Fine-tuning parameters do not match

[JuShuai-gif]

Why does the provided fine-tuning example dataset contain two images with state_dim=8 and action_dim=7, while the provided weights feature three image inputs with state_dim=16 and action_dim=16? Will this discrepancy have an impact? Or can this difference be ignored after freezing the parameters?

[ModiShi]

The state_dim=16 and action_dim=16 configuration comes from our pre-training on Agibot World, while the fine-tuning example uses a different setting from Libero. The model will automatically adapt the relevant parameters to match your new task dimensions during fine-tuning.