Jiahao-Wang/lingbot-va-robotwin-8task-lora-attn-r16-step10000
0
LingBot-VA RoboTwin 8-Task Attention-LoRA Step 10000
This is a LingBot-VA inference bundle post-trained with attention-only LoRA on eight RoboTwin 2.0 tasks:
The model was initialized from the LingBot-VA base checkpoint and trained for 10000 optimizer steps with effective batch size 8. The LoRA adapter used rank 16, alpha 32, and dropout 0.05 on self/cross-attention q/k/v/out projections.
contains merged weights and can be used directly by . is included for analysis and reproducibility.