This repository hosts the checkpoints and datasets for Supply And Demand Loss (SADL), described in SADL: On the Demand-Supply Geometry of Representation Collapse. The official PyTorch training and evaluation code is in zju3dv/SADL on GitHub. Project page.
Models are trained on PushT, Cube, Reacher and TwoRoom with one BatchNorm population shared by all tasks and GPUs, for two objectives:
checkpoints/<model>/<training>/seed<seed>/<task>/
weights_epoch_1.pt model weights and BatchNorm statistics
config.yaml model configuration
action_stats.json action normalization of the task
training_config.json training configuration of the run
datasets/
pusht_expert_train.h5 cube_single_expert.h5 reacher.h5 tworoom.h5
SHA256SUMS
CHECKPOINTS_SHA256.txt
<model> is intact or lewm; <training> is four_task (one model trained on all four tasks; the four task
folders share every weight and normalization statistic and differ only in action dimension and action scaling) or
single_task (one model per task); seeds are 3072, 3073 and 3074. Every checkpoint is trained for one epoch.
Training checkpoints for J, E, P, and A are available separately. Each letter is an epoch-1 INTACT-Lite model trained jointly on PushT, Cube, Reacher, and TwoRoom.
git clone https://github.com/zju3dv/SADL.git && cd SADL
pip install -r requirements.txt
hf download DavidSunok/SADL --local-dir hf
python evaluate.py --checkpoint hf/checkpoints/intact/four_task/seed3072/reacher --task reacher \
--data-dir hf/datasets --planner direct
python evaluate.py --checkpoint hf/checkpoints/lewm/single_task/seed3072/pusht --task pusht \
--data-dir hf/datasets --planner cem
Success rate (%), mean +- SD over the three training seeds; each seed is evaluated with seeds 0, 1 and 42, 100 episodes each, on an NVIDIA RTX 4090 (goal 25 steps ahead, at most 50 actions, horizon 5, replanning every 5 actions).
| Model | Training | Planner | PushT | Cube | Reacher | TwoRoom | Mean |
|---|---|---|---|---|---|---|---|
| INTACT | four tasks | Direct | 85.7 +- 0.9 | 100.0 +- 0.0 | 93.4 +- 1.7 | 100.0 +- 0.0 | 94.8 +- 0.4 |
| INTACT | four tasks | Guarded A | 88.1 +- 0.8 | 99.6 +- 0.4 | 93.1 +- 0.7 | 99.8 +- 0.4 | 95.1 +- 0.3 |
| INTACT | single task | Direct | 85.1 +- 0.4 | 100.0 +- 0.0 | 94.9 +- 2.0 | 99.7 +- 0.6 | 94.9 +- 0.5 |
| INTACT | single task | Guarded A | 89.8 +- 0.2 | 99.6 +- 0.2 | 93.7 +- 3.2 | 99.3 +- 0.9 | 95.6 +- 0.9 |
| LeWM | four tasks | CEM | 64.7 +- 18.5 | 59.6 +- 1.3 | 46.7 +- 26.1 | 94.4 +- 1.0 | 66.3 +- 9.2 |
| LeWM | single task | CEM | 83.1 +- 1.6 | 58.9 +- 1.4 | 81.0 +- 4.3 | 91.3 +- 1.5 | 78.6 +- 1.9 |
The HDF5 files hold expert demonstrations with uint8 RGB frames (224 x 224) and actions. PushT, Cube, Reacher and TwoRoom come from the LeWorldModel benchmark. The original dataset cards specify the MIT license: PushT, Cube, Reacher, and TwoRoom. The upstream copyright and permission notice is included in datasets/LeWM-MIT-LICENSE.txt.
The MIT license applies to the released weights. The upstream dataset license is linked above.
checkpoints/intact/four_task/seed3072/cube/weights_epoch_1.pt 36 MBcheckpoints/intact/four_task/seed3072/pusht/weights_epoch_1.pt 36 MBcheckpoints/intact/four_task/seed3072/reacher/weights_epoch_1.pt 36 MBcheckpoints/intact/four_task/seed3072/tworoom/weights_epoch_1.pt 36 MBcheckpoints/intact/four_task/seed3073/cube/weights_epoch_1.pt 36 MBcheckpoints/intact/four_task/seed3073/pusht/weights_epoch_1.pt 36 MBcheckpoints/intact/four_task/seed3073/reacher/weights_epoch_1.pt 36 MBcheckpoints/intact/four_task/seed3073/tworoom/weights_epoch_1.pt 36 MBcheckpoints/intact/four_task/seed3074/cube/weights_epoch_1.pt 36 MBcheckpoints/intact/four_task/seed3074/pusht/weights_epoch_1.pt 36 MBcheckpoints/intact/four_task/seed3074/reacher/weights_epoch_1.pt 36 MBcheckpoints/intact/four_task/seed3074/tworoom/weights_epoch_1.pt 36 MBcheckpoints/intact/single_task/seed3072/cube/weights_epoch_1.pt 36 MBcheckpoints/intact/single_task/seed3073/cube/weights_epoch_1.pt 36 MBcheckpoints/intact/single_task/seed3074/cube/weights_epoch_1.pt 36 MBcheckpoints/lewm/four_task/seed3072/cube/weights_epoch_1.pt 36 MBcheckpoints/lewm/four_task/seed3072/pusht/weights_epoch_1.pt 36 MBcheckpoints/lewm/four_task/seed3072/reacher/weights_epoch_1.pt 36 MBcheckpoints/lewm/four_task/seed3072/tworoom/weights_epoch_1.pt 36 MBcheckpoints/lewm/four_task/seed3073/cube/weights_epoch_1.pt 36 MBcheckpoints/lewm/four_task/seed3073/pusht/weights_epoch_1.pt 36 MBcheckpoints/lewm/four_task/seed3073/reacher/weights_epoch_1.pt 36 MBcheckpoints/lewm/four_task/seed3073/tworoom/weights_epoch_1.pt 36 MBcheckpoints/lewm/four_task/seed3074/cube/weights_epoch_1.pt 36 MBcheckpoints/lewm/four_task/seed3074/pusht/weights_epoch_1.pt 36 MBcheckpoints/lewm/four_task/seed3074/reacher/weights_epoch_1.pt 36 MBcheckpoints/lewm/four_task/seed3074/tworoom/weights_epoch_1.pt 36 MBcheckpoints/lewm/single_task/seed3072/cube/weights_epoch_1.pt 36 MBcheckpoints/lewm/single_task/seed3073/cube/weights_epoch_1.pt 36 MBcheckpoints/lewm/single_task/seed3074/cube/weights_epoch_1.pt 36 MBcheckpoints/intact/single_task/seed3072/pusht/weights_epoch_1.pt 36 MBcheckpoints/intact/single_task/seed3072/reacher/weights_epoch_1.pt 36 MBcheckpoints/intact/single_task/seed3072/tworoom/weights_epoch_1.pt 36 MBcheckpoints/intact/single_task/seed3073/pusht/weights_epoch_1.pt 36 MBcheckpoints/intact/single_task/seed3073/reacher/weights_epoch_1.pt 36 MBcheckpoints/intact/single_task/seed3073/tworoom/weights_epoch_1.pt 36 MBcheckpoints/intact/single_task/seed3074/pusht/weights_epoch_1.pt 36 MBcheckpoints/intact/single_task/seed3074/reacher/weights_epoch_1.pt 36 MBcheckpoints/intact/single_task/seed3074/tworoom/weights_epoch_1.pt 36 MBcheckpoints/lewm/single_task/seed3072/pusht/weights_epoch_1.pt 36 MBcheckpoints/lewm/single_task/seed3072/reacher/weights_epoch_1.pt 36 MBcheckpoints/lewm/single_task/seed3072/tworoom/weights_epoch_1.pt 36 MBcheckpoints/lewm/single_task/seed3073/pusht/weights_epoch_1.pt 36 MBcheckpoints/lewm/single_task/seed3073/reacher/weights_epoch_1.pt 36 MBcheckpoints/lewm/single_task/seed3073/tworoom/weights_epoch_1.pt 36 MBcheckpoints/lewm/single_task/seed3074/pusht/weights_epoch_1.pt 36 MBcheckpoints/lewm/single_task/seed3074/reacher/weights_epoch_1.pt 36 MBcheckpoints/lewm/single_task/seed3074/tworoom/weights_epoch_1.pt 36 MBJEPA-Training/A/multitask_bundle_epoch_1.pt 29 MBJEPA-Training/E/multitask_bundle_epoch_1.pt 29 MBJEPA-Training/J/multitask_bundle_epoch_1.pt 29 MBJEPA-Training/P/multitask_bundle_epoch_1.pt 29 MB