Grimoire
Tree — 03
peft/3

Fine-tuning

Adapting a pretrained model without paying to pretrain it. Mostly a study in which parameters you are allowed to leave frozen, and what the frozen ones cost you in precision.

Three entries. One is a stub and is marked as such — read them as open questions rather than answers.

Entries
03.1
Adapter Mergingstub
Task arithmetic, TIES and DARE over independently trained deltas.
O(d)03.2
LoRA
Low-rank additive update to frozen weight matrices during adaptation.
O(r·d)03.3
QLoRA
4-bit NF4 base weights with paged optimiser state and LoRA adapters.
O(r·d)