Update README.md

README.md

@@ -1,8 +1,8 @@
----
-license: apache-2.0
-datasets:
-- ILSVRC/imagenet-1k
----
+---
+license: apache-2.0
+datasets:
+- ILSVRC/imagenet-1k
+---
 
 # Efficient Training of Diffusion Mixture-of-Experts Models: A Practical Recipe
 
@@ -13,34 +13,13 @@ datasets:
 </p>
 
 
-## 1. 🔥 Updates
-- __[2025.12.15]__: Release the codes of [DSMoE](./DSMoE) and [JiTMoE](./JiTMoE).
-
-
-## 2. 📖 Introduction
+## 📖 Introduction
 
 We release the MoE Transformer that can be applied to both latent and pixel-space diffusion frameworks, employing DeepSeek-style expert modules, alternative intermediate widths, varying expert counts, and enhanced attention positional encodings. The models are already relased to Huggingface. <br>
 
-## 3. Preparation
-
-### 3.1 Dataset
-Download [ImageNet](http://image-net.org/download) dataset, and place it in your `IMAGENET_PATH`.
-
-### 3.2 Installation
-
-Please follow the installations of [DiffMoE](https://github.com/KlingTeam/DiffMoE) and [JiT](https://github.com/LTH14/JiT), respectively.
-
-### 3.3 Training
-
-See details in [DSMoE](./DSMoE) and [JiTMoE](./JiTMoE) respectively.
-
-### 3.4 Evaluation
-
-We follow the evaluation protocols provided by [DiffMoE](https://github.com/KlingTeam/DiffMoE) and [JiT](https://github.com/LTH14/JiT). 
-
-## 4. Main results
+## Main results
 
-### 4.1 Latent diffusion framework
+### Latent diffusion framework
 
  - Ours DSMoE v.s. [DiffMoE](https://arxiv.org/pdf/2503.14487) on 700K training steps with CFG = 1.0 (* refers to the reported results in the official paper):
 
@@ -67,4 +46,32 @@ We follow the evaluation protocols provided by [DiffMoE](https://github.com/Klin
 |DiffMoE-B-E16|130M|4.87|183.43|
 |DSMoE-B-E16|132M|4.50|186.79|
 |DSMoE-B-E48|118M|4.27|191.03|
-|DiffMoE-L-E16|458M|2.84|256.57|
+|DiffMoE-L-E16|458M|2.84|256.57|
+|DSMoE-L-E16|465M|2.59|272.55|
+|DSMoE-L-E48|436M|2.55|278.35|
+|DSMoE-3B-E16|965M|2.38|304.93|
+
+
+### Pixel-space diffusion framework 
+
+-  Ours JiTMoE v.s. [JiT](https://arxiv.org/pdf/2511.13720) on 200 training epochs with CFG interval (* refers to the reported results in the official paper):
+
+| Model Name                 | # Act. Params | FID-50K↓  | Inception Score↑ |
+|----------------------------|-------------------------|---------|----------------|
+|JiT-B/16|131M|4.81 (4.37*)| 222.32 (-)|
+|JiTMoE-B/16-E16|133M|4.23| 245.53|
+|JiT-L/16|459M| 3.19 (2.79*)| 309.72 (-)|
+|JiTMoE-L/16-E16|465M|3.10| 311.34|
+
+
+## 🌟 Citation
+
+```
+@article{liu2025efficient,
+  title={Efficient Training of Diffusion Mixture-of-Experts Models: A Practical Recipe},
+  author={Liu, Yahui and Yue, Yang and Zhang, Jingyuan and Sun, Chenxi and Zhou, Yang and Zeng, Wencong and Tang, Ruiming and Zhou, Guorui},
+  journal={arXiv preprint arXiv:2512.01252},
+  year={2025}
+}
+```
+