Update weights/SimWAM.pt to the exact IL warm-start of SimWAM-RL.pt
Browse filesThe previously released SimWAM.pt was IL step_044400, while SimWAM-RL.pt was
warm-started from IL step_044000. Replace it with step_044000 so the released
supervised / RL checkpoint pair is self-consistent: all 1409 non-LoRA tensors of
SimWAM-RL.pt are bit-identical to this file, and only the 240 LoRA-target
action-expert q/k/v/o weights differ.
NAVSIM navtest PDMS: 90.29 (this file) vs 90.25 (previous, step_044400).
Previous blob sha256 1d05ec214008efa15c5b2fc5029e139fecc6cd0865eb7ca0160a28708db9398b
remains reachable at commit a862a0dc8576be777872c128266a5573d8ae3c40.
- weights/SimWAM.pt +1 -1
weights/SimWAM.pt
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 12041719353
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:401e61c902b722f11ba63479efa65c8aff855be5f8d147975030f3513da37ec9
|
| 3 |
size 12041719353
|