Reproducible recipe

Parameter-efficient fine-tuning variants

Trade adapter capacity, initialization, and storage for one narrow fine-tuning target.. Plain LoRA has a measured geometry, capacity, quantization, or many-adapter limitation.

Status
validated with caveat
Learning path
post training
Regression slices
5
Execution
Fresh experiment required

Target

Trade adapter capacity, initialization, and storage for one narrow fine-tuning target.

Failure this recipe addresses

Plain LoRA has a measured geometry, capacity, quantization, or many-adapter limitation.

Data contract

One frozen target with the same split, steps, seed, and module set across variants.

Method or policy

Compare DoRA, QLoRA, VeRA, AdaLoRA, RsLoRA, PISSA, or LoftQ one axis at a time.

Evaluation contract

Target score and effective update versus plain LoRA.

  • load parity
  • memory
  • trainable parameters
  • breadth
  • wall time

Budget and stop rule

Smokes are retained; any comparative run is a fresh controlled experiment.

Stop when the variant does not solve the diagnosed plain-LoRA limitation.

Decision rule

Keep plain LoRA unless a variant earns a measured target or systems advantage.

Learning exercise

Create a one-page comparison of LoRA, DoRA, QLoRA, and VeRA parameter/memory assumptions.

Select a PEFT variant from the bottleneck, not from its name.

How to read this dossier

This canonical page separates the retained repository decision from the material that informed it. A source may describe an external claim, historical measurement, local observation, or planned procedure; those evidence classes are not interchangeable. Follow the provenance links before reusing a number or method.

Status and disposition describe what PostTrainLLM retained when this record was indexed. They are not a live product promise, a newly run benchmark, or permission to restart historical work. Unknown or unmeasured fields remain unknown rather than being treated as zero.

To reuse the record, first name the exact claim you need and trace it to the linked source. Then check whether the original environment, model revision, data split, evaluator, hardware, and budget match the proposed use. If they do not, treat the record as a hypothesis or design reference and run the smallest fresh comparison that can falsify it. Preserve negative outcomes and regressions beside any improvement; a local win on one slice does not silently become a general capability claim.

Source provenance

The normalized record comes from docs/recipes/registry.json. The links below are the tracked evidence and explanatory sources preserved with the record.