lerobot/pi05_libero_base
Partial fine-tune vs lerobot/pi05_base. Partial change: 84% of the weights changed (attention, embeddings, lm_head, mlp, norms, other), in 27 of 27 layers.
Tensors changed: other 6/6, lm_head 1/2, Norms 221/221, MLP 217/220, Attention 359/360, Embeddings 3/3
3 derivatives checked
- lerobot/pi05_libero_finetuned_v044: Partial fine-tune
- lerobot/pi05_libero_finetuned_quantiles_v044: Partial fine-tune
- vrfai/pi05-libero-gguf: Not comparable yet
5 other repos hold byte-for-byte identical weight files.