Update README.md
Browse files
README.md
CHANGED
|
@@ -7,8 +7,8 @@ license: cc-by-nc-4.0
|
|
| 7 |
This is a scaled-up version of the checkpoint originally presented in our preprint. This model has 12 layers with 12 attention heads each.
|
| 8 |
|
| 9 |
Trained with larger dataset of multiple initial conditions per system, with mixed periods as well.
|
| 10 |
-
Specifically, using 8 out of the 16 initial conditions (ICs) per system that we provide in [skew-mixedp-ic16 dataset](https://huggingface.co/datasets/GilpinLab/skew-mixedp-ic16)
|
| 11 |
-
|
| 12 |
*Panda*: Patched Attention for Nonlinear Dynamics.
|
| 13 |
|
| 14 |
Paper abstract:
|
|
|
|
| 7 |
This is a scaled-up version of the checkpoint originally presented in our preprint. This model has 12 layers with 12 attention heads each.
|
| 8 |
|
| 9 |
Trained with larger dataset of multiple initial conditions per system, with mixed periods as well.
|
| 10 |
+
Specifically, using 8 out of the 16 initial conditions (ICs) per system that we provide in our [skew-mixedp-ic16 dataset](https://huggingface.co/datasets/GilpinLab/skew-mixedp-ic16)
|
| 11 |
+
We trained this model with per-device batch size 384, across 6 AMD MI100X GPUs, for 800k iterations.
|
| 12 |
*Panda*: Patched Attention for Nonlinear Dynamics.
|
| 13 |
|
| 14 |
Paper abstract:
|