openwam-review/OpenWAM-Alpha-Checkpoints

OpenWAM Alpha checkpoints

Anonymous supplementary checkpoints for review.

All 14 checkpoint groups are available.

Code and download instructions: https://github.com/openwam-review/OpenWAM

Each group contains its saved configuration, safetensors weights, and tokenizer. Task checkpoints also contain normalization statistics. The foundation checkpoint is intended for fine-tuning, not direct robot deployment.

Configurations use local paths and disable W&B by default. Weight tensors and normalization statistics are unchanged. Run deployment from the code repository root. The dexterous-hand training adapter and some original pretraining data adapters are not bundled in the code release; providing these checkpoints does not imply those training pipelines are included. Physical-robot use requires an appropriate observation/action adapter.

Checkpoint groups

• OpenWAM-Alpha-Pretrain-Foundation-Model
• OpenWAM-Alpha-Real-Dexterous-Hand
• OpenWAM-Alpha-Real-RoboDojo-ARX-X5
• OpenWAM-Alpha-Real-RoboDojo-Piper
• OpenWAM-Alpha-Real-RoboDojo-Piper-X
• OpenWAM-Alpha-Real-Single-Arm-Franka
• OpenWAM-Alpha-Sim-EBench
• OpenWAM-Alpha-Sim-LIBERO
• OpenWAM-Alpha-Sim-RoboCasa-GR1
• OpenWAM-Alpha-Sim-RoboCasa365
• OpenWAM-Alpha-Sim-RoboDojo
• OpenWAM-Alpha-Sim-RoboTwin-Clean2Random
• OpenWAM-Alpha-Sim-RoboTwin-Full
• OpenWAM-Alpha-Sim-VLABench