Remark. Architecture and training are separable records [ftip-00KS]
✍️source

Weight sharing, latent objectives, and recirculation can alter optimization without changing the declared external interface. The comparison must retain architecture, training intervention, and systems coordinates separately.

Remark. Architecture and training are separable records [ftip-00KS]
✍️source

Depth reuse and latent recurrence [ftip-00KG]

Definition. Computation-axis record [ftip-00KH]

Remark. Looped Transformers as a source result [ftip-00KI]

Remark. Parameter sharing and width tradeoffs [ftip-00KJ]

Remark. Virtual logical depth and scaling [ftip-00KK]

Remark. Recirculation as an inference intervention [ftip-00KL]

Remark. Latent feedback with a preserved Transformer interface [ftip-00KM]

Remark. Latent prediction as a training intervention [ftip-00KN]

Example. Equal parameters do not fix effective depth [ftip-00KO]

Definition. Depth-reuse model-instance record [ftip-00KP]

Example. Fixed-recipe depth comparison [ftip-00KQ]

Remark. Scaling surfaces rather than a single law [ftip-00KR]

Remark. Architecture and training are separable records [ftip-00KS]
✍️source

Remark. Depth reuse across passes, parameters, and tasks [ftip-00KT]