Skip to content

lora_stub

diapason.learning.spec_search.execute.appliers.lora_stub

LoRA fine-tuning stub — deferred to v2.

The planner can emit LORA_FINETUNE edits so the diagnosis surfaces "this should be a weight update" pressure to the user, but the executor refuses them in v1.

See spec §4.1.

LoraStubApplier

Bases: EditApplier

Refuses LORA_FINETUNE with a clear v2 message.