YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
tvshift WAM-IDM (Wan2.2 init, step 5000)
LingBot-VA reimplemented as a dual-stream MoT (video 4.91B + action 0.73B, idm attention layout), initialised from Wan2.2-TI2V-5B and trained 5000 steps on LIBERO "tvshift" (camera angle x task pairs).
lingbot-va-tvshift-mot-idm-code.zip: code, configs, scripts, implementation notestransformer/: step-5000 weights (diffusion_pytorch_model.safetensors, bf16) + config. VAE / tokenizer / text encoder come from lingbot-va-base.eval_summary_wan22_step5000.json: per-combination results (10 episodes each)- Data: https://huggingface.co/datasets/dinyu/tvshift-libero-lerobot
Training angles: 0deg Spatial idx 1,3 / 45deg Goal idx 0,7 / 270deg Object idx 0,1 / 345deg Goal idx 2,5. Results: seen 62/80 (77.5%), novel 132/240 (55.0%).
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support