EasyWAM-Unified-Wan22-LoRA-128

EasyWAM-Unified places video, action, and robot-state tokens in a single Video DiT and jointly predicts future video and actions. This checkpoint was trained on LIBERO using rank-128 LoRA (r=128, alpha=128) with Wan2.2-TI2V-5B as the backbone.

This checkpoint contains LoRA and other trainable EasyWAM parameters, not a standalone full model. It must be loaded with the matching EasyWAM LoRA task configuration and base components.

Results

Success rate (%) under the EasyWAM LIBERO evaluation protocol:

Model	Spatial	Object	Goal	Long	Avg.
Full-Parameter
EasyWAM-Unified	99.0	99.4	99.2	98.2	99.0
EasyWAM-MoT	97.8	98.4	97.6	95.6	97.4
EasyWAM-Hidden	99.4	100.0	97.0	97.8	98.6
LoRA (Rank 128)
🔥 EasyWAM-Unified	84.0	97.8	92.0	81.2	88.8
EasyWAM-MoT	96.8	98.8	94.4	90.4	95.1
EasyWAM-Hidden	96.8	99.4	92.6	86.8	93.9

Success rate (%) under the LIBERO-Plus evaluation protocol:

Model	Background	Camera	Language	Layout	Light	Noise	Robot	Avg.
EasyWAM-Unified	55.8	33.7	93.7	80.6	92.2	50.2	71.4	67.5
EasyWAM-MoT	52.8	20.6	80.4	65.2	85.1	51.5	49.7	56.8
EasyWAM-Hidden	56.8	49.2	95.3	81.0	90.4	58.2	77.4	72.4

Download

hf download OpenMOSS-Team/EasyWAM-Unified-Wan22-LoRA-128 \
  easywam_unified_wan22_lora_128.pt \
  --local-dir ./checkpoints

Evaluation

Prepare Wan2.2, LIBERO, and the matching dataset_stats.json as described in the EasyWAM LIBERO guide, then run:

python experiments/libero/run_libero_manager.py \
  task=libero_easywam_unified_wan22_lora \
  ckpt=./checkpoints/easywam_unified_wan22_lora_128.pt \
  EVALUATION.dataset_stats_path=<path-to-matching-dataset_stats.json>

EasyWAM creates the matching LoRA modules from the task config and merges them for evaluation when loading this checkpoint.

Checkpoint Details

• Architecture: EasyWAM-Unified
• Backbone: Wan2.2-TI2V-5B
• Training: LoRA, rank 128, alpha 128
• Dataset: LIBERO, two cameras at 224 px
• Training steps: 20,000
• Action dimension: 7
• State dimension: 8
• Format: EasyWAM PyTorch checkpoint (.pt)

License and Citation

EasyWAM code is released under the MIT License. Use of this checkpoint is also subject to the terms of its base model and training data. If EasyWAM is useful in your research, please cite:

@misc{easywam2026,
  title  = {EasyWAM: A Unified and Efficient Framework for Training and Evaluating World Action Models},
  author = {EasyWAM-Team},
  year   = {2026},
  url    = {https://github.com/OpenMOSS/EasyWAM}
}