Fix QK norm to apply before RoPE, retrain released checkpoints

README.md

@@ -22,8 +22,8 @@ Reconstruction quality, measured by applying downstream task heads to the recons
 | Method | Horizon | VSPW mIoU (↑) | Cityscapes mIoU (↑) | KITTI RMSE (↓) |
 |--------|---------|---------------|---------------------|----------------|
 | *Present (upper bound)* | — | *58.4* | *70.5* | *2.79* |
-| DeltaTok | Short (1 frame) | 58.6 | 69.6 | 2.78 |
-| DeltaTok | Mid (3 frames)* | 58.5 | 67.9 | 2.86 |
+| DeltaTok | Short (1 frame) | 58.5 | 69.7 | 2.80 |
+| DeltaTok | Mid (3 frames)* | 58.5 | 68.1 | 2.88 |
 
 *Parallel encoding from ground-truth frames with autoregressive decoding from previous reconstructions.
 

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:725678517be3230e9dc4ddc056004292e0653b53ffc92669426b4ccc4a94ffe7
-size 681697747
+oid sha256:4fb478e81e70e743472ec59f959573476f5acea205aaa918c9934211126d35a9
+size 681704533