Add resource links (interactive paper, live demo, dataset) to top of model card
Browse files
README.md
CHANGED
|
@@ -16,6 +16,12 @@ Pretrained and finetuned checkpoints for the 350M-parameter generative world mod
|
|
| 16 |
|
| 17 |
by [Nicklas Hansen](https://www.nicklashansen.com) and [Xiaolong Wang](https://xiaolonw.github.io) (UC San Diego).
|
| 18 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
The world model follows the architecture and two-stage training recipe of [Dreamer 4](https://arxiv.org/abs/2509.24527), adapted for large-scale multi-task continuous control, and is trained on **MMBench2** — a 427-hour, 210-task dataset for visual world modeling (see the [dataset repository](https://huggingface.co/datasets/nicklashansen/mmbench2)). Each variant is a `(tokenizer.pt, dynamics.pt)` pair at 224×224 resolution:
|
| 20 |
|
| 21 |
- **tokenizer** — a causal video tokenizer (≈50M-parameter encoder + ≈50M-parameter decoder, projecting to a 64-dim tanh-bounded latent).
|
|
|
|
| 16 |
|
| 17 |
by [Nicklas Hansen](https://www.nicklashansen.com) and [Xiaolong Wang](https://xiaolonw.github.io) (UC San Diego).
|
| 18 |
|
| 19 |
+
🌐 **Interactive paper:** https://www.nicklashansen.com/mmbench2
|
| 20 |
+
|
| 21 |
+
🕹️ **Live demo:** https://www.nicklashansen.com/mmbench2/#live-demo
|
| 22 |
+
|
| 23 |
+
📦 **Dataset:** https://huggingface.co/datasets/nicklashansen/mmbench2
|
| 24 |
+
|
| 25 |
The world model follows the architecture and two-stage training recipe of [Dreamer 4](https://arxiv.org/abs/2509.24527), adapted for large-scale multi-task continuous control, and is trained on **MMBench2** — a 427-hour, 210-task dataset for visual world modeling (see the [dataset repository](https://huggingface.co/datasets/nicklashansen/mmbench2)). Each variant is a `(tokenizer.pt, dynamics.pt)` pair at 224×224 resolution:
|
| 26 |
|
| 27 |
- **tokenizer** — a causal video tokenizer (≈50M-parameter encoder + ≈50M-parameter decoder, projecting to a 64-dim tanh-bounded latent).
|