = heads
+        self.scale = dim_head ** -0.5
+
+        self.norm = nn.LayerNorm(dim)
+        self.attend = nn.Softmax(dim = -1)
+        self.dropout = nn.Dropout(dropout)
+
+        self.to_qk = nn.Linear(dim, inner_dim * 2, bias = False)
+        self.to_v = nn.Linear(dim, inner_dim, bias = False)
+
+        self.to_out = nn.Sequential(
+            nn.Linear(inner_dim, dim),
+            nn.Dropout(dropout)
+        ) if project_out else nn.Identity()
+
+    def forward(self, x, polar_pos_emb = None):
+        x = self.norm(x)
+        qkv = (*self.to_qk(x).chunk(2, dim = -1), self.to_v(x))
+
+        q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b h n d', h = self.heads), qkv)
+
+        if exists(polar_pos_emb):
+            freqs, bias = polar_pos_emb
+            q = apply_polar_pos_emb(q, freqs)
+            k = apply_polar_pos_emb(k, freqs + bias)
+
+        dots = torch.matmul(q, k.transpose(-1, -2)) * self.scale
+
+        attn = self.attend(dots)
+        attn = self.dropout(attn)
+
+        out = torch.matmul(attn, v)
+        out = rearrange(out, 'b h n d -> b n (h d)')
+        return self.to_out(out)
+
+class Transformer(Module):
+    def __init__(self, dim, depth, heads, dim_head, mlp_dim, dropout = 0., polar_emb = None):
+        super().__init__()
+        self.norm = nn.LayerNorm(dim)
+
+        self.polar_emb = polar_emb
+
+        self.layers = ModuleList([])
+
+        for _ in range(depth):
+            self.layers.append(ModuleList([
+                Attention(dim, heads = heads, dim_head = dim_head, dropout = dropout),
+                FeedForward(dim, mlp_dim, dropout = dropout)
+            ]))
+
+    def forward(self, x, pos = None):
+
+        # pope embedding
+
+        polar_pos_emb = None
+        if exists(pos) and exists(self.polar_emb):
+            polar_pos_emb = self.polar_emb(pos)
+
+        # transformer layers
+
+        for attn, ff in self.layers:
+            x = attn(x, polar_pos_emb) + x
+            x = ff(x) + x
+
+        return self.norm(x)
+
+class ViTND(Module):
+    def __init__(
+        self,
+        *,
+        ndim: int,
+        input_shape: int | tuple[int, ...],
+        patch_size: int | tuple[int, ...],
+        num_classes: int,
+        dim: int,
+        depth: int,
+        heads: int,
+        mlp_dim: int,
+        channels: int = 3,
+        dim_head: int = 64,
+        dropout: float = 0.,
+        emb_dropout: float = 0.,
+        pope_min_freq: float = 1.0,
+        pope_max_freq: float = 10000.0,
+        pope_p_zero_freqs: float = 0.0,
+        pope_init_learned_bias_uniform = False
+    ):
+        super().__init__()
+
+        assert 1 <= ndim <= 7, 'ndim must be between 1 and 7'
+
+        self.ndim = ndim
+
+        input_shape = ensure_tuple(input_shape, ndim)
+        patch_size = ensure_tuple(patch_size, ndim)
+
+        for i, (inp_dim, patch_dim) in enumerate(zip(input_shape, patch_size)):
+            assert inp_dim % patch_dim == 0, f'Input dimension {i} ({inp_dim}) must be divisible by patch size ({patch_dim})'
+
+        num_patches_per_dim = [inp_dim // patch_dim for inp_dim, patch_dim in zip(input_shape, patch_size)]
+        num_patches = 1
+        for n in num_patches_per_dim:
+            num_patches *= n
+
+        patch_dim = channels
+        for p in patch_size:
+            patch_dim *= p
+
+        dim_names = 'fghijkl'[:ndim]
+
+        input_dims = [f'({d} p{i})' for i, d in enumerate(dim_names)]
+        patch_dims = [f'p{i}' for i in range(ndim)]
+
+        input_pattern = f'b c {join(input_dims)}'
+        output_pattern = f'b {join(dim_names)} ({join(patch_dims)} c)'
+        rearrange_str = f'{input_pattern} -> {output_pattern}'
+
+        rearrange_kwargs = {f'p{i}': p for i, p in enumerate(patch_size)}
+
+        self.to_patch_embedding = nn.Sequential(
+            Rearrange(rearrange_str, **rearrange_kwargs),
+            nn.Linear(patch_dim, dim),
+            nn.LayerNorm(dim),
+        )
+
+        self.dropout = nn.Dropout(emb_dropout)
+
+        # golden gate pope
+
+        self.polar_emb = GoldenGatePoPENd(
+            dim_pos = ndim,
+            heads = heads,
+            dim_head = dim_head,
+            min_freq = pope_min_freq,
+            max_freq = pope_max_freq,
+            p_zero_freqs = pope_p_zero_freqs,
+            init_learned_bias_uniform = pope_init_learned_bias_uniform
+        )
+
+        self.transformer = Transformer(dim, depth, heads, dim_head, mlp_dim, dropout, polar_emb = self.polar_emb)
+
+        self.to_latent = nn.Identity()
+        self.mlp_head = nn.Linear(dim, num_classes)
+
+    def muon_parameters(self):
+        params = []
+
+        for m in self.modules():
+            if isinstance(m, Attention):
+                params.extend([
+                    m.to_v.weight,
+                    m.to_out[0].weight
+                ])
+            elif isinstance(m, FeedForward):
+                params.extend([
+                    m.net[1].weight,
+                    m.net[-2].weight
+                ])
+
+        return params
+
+    def forward(
+        self,
+        x,
+        return_embed = False
+    ):
+        x = self.to_patch_embedding(x) # (b, *spatial_dims, patch_dim)
+
+        batch, *spatial_dims, _, device = *x.shape, x.device
+
+        # Generate position coordinates
+
+        grids = [arange(d, device = device, dtype = torch.float32) for d in spatial_dims]
+        grid = torch.meshgrid(*grids, indexing = 'ij')
+        pos = stack(grid, dim = -1)  # (*spatial_dims, ndim)
+
+        # flatten spatial dimensions for attention with nd rotary
+
+        pos = repeat(pos, '... p -> b (...) p', b = batch)
+        x, packed_shape = pack([x], 'b * d')
+
+        x = self.dropout(x)
+
+        embed = self.transformer(x, pos)
+
+        # return the embed with reconstituted patch shape
+
+        if return_embed:
+            embed, = unpack(embed, packed_shape, 'b * d')
+            return embed
+
+        # pooling to logits
+
+        pooled = reduce(embed, 'b n d -> b d', 'mean')
+
+        pooled = self.to_latent(pooled)
+        return self.mlp_head(pooled)
+
+if __name__ == '__main__':
+
+    model = ViTND(
+        ndim = 5,
+        input_shape = (4, 8, 16, 32, 64),
+        patch_size = (2, 2, 4, 4, 8),
+        num_classes = 1000,
+        dim = 512,
+        depth = 6,
+        heads = 8,
+        mlp_dim = 2048,
+        channels = 3,
+        dropout = 0.1,
+        emb_dropout = 0.1
+    )
+
+    data = torch.randn(3, 3, 4, 8, 16, 32, 64)
+
+    logits = model(data)
+
+    embed = model(data, return_embed = True)
+    assert embed.shape == (3, 2, 4, 4, 8, 8, 512)

vit_pytorch/vit_nd_rotary.py

@@ -75,23 +75,23 @@ class GoldenGateRoPENd(Module):
         # input shape: (b, h, n, d) where d = head_dim
         # pos shape: (b, n, p) where p = pos_dim
         # self.freqs shape: (h, f, p) where f = d // 2
-        
+
         x, y = input.float().chunk(2, dim = -1)  # both (b, h, n, f)
-        
+
         # Expand dimensions for broadcasting
         freqs = rearrange(self.freqs, 'h f p -> 1 h 1 f p')
         positions = rearrange(pos.float(), 'b n p -> b 1 n 1 p')
-        
+
         # Compute theta for each (batch, head, seq, freq)
         theta = reduce(freqs * positions, 'b h n f p -> b h n f', 'sum')
-        
+
         cos_theta = torch.cos(theta)
         sin_theta = torch.sin(theta)
-        
+
         # Apply rotation
         x_out = x * cos_theta - y * sin_theta
         y_out = x * sin_theta + y * cos_theta
-        
+
         output = cat((x_out, y_out), dim=-1)
         return output.type_as(input)
 
@@ -108,7 +108,7 @@ class FeedForward(Module):
             nn.Linear(hidden_dim, dim),
             nn.Dropout(dropout)
         )
-    
+
     def forward(self, x):
         return self.net(x)
 
@@ -117,15 +117,15 @@ class Attention(Module):
         super().__init__()
         inner_dim = dim_head * heads
         project_out = not (heads == 1 and dim_head == dim)
-        
+
         self.heads = heads
         self.scale = dim_head ** -0.5
         self.rotary_emb = rotary_emb
-        
+
         self.norm = nn.LayerNorm(dim)
         self.attend = nn.Softmax(dim = -1)
         self.dropout = nn.Dropout(dropout)
-        
+
         self.to_qk = nn.Linear(dim, inner_dim * 2, bias = False)
         self.to_v = nn.Linear(dim, inner_dim, bias = False)
 
@@ -133,24 +133,24 @@ class Attention(Module):
             nn.Linear(inner_dim, dim),
             nn.Dropout(dropout)
         ) if project_out else nn.Identity()
-    
+
     def forward(self, x, pos = None):
         x = self.norm(x)
         qkv = (*self.to_qk(x).chunk(2, dim = -1), self.to_v(x))
 
         q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b h n d', h = self.heads), qkv)
-        
+
         # Apply rotary embeddings if available
         if exists(self.rotary_emb):
             assert exists(pos)
             q = self.rotary_emb(q, pos)
             k = self.rotary_emb(k, pos)
-        
+
         dots = torch.matmul(q, k.transpose(-1, -2)) * self.scale
-        
+
         attn = self.attend(dots)
         attn = self.dropout(attn)
-        
+
         out = torch.matmul(attn, v)
         out = rearrange(out, 'b h n d -> b n (h d)')
         return self.to_out(out)
@@ -165,7 +165,7 @@ class Transformer(Module):
                 Attention(dim, heads = heads, dim_head = dim_head, dropout = dropout, rotary_emb = rotary_emb),
                 FeedForward(dim, mlp_dim, dropout = dropout)
             ]))
-    
+
     def forward(self, x, pos = None):
         for attn, ff in self.layers:
             x = attn(x, pos) + x
@@ -193,45 +193,45 @@ class ViTND(Module):
         rope_p_zero_freqs: float = 0.0
     ):
         super().__init__()
-        
+
         assert 1 <= ndim <= 7, 'ndim must be between 1 and 7'
-        
+
         self.ndim = ndim
-        
+
         input_shape = ensure_tuple(input_shape, ndim)
         patch_size = ensure_tuple(patch_size, ndim)
-        
+
         for i, (inp_dim, patch_dim) in enumerate(zip(input_shape, patch_size)):
             assert inp_dim % patch_dim == 0, f'Input dimension {i} ({inp_dim}) must be divisible by patch size ({patch_dim})'
-        
+
         num_patches_per_dim = [inp_dim // patch_dim for inp_dim, patch_dim in zip(input_shape, patch_size)]
         num_patches = 1
         for n in num_patches_per_dim:
             num_patches *= n
-        
+
         patch_dim = channels
         for p in patch_size:
             patch_dim *= p
-        
+
         dim_names = 'fghijkl'[:ndim]
-        
+
         input_dims = [f'({d} p{i})' for i, d in enumerate(dim_names)]
         patch_dims = [f'p{i}' for i in range(ndim)]
-        
+
         input_pattern = f'b c {join(input_dims)}'
         output_pattern = f'b {join(dim_names)} ({join(patch_dims)} c)'
         rearrange_str = f'{input_pattern} -> {output_pattern}'
-        
+
         rearrange_kwargs = {f'p{i}': p for i, p in enumerate(patch_size)}
-        
+
         self.to_patch_embedding = nn.Sequential(
             Rearrange(rearrange_str, **rearrange_kwargs),
             nn.Linear(patch_dim, dim),
             nn.LayerNorm(dim),
         )
-        
+
         self.dropout = nn.Dropout(emb_dropout)
-        
+
         # Create rotary embeddings
         self.rotary_emb = GoldenGateRoPENd(
             dim_pos = ndim,
@@ -241,12 +241,12 @@ class ViTND(Module):
             rope_max_freq = rope_max_freq,
             rope_p_zero_freqs = rope_p_zero_freqs
         )
-        
+
         self.transformer = Transformer(dim, depth, heads, dim_head, mlp_dim, dropout, rotary_emb = self.rotary_emb)
-        
+
         self.to_latent = nn.Identity()
         self.mlp_head = nn.Linear(dim, num_classes)
-    
+
     def muon_parameters(self):
         params = []
 
@@ -270,9 +270,9 @@ class ViTND(Module):
         return_embed = False
     ):
         x = self.to_patch_embedding(x) # (b, *spatial_dims, patch_dim)
-        
+
         batch, *spatial_dims, _, device = *x.shape, x.device
-        
+
         # Generate position coordinates
 
         grids = [arange(d, device = device, dtype = torch.float32) for d in spatial_dims]
@@ -280,12 +280,12 @@ class ViTND(Module):
         pos = stack(grid, dim = -1)  # (*spatial_dims, ndim)
 
         # flatten spatial dimensions for attention with nd rotary
-        
+
         pos = repeat(pos, '... p -> b (...) p', b = batch)
         x, packed_shape = pack([x], 'b * d')
 
         x = self.dropout(x)
-        
+
         embed = self.transformer(x, pos)
 
         # return the embed with reconstituted patch shape
@@ -303,7 +303,7 @@ class ViTND(Module):
 
 
 if __name__ == '__main__':
-  
+
     model = ViTND(
         ndim = 5,
         input_shape = (4, 8, 16, 32, 64),

vit_pytorch/vit_with_decorr.py

@@ -155,7 +155,7 @@ class ViT(Module):
     def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, pool = 'cls', channels = 3, dim_head = 64, dropout = 0., emb_dropout = 0., decorr_sample_frac = 1.):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
 
@@ -231,4 +231,4 @@ if __name__ == '__main__':
 
     decorr_loss = DecorrelationLoss(0.0001, soft_validate_num_sampled = True)
     out = decorr_loss(hiddens)
-    assert out.item() == 0
+    assert out.item() == 0

vit_pytorch/vit_with_keel_post_ln.py

@@ -0,0 +1,217 @@
+from __future__ import annotations
+
+import torch
+from torch import nn
+from torch.nn import Module, ModuleList
+
+from einops import rearrange, repeat
+from einops.layers.torch import Rearrange
+
+# functions
+
+def exists(v):
+    return v is not None
+
+def default(v, d):
+    return v if exists(v) else d
+
+def pair(t):
+    return t if isinstance(t, tuple) else (t, t)
+
+# classes
+
+class FeedForward(Module):
+    def __init__(self, dim, hidden_dim, dropout = 0.):
+        super().__init__()
+        self.net = nn.Sequential(
+            nn.LayerNorm(dim, bias = False),
+            nn.Linear(dim, hidden_dim),
+            nn.GELU(),
+            nn.Dropout(dropout),
+            nn.Linear(hidden_dim, dim),
+            nn.Dropout(dropout)
+        )
+
+    def forward(self, x):
+        return self.net(x)
+
+class Attention(Module):
+    def __init__(self, dim, heads = 8, dim_head = 64, dropout = 0.):
+        super().__init__()
+        inner_dim = dim_head *  heads
+        project_out = not (heads == 1 and dim_head == dim)
+
+        self.norm = nn.LayerNorm(dim, bias = False)
+
+        self.heads = heads
+        self.scale = dim_head ** -0.5
+
+        self.attend = nn.Softmax(dim = -1)
+        self.dropout = nn.Dropout(dropout)
+
+        self.to_qkv = nn.Linear(dim, inner_dim * 3, bias = False)
+
+        self.to_out = nn.Sequential(
+            nn.Linear(inner_dim, dim),
+            nn.Dropout(dropout)
+        ) if project_out else nn.Identity()
+
+    def forward(self, x):
+        x = self.norm(x)
+
+        qkv = self.to_qkv(x).chunk(3, dim = -1)
+        q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b h n d', h = self.heads), qkv)
+
+        dots = torch.matmul(q, k.transpose(-1, -2)) * self.scale
+
+        attn = self.attend(dots)
+        attn = self.dropout(attn)
+
+        out = torch.matmul(attn, v)
+        out = rearrange(out, 'b h n d -> b n (h d)')
+        return self.to_out(out)
+
+class Transformer(Module):
+    def __init__(
+        self,
+        dim,
+        depth,
+        heads,
+        dim_head,
+        mlp_dim,
+        dropout = 0.,
+        keel_residual_scale = None
+    ):
+        super().__init__()
+        assert depth > 1
+
+        self.layers = ModuleList([])
+
+        for _ in range(depth):
+            self.layers.extend([
+                Attention(dim, heads = heads, dim_head = dim_head, dropout = dropout),
+                FeedForward(dim, mlp_dim, dropout = dropout)
+            ])
+
+        num_layers = depth * 2
+        self.keel_residual_scale = default(keel_residual_scale, num_layers)
+
+        self.post_norms = ModuleList([nn.LayerNorm(dim, bias = False) for _ in range(num_layers - 1)])
+
+    def forward(self, x):
+        residual_scale = self.keel_residual_scale
+
+        for layer_ind, layer in enumerate(self.layers):
+            first_layer = layer_ind == 0
+
+            residual = x
+
+            out = layer(x)
+
+            if first_layer:
+                x = out + residual
+                continue
+
+            post_norm = self.post_norms[layer_ind - 1]
+
+            x = post_norm(out + residual * residual_scale)
+
+        return x
+
+class ViT(Module):
+    def __init__(
+        self,
+        *,
+        image_size,
+        patch_size,
+        num_classes,
+        dim,
+        depth,
+        heads,
+        mlp_dim,
+        pool = 'cls',
+        channels = 3,
+        dim_head = 64,
+        dropout = 0.,
+        emb_dropout = 0.,
+        keel_residual_scale = None
+    ):
+        super().__init__()
+        image_height, image_width = pair(image_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
+
+        assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
+
+        num_patches = (image_height // patch_height) * (image_width // patch_width)
+        patch_dim = channels * patch_height * patch_width
+
+        assert pool in {'cls', 'mean'}, 'pool type must be either cls (cls token) or mean (mean pooling)'
+        num_cls_tokens = 1 if pool == 'cls' else 0
+
+        self.to_patch_embedding = nn.Sequential(
+            Rearrange('b c (h p1) (w p2) -> b (h w) (p1 p2 c)', p1 = patch_height, p2 = patch_width),
+            nn.LayerNorm(patch_dim),
+            nn.Linear(patch_dim, dim),
+            nn.LayerNorm(dim),
+        )
+
+        self.cls_token = nn.Parameter(torch.randn(num_cls_tokens, dim))
+        self.pos_embedding = nn.Parameter(torch.randn(num_patches + num_cls_tokens, dim))
+
+        self.dropout = nn.Dropout(emb_dropout)
+
+        self.transformer = Transformer(
+            dim,
+            depth,
+            heads,
+            dim_head,
+            mlp_dim,
+            dropout,
+            keel_residual_scale = keel_residual_scale
+        )
+
+        self.pool = pool
+        self.to_latent = nn.Identity()
+
+        self.mlp_head = nn.Linear(dim, num_classes) if num_classes > 0 else None
+
+    def forward(self, img):
+        batch = img.shape[0]
+        x = self.to_patch_embedding(img)
+
+        cls_tokens = repeat(self.cls_token, '... d -> b ... d', b = batch)
+        x = torch.cat((cls_tokens, x), dim = 1)
+
+        seq = x.shape[1]
+
+        x = x + self.pos_embedding[:seq]
+        x = self.dropout(x)
+
+        x = self.transformer(x)
+
+        if not exists(self.mlp_head):
+            return x
+
+        x = x.mean(dim = 1) if self.pool == 'mean' else x[:, 0]
+
+        x = self.to_latent(x)
+        return self.mlp_head(x)
+
+if __name__ == '__main__':
+    v = ViT(
+        image_size = 256,
+        patch_size = 32,
+        num_classes = 1000,
+        dim = 1024,
+        depth = 6,
+        heads = 16,
+        mlp_dim = 2048,
+        dropout = 0.1,
+        emb_dropout = 0.1
+    )
+
+    img = torch.randn(1, 3, 256, 256)
+
+    preds = v(img)
+
+    assert preds.shape == (1, 1000)

vit_pytorch/vit_with_patch_dropout.py

@@ -97,7 +97,7 @@ class ViT(nn.Module):
     def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, pool = 'cls', channels = 3, dim_head = 64, dropout = 0., emb_dropout = 0., patch_dropout = 0.25):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
 

vit_pytorch/vit_with_patch_merger.py

@@ -108,7 +108,7 @@ class ViT(nn.Module):
     def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, patch_merge_layer = None, patch_merge_num_tokens = 8, channels = 3, dim_head = 64, dropout = 0., emb_dropout = 0.):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(patch_size)
+        self.patch_size = patch_height, patch_width = pair(patch_size)
 
         assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
 

vit_pytorch/vivit.py

@@ -1,5 +1,10 @@
+from collections import namedtuple
+
 import torch
-from torch import nn
+from torch import nn, cat
+import torch.nn.functional as F
+from torch.nn import Module, ModuleList
+from torch.nn.attention import SDPBackend, sdpa_kernel
 
 from einops import rearrange, repeat, reduce
 from einops.layers.torch import Rearrange
@@ -9,12 +14,15 @@ from einops.layers.torch import Rearrange
 def exists(val):
     return val is not None
 
+def divisible_by(num, den):
+    return (num % den) == 0
+
 def pair(t):
     return t if isinstance(t, tuple) else (t, t)
 
 # classes
 
-class FeedForward(nn.Module):
+class FeedForward(Module):
     def __init__(self, dim, hidden_dim, dropout = 0.):
         super().__init__()
         self.net = nn.Sequential(
@@ -28,9 +36,11 @@ class FeedForward(nn.Module):
     def forward(self, x):
         return self.net(x)
 
-class Attention(nn.Module):
-    def __init__(self, dim, heads = 8, dim_head = 64, dropout = 0.):
+class Attention(Module):
+    def __init__(self, dim, heads = 8, dim_head = 64, dropout = 0., use_flash_attn = True):
         super().__init__()
+        self.use_flash_attn = use_flash_attn
+        self.dropout_p = dropout
         inner_dim = dim_head *  heads
         project_out = not (heads == 1 and dim_head == dim)
 
@@ -48,61 +58,100 @@ class Attention(nn.Module):
             nn.Dropout(dropout)
         ) if project_out else nn.Identity()
 
-    def forward(self, x):
+    def flash_attn(self, q, k, v, mask = None):
+
+        with sdpa_kernel([SDPBackend.MATH, SDPBackend.EFFICIENT_ATTENTION, SDPBackend.FLASH_ATTENTION, SDPBackend.CUDNN_ATTENTION]):
+
+            out = F.scaled_dot_product_attention(
+                q, k, v,
+                attn_mask = mask,
+                dropout_p = self.dropout_p,
+                is_causal = False,
+                scale = self.scale
+            )
+
+        return out
+
+    def forward(self, x, mask = None):
+        batch, seq, _ = x.shape
+
         x = self.norm(x)
         qkv = self.to_qkv(x).chunk(3, dim = -1)
         q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b h n d', h = self.heads), qkv)
 
-        dots = torch.matmul(q, k.transpose(-1, -2)) * self.scale
+        if exists(mask):
+            mask = rearrange(mask, 'b j -> b 1 1 j')
 
-        attn = self.attend(dots)
-        attn = self.dropout(attn)
+        if self.use_flash_attn:
+            out =  self.flash_attn(q, k, v, mask = mask)
+
+        else:
+            dots = torch.matmul(q, k.transpose(-1, -2)) * self.scale
+
+            if exists(mask):
+                dots = dots.masked_fill(~mask, -torch.finfo(dots.dtype).max)
+
+            attn = self.attend(dots)
+            attn = self.dropout(attn)
+
+            out = torch.matmul(attn, v)
 
-        out = torch.matmul(attn, v)
         out = rearrange(out, 'b h n d -> b n (h d)')
         return self.to_out(out)
 
-class Transformer(nn.Module):
-    def __init__(self, dim, depth, heads, dim_head, mlp_dim, dropout = 0.):
+class Transformer(Module):
+    def __init__(self, dim, depth, heads, dim_head, mlp_dim, dropout = 0., use_flash_attn = True):
         super().__init__()
+        self.use_flash_attn = use_flash_attn
+
         self.norm = nn.LayerNorm(dim)
-        self.layers = nn.ModuleList([])
+        self.layers = ModuleList([])
         for _ in range(depth):
             self.layers.append(nn.ModuleList([
-                Attention(dim, heads = heads, dim_head = dim_head, dropout = dropout),
+                Attention(dim, heads = heads, dim_head = dim_head, dropout = dropout, use_flash_attn = use_flash_attn),
                 FeedForward(dim, mlp_dim, dropout = dropout)
             ]))
-    def forward(self, x):
+
+    def forward(self, x, mask = None):
+
         for attn, ff in self.layers:
-            x = attn(x) + x
+            x = attn(x, mask = mask) + x
             x = ff(x) + x
+
         return self.norm(x)
 
-class FactorizedTransformer(nn.Module):
-    def __init__(self, dim, depth, heads, dim_head, mlp_dim, dropout = 0.):
+class FactorizedTransformer(Module):
+    def __init__(self, dim, depth, heads, dim_head, mlp_dim, dropout = 0., use_flash_attn = True):
         super().__init__()
+        self.use_flash_attn = use_flash_attn
+
         self.norm = nn.LayerNorm(dim)
         self.layers = nn.ModuleList([])
+
         for _ in range(depth):
             self.layers.append(nn.ModuleList([
-                Attention(dim, heads = heads, dim_head = dim_head, dropout = dropout),
-                Attention(dim, heads = heads, dim_head = dim_head, dropout = dropout),
+                Attention(dim, heads = heads, dim_head = dim_head, dropout = dropout, use_flash_attn = use_flash_attn),
+                Attention(dim, heads = heads, dim_head = dim_head, dropout = dropout, use_flash_attn = use_flash_attn),
                 FeedForward(dim, mlp_dim, dropout = dropout)
             ]))
 
-    def forward(self, x):
-        b, f, n, _ = x.shape
+    def forward(self, x, mask = None):
+        batch, frames, seq, _ = x.shape
+
+        if exists(mask):
+            mask = repeat(mask, 'b ... -> (b space) ...', space = x.shape[2])
+
         for spatial_attn, temporal_attn, ff in self.layers:
             x = rearrange(x, 'b f n d -> (b f) n d')
             x = spatial_attn(x) + x
-            x = rearrange(x, '(b f) n d -> (b n) f d', b=b, f=f)
-            x = temporal_attn(x) + x
+            x = rearrange(x, '(b f) n d -> (b n) f d', b = batch, f = frames)
+            x = temporal_attn(x, mask = mask) + x
             x = ff(x) + x
-            x = rearrange(x, '(b n) f d -> b f n d', b=b, n=n)
+            x = rearrange(x, '(b n) f d -> b f n d', b = batch, n = seq)
 
         return self.norm(x)
 
-class ViT(nn.Module):
+class ViViT(Module):
     def __init__(
         self,
         *,
@@ -122,13 +171,14 @@ class ViT(nn.Module):
         dropout = 0.,
         emb_dropout = 0.,
         variant = 'factorized_encoder',
+        use_flash_attn: bool = True,
     ):
         super().__init__()
         image_height, image_width = pair(image_size)
-        patch_height, patch_width = pair(image_patch_size)
+        self.patch_size = patch_height, patch_width = pair(image_patch_size)
 
-        assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
-        assert frames % frame_patch_size == 0, 'Frames must be divisible by frame patch size'
+        assert divisible_by(image_height, patch_height) and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
+        assert divisible_by(frames, frame_patch_size), 'Frames must be divisible by frame patch size'
         assert variant in ('factorized_encoder', 'factorized_self_attention'), f'variant = {variant} is not implemented'
 
         num_image_patches = (image_height // patch_height) * (image_width // patch_width)
@@ -138,6 +188,8 @@ class ViT(nn.Module):
 
         assert pool in {'cls', 'mean'}, 'pool type must be either cls (cls token) or mean (mean pooling)'
 
+        self.frame_patch_size = frame_patch_size
+
         self.global_average_pool = pool == 'mean'
 
         self.to_patch_embedding = nn.Sequential(
@@ -154,11 +206,11 @@ class ViT(nn.Module):
 
         if variant == 'factorized_encoder':
             self.temporal_cls_token = nn.Parameter(torch.randn(1, 1, dim)) if not self.global_average_pool else None
-            self.spatial_transformer = Transformer(dim, spatial_depth, heads, dim_head, mlp_dim, dropout)
-            self.temporal_transformer = Transformer(dim, temporal_depth, heads, dim_head, mlp_dim, dropout)
+            self.spatial_transformer = Transformer(dim, spatial_depth, heads, dim_head, mlp_dim, dropout, use_flash_attn)
+            self.temporal_transformer = Transformer(dim, temporal_depth, heads, dim_head, mlp_dim, dropout, use_flash_attn)
         elif variant == 'factorized_self_attention':
             assert spatial_depth == temporal_depth, 'Spatial and temporal depth must be the same for factorized self-attention'
-            self.factorized_transformer = FactorizedTransformer(dim, spatial_depth, heads, dim_head, mlp_dim, dropout)
+            self.factorized_transformer = FactorizedTransformer(dim, spatial_depth, heads, dim_head, mlp_dim, dropout, use_flash_attn)
 
         self.pool = pool
         self.to_latent = nn.Identity()
@@ -166,25 +218,36 @@ class ViT(nn.Module):
         self.mlp_head = nn.Linear(dim, num_classes)
         self.variant = variant
 
-    def forward(self, video):
-        x = self.to_patch_embedding(video)
-        b, f, n, _ = x.shape
+    def forward(self, video, mask = None):
+        device = video.device
 
-        x = x + self.pos_embedding[:, :f, :n]
+        x = self.to_patch_embedding(video)
+        batch, frames, seq, _ = x.shape
+
+        x = x + self.pos_embedding[:, :frames, :seq]
 
         if exists(self.spatial_cls_token):
-            spatial_cls_tokens = repeat(self.spatial_cls_token, '1 1 d -> b f 1 d', b = b, f = f)
-            x = torch.cat((spatial_cls_tokens, x), dim = 2)
+            spatial_cls_tokens = repeat(self.spatial_cls_token, '1 1 d -> b f 1 d', b = batch, f = frames)
+            x = cat((spatial_cls_tokens, x), dim = 2)
 
         x = self.dropout(x)
 
+        # maybe temporal mask
+
+        temporal_mask = None
+
+        if exists(mask):
+            temporal_mask = reduce(mask, 'b (f patch) -> b f', 'all', patch = self.frame_patch_size)
+
+        # the two variants
+
         if self.variant == 'factorized_encoder':
             x = rearrange(x, 'b f n d -> (b f) n d')
 
             # attend across space
 
             x = self.spatial_transformer(x)
-            x = rearrange(x, '(b f) n d -> b f n d', b = b)
+            x = rearrange(x, '(b f) n d -> b f n d', b = batch)
 
             # excise out the spatial cls tokens or average pool for temporal attention
 
@@ -193,22 +256,50 @@ class ViT(nn.Module):
             # append temporal CLS tokens
 
             if exists(self.temporal_cls_token):
-                temporal_cls_tokens = repeat(self.temporal_cls_token, '1 1 d-> b 1 d', b = b)
+                temporal_cls_tokens = repeat(self.temporal_cls_token, '1 1 d-> b 1 d', b = batch)
 
-                x = torch.cat((temporal_cls_tokens, x), dim = 1)
-            
+                x = cat((temporal_cls_tokens, x), dim = 1)
+
+                if exists(temporal_mask):
+                    temporal_mask = F.pad(temporal_mask, (1, 0), value = True)
 
             # attend across time
 
-            x = self.temporal_transformer(x)
+            x = self.temporal_transformer(x, mask = temporal_mask)
 
             # excise out temporal cls token or average pool
 
             x = x[:, 0] if not self.global_average_pool else reduce(x, 'b f d -> b d', 'mean')
 
         elif self.variant == 'factorized_self_attention':
-            x = self.factorized_transformer(x)
+
+            x = self.factorized_transformer(x, mask = temporal_mask)
+
             x = x[:, 0, 0] if not self.global_average_pool else reduce(x, 'b f n d -> b d', 'mean')
 
         x = self.to_latent(x)
         return self.mlp_head(x)
+
+# main
+
+if __name__ == '__main__':
+
+    vivit = ViViT(
+        dim = 512,
+        spatial_depth = 2,
+        temporal_depth = 2,
+        heads = 4,
+        mlp_dim = 2048,
+        image_size = 256,
+        image_patch_size = 16,
+        frames = 8,
+        frame_patch_size = 2,
+        num_classes = 1000,
+        variant = 'factorized_encoder',
+    )
+
+    video = torch.randn(3, 3, 8, 256, 256)
+    mask = torch.randint(0, 2, (3, 8)).bool()
+
+    logits = vivit(video, mask = None)
+    assert logits.shape == (3, 1000)

vit_pytorch/vivit_with_moss.py

@@ -0,0 +1,386 @@
+# https://openreview.net/forum?id=Co6SCyBIjo
+# applied at https://arxiv.org/abs/2605.03269 - 50-85% jump in pick-place moving conveyer belt
+
+import torch
+from torch import nn
+from torch.nn import Module, ModuleList
+import torch.nn.functional as F
+from torch.nn.attention import SDPBackend, sdpa_kernel
+
+from einops import rearrange, repeat, reduce, einsum
+from einops.layers.torch import Rearrange
+
+# helpers
+
+def exists(val):
+    return val is not None
+
+
+def divisible_by(num, den):
+    return (num % den) == 0
+
+def is_odd(n):
+    return not divisible_by(n, 2)
+
+def pair(t):
+    return t if isinstance(t, tuple) else (t, t)
+
+def l2norm(t):
+    return F.normalize(t, dim = -1)
+
+# normalization helpers
+
+class ChanLayerNorm(Module):
+    def __init__(self, dim, eps = 1e-5):
+        super().__init__()
+        self.eps = eps
+        self.gamma = nn.Parameter(torch.ones(1, dim, 1, 1))
+
+    def forward(self, x):
+        var = torch.var(x, dim = 1, unbiased = False, keepdim = True)
+        mean = torch.mean(x, dim = 1, keepdim = True)
+        return (x - mean) * (var + self.eps).rsqrt() * self.gamma
+
+# classes
+
+class FeedForward(Module):
+    def __init__(self, dim, hidden_dim, dropout = 0.):
+        super().__init__()
+        self.net = nn.Sequential(
+            nn.LayerNorm(dim),
+            nn.Linear(dim, hidden_dim),
+            nn.GELU(),
+            nn.Dropout(dropout),
+            nn.Linear(hidden_dim, dim),
+            nn.Dropout(dropout)
+        )
+    def forward(self, x):
+        return self.net(x)
+
+class Attention(Module):
+    def __init__(self, dim, heads = 8, dim_head = 64, dropout = 0., use_flash_attn = True):
+        super().__init__()
+        self.use_flash_attn = use_flash_attn
+        self.dropout_p = dropout
+        inner_dim = dim_head * heads
+        project_out = not (heads == 1 and dim_head == dim)
+
+        self.heads = heads
+        self.scale = dim_head ** -0.5
+
+        self.norm = nn.LayerNorm(dim)
+        self.attend = nn.Softmax(dim = -1)
+        self.dropout = nn.Dropout(dropout)
+
+        self.to_qkv = nn.Linear(dim, inner_dim * 3, bias = False)
+
+        self.to_out = nn.Sequential(
+            nn.Linear(inner_dim, dim),
+            nn.Dropout(dropout)
+        ) if project_out else nn.Identity()
+
+    def flash_attn(self, q, k, v, mask = None):
+        with sdpa_kernel([SDPBackend.MATH, SDPBackend.EFFICIENT_ATTENTION, SDPBackend.FLASH_ATTENTION, SDPBackend.CUDNN_ATTENTION]):
+            out = F.scaled_dot_product_attention(
+                q, k, v,
+                attn_mask = mask,
+                dropout_p = self.dropout_p,
+                is_causal = False,
+                scale = self.scale
+            )
+        return out
+
+    def forward(self, x, mask = None):
+        x = self.norm(x)
+        qkv = self.to_qkv(x).chunk(3, dim = -1)
+        q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b h n d', h = self.heads), qkv)
+
+        if exists(mask):
+            mask = rearrange(mask, 'b j -> b 1 1 j')
+
+        if self.use_flash_attn:
+            out = self.flash_attn(q, k, v, mask = mask)
+        else:
+            dots = einsum(q, k, 'b h i d, b h j d -> b h i j') * self.scale
+            if exists(mask):
+                dots = dots.masked_fill(~mask, -torch.finfo(dots.dtype).max)
+
+            attn = self.attend(dots)
+            attn = self.dropout(attn)
+            out = torch.matmul(attn, v)
+
+        out = rearrange(out, 'b h n d -> b n (h d)')
+        return self.to_out(out)
+
+class Transformer(Module):
+    def __init__(self, dim, depth, heads, dim_head, mlp_dim, dropout = 0., use_flash_attn = True):
+        super().__init__()
+        self.norm = nn.LayerNorm(dim)
+        self.layers = ModuleList([])
+        for _ in range(depth):
+            self.layers.append(ModuleList([
+                Attention(dim, heads = heads, dim_head = dim_head, dropout = dropout, use_flash_attn = use_flash_attn),
+                FeedForward(dim, mlp_dim, dropout = dropout)
+            ]))
+
+    def forward(self, x, mask = None):
+        for attn, ff in self.layers:
+            x = attn(x, mask = mask) + x
+            x = ff(x) + x
+        return self.norm(x)
+
+# moss specific classes
+
+class STSSEncoder(Module):
+    def __init__(self, dim, local_time = 3, local_height = 3, local_width = 3, hidden_dim = 64):
+        super().__init__()
+
+        self.spatial_to_hidden = nn.Linear(local_height * local_width, hidden_dim)
+
+        self.conv = nn.Sequential(
+            nn.Conv2d(hidden_dim, hidden_dim, 3, padding = 1),
+            ChanLayerNorm(hidden_dim),
+            nn.GELU(),
+            nn.Conv2d(hidden_dim, hidden_dim, 3, padding = 1),
+            ChanLayerNorm(hidden_dim),
+            nn.GELU()
+        )
+
+        self.time_to_out = nn.Linear(local_time * hidden_dim, dim)
+
+    def forward(self, sim):
+        b, t, h, w, lt, lh, lw = sim.shape
+
+        x = rearrange(sim, 'b t h w lt lh lw -> b t h w lt (lh lw)')
+        x = self.spatial_to_hidden(x)
+
+        x = rearrange(x, 'b t h w lt d -> (b t lt) d h w')
+        x = self.conv(x)
+        x = rearrange(x, '(b t lt) d h w -> b t h w (lt d)', b = b, t = t, lt = lt)
+
+        return self.time_to_out(x)
+
+class MOSS(Module):
+    def __init__(
+        self,
+        dim,
+        local_time = 3,
+        local_height = 3,
+        local_width = 3,
+        hidden_dim = 64,
+        orders = 2,
+        causal = False
+    ):
+        super().__init__()
+        assert is_odd(local_time) and is_odd(local_height) and is_odd(local_width), 'MOSS local dimensions must be odd'
+
+        self.local_time = local_time
+        self.local_height = local_height
+        self.local_width = local_width
+        self.causal = causal
+
+        self.encoders = ModuleList([STSSEncoder(dim, local_time, local_height, local_width, hidden_dim) for _ in range(orders)])
+        self.to_order_out = ModuleList([nn.Linear(dim, dim) for _ in range(orders)])
+        self.to_out = nn.Linear(dim, dim)
+
+    def stss_transform(self, x):
+        lt, lh, lw = self.local_time, self.local_height, self.local_width
+
+        x = l2norm(x)
+        x = rearrange(x, 'b t h w c -> b c t h w')
+
+        pad_h, pad_w = lh // 2, lw // 2
+        pad_t_past, pad_t_future = (lt - 1, 0) if self.causal else (lt // 2, lt // 2)
+
+        padded_x = F.pad(x, (pad_w, pad_w, pad_h, pad_h, pad_t_past, pad_t_future))
+        windows = padded_x.unfold(2, lt, 1).unfold(3, lh, 1).unfold(4, lw, 1)
+
+        return einsum(x, windows, 'b c t h w, b c t h w l u v -> b t h w l u v')
+
+    def forward(self, x):
+        out = self.to_out(x)
+
+        for encoder, to_order_out in zip(self.encoders, self.to_order_out):
+            sim = self.stss_transform(x)
+            x = encoder(sim)
+            out = out + to_order_out(x)
+
+        return out
+
+# main architecture
+
+class ViViT(Module):
+    def __init__(
+        self,
+        *,
+        image_size,
+        image_patch_size,
+        frames,
+        frame_patch_size,
+        num_classes,
+        dim,
+        spatial_depth,
+        temporal_depth,
+        heads,
+        mlp_dim,
+        pool = 'cls',
+        channels = 3,
+        dim_head = 64,
+        dropout = 0.,
+        emb_dropout = 0.,
+        use_flash_attn: bool = True,
+        moss_local_time = 3,
+        moss_local_height = 3,
+        moss_local_width = 3,
+        moss_hidden_dim = 64,
+        moss_orders = 2,
+        moss_causal = True,
+    ):
+        super().__init__()
+        image_height, image_width = pair(image_size)
+        self.patch_size = patch_height, patch_width = pair(image_patch_size)
+
+        assert divisible_by(image_height, patch_height) and divisible_by(image_width, patch_width), 'Image dimensions must be divisible by the patch size.'
+        assert divisible_by(frames, frame_patch_size), 'Frames must be divisible by frame patch size'
+        assert pool in {'cls', 'mean'}, 'pool type must be either cls (cls token) or mean (mean pooling)'
+
+        num_image_patches = (image_height // patch_height) * (image_width // patch_width)
+        num_frame_patches = frames // frame_patch_size
+        patch_dim = channels * patch_height * patch_width * frame_patch_size
+
+        self.frame_patch_size = frame_patch_size
+        self.patch_h = image_height // patch_height
+        self.patch_w = image_width // patch_width
+        self.global_average_pool = pool == 'mean'
+
+        self.to_patch_embedding = nn.Sequential(
+            Rearrange('b c (f pf) (h p1) (w p2) -> b f (h w) (pf p1 p2 c)', p1 = patch_height, p2 = patch_width, pf = frame_patch_size),
+            nn.LayerNorm(patch_dim),
+            nn.Linear(patch_dim, dim),
+            nn.LayerNorm(dim)
+        )
+
+        self.pos_embedding = nn.Parameter(torch.randn(1, num_frame_patches, num_image_patches, dim))
+        self.dropout = nn.Dropout(emb_dropout)
+
+        self.has_cls = not self.global_average_pool
+        self.spatial_cls_token = nn.Parameter(torch.randn(1, 1, dim)) if self.has_cls else None
+        self.temporal_cls_token = nn.Parameter(torch.randn(1, 1, dim)) if self.has_cls else None
+
+        self.spatial_transformer = Transformer(dim, spatial_depth, heads, dim_head, mlp_dim, dropout, use_flash_attn)
+        self.temporal_transformer = Transformer(dim, temporal_depth, heads, dim_head, mlp_dim, dropout, use_flash_attn)
+
+        self.moss = MOSS(
+            dim,
+            local_time = moss_local_time,
+            local_height = moss_local_height,
+            local_width = moss_local_width,
+            hidden_dim = moss_hidden_dim,
+            orders = moss_orders,
+            causal = moss_causal
+        )
+
+        self.mlp_head = nn.Sequential(
+            nn.LayerNorm(dim),
+            nn.Linear(dim, num_classes)
+        )
+
+    def forward(self, video, mask = None):
+        x = self.to_patch_embedding(video)
+        batch, frames, seq, _ = x.shape
+
+        x = x + self.pos_embedding[:, :frames, :seq]
+
+        if self.has_cls:
+            spatial_cls_tokens = repeat(self.spatial_cls_token, '1 1 d -> b f 1 d', b = batch, f = frames)
+            x = torch.cat((spatial_cls_tokens, x), dim = 2)
+
+        x = self.dropout(x)
+
+        # temporal mask
+
+        temporal_mask = None
+        if exists(mask):
+            temporal_mask = reduce(mask, 'b (f patch) -> b f', 'all', patch = self.frame_patch_size)
+
+        x = rearrange(x, 'b f n d -> (b f) n d')
+
+        # attend across space
+        x = self.spatial_transformer(x)
+        x = rearrange(x, '(b f) n d -> b f n d', b = batch)
+
+        # moss integration over spatial patch tokens
+
+        if self.has_cls:
+            spatial_cls_tokens, patch_tokens = x[:, :, :1], x[:, :, 1:]
+        else:
+            patch_tokens = x
+
+        patch_tokens = rearrange(patch_tokens, 'b f (h w) d -> b f h w d', h = self.patch_h, w = self.patch_w)
+        patch_tokens = self.moss(patch_tokens)
+        patch_tokens = rearrange(patch_tokens, 'b f h w d -> b f (h w) d')
+
+        # pool spatial features
+
+        moss_pooled = reduce(patch_tokens, 'b f n d -> b f d', 'mean')
+
+        if self.has_cls:
+            x = rearrange(spatial_cls_tokens, 'b f 1 d -> b f d') + moss_pooled
+        else:
+            x = moss_pooled
+
+        # append temporal cls tokens
+
+        if self.has_cls:
+            temporal_cls_tokens = repeat(self.temporal_cls_token, '1 1 d -> b 1 d', b = batch)
+            x = torch.cat((temporal_cls_tokens, x), dim = 1)
+
+            if exists(temporal_mask):
+                temporal_mask = F.pad(temporal_mask, (1, 0), value = True)
+
+        # attend across time
+
+        x = self.temporal_transformer(x, mask = temporal_mask)
+
+        # temporal pooling
+
+        x = x[:, 0] if self.has_cls else reduce(x, 'b f d -> b d', 'mean')
+
+        return self.mlp_head(x)
+
+if __name__ == '__main__':
+    vivit = ViViT(
+        dim = 512,
+        spatial_depth = 2,
+        temporal_depth = 2,
+        heads = 4,
+        mlp_dim = 2048,
+        image_size = 256,
+        image_patch_size = 32,
+        frames = 8,
+        frame_patch_size = 2,
+        num_classes = 1000,
+    )
+
+    video = torch.randn(2, 3, 8, 256, 256)
+    mask = torch.randint(0, 2, (2, 8)).bool()
+
+    logits = vivit(video, mask = None)
+    assert logits.shape == (2, 1000)
+
+    logits = vivit(video, mask = mask)
+    assert logits.shape == (2, 1000)
+
+    moss = MOSS(
+        dim = 512,
+        local_time = 3,
+        local_height = 3,
+        local_width = 3,
+        hidden_dim = 64,
+        orders = 2,
+        causal = True
+    )
+
+    moss_input = torch.randn(2, 8, 16, 16, 512) # (batch, frames, height, width, dim)
+    moss_output = moss(moss_input)
+    assert moss_output.shape == (2, 8, 16, 16, 512)