Update README.md
Browse files
README.md
CHANGED
|
@@ -12,11 +12,7 @@ We introduce MMaDA, a novel class of multimodal diffusion foundation models desi
|
|
| 12 |
2. MMaDA introduces a mixed long chain-of-thought (CoT) fine-tuning strategy that curates a unified CoT format across modalities.
|
| 13 |
3. MMaDA adopts a unified policy-gradient-based RL algorithm, which we call UniGRPO, tailored for diffusion foundation models. Utilizing diversified reward modeling, UniGRPO unifies post-training across both reasoning and generation tasks, ensuring consistent performance improvements.
|
| 14 |
|
| 15 |
-
[Paper](https://arxiv.org/abs/2505.15809)
|
| 16 |
-
|
| 17 |
-
[Code](https://github.com/Gen-Verse/MMaDA)
|
| 18 |
-
|
| 19 |
-
[Demo](https://huggingface.co/spaces/Gen-Verse/MMaDA)
|
| 20 |
|
| 21 |
# Citation
|
| 22 |
|
|
|
|
| 12 |
2. MMaDA introduces a mixed long chain-of-thought (CoT) fine-tuning strategy that curates a unified CoT format across modalities.
|
| 13 |
3. MMaDA adopts a unified policy-gradient-based RL algorithm, which we call UniGRPO, tailored for diffusion foundation models. Utilizing diversified reward modeling, UniGRPO unifies post-training across both reasoning and generation tasks, ensuring consistent performance improvements.
|
| 14 |
|
| 15 |
+
[Paper](https://arxiv.org/abs/2505.15809) | [Code](https://github.com/Gen-Verse/MMaDA) | [Demo](https://huggingface.co/spaces/Gen-Verse/MMaDA)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
|
| 17 |
# Citation
|
| 18 |
|