Reference library
DoRA: Weight-Decomposed Low-Rank Adaptation
Paper summary
DoRA separates weight magnitude and direction, using low-rank updates for the directional component. Comparisons with LoRA and full fine-tuning evaluate adaptation accuracy and stability without additional inference overhead after merging.
Relevance to the project
Offers a structured parameter-update comparison for future model adaptation. The project should independently test any restricted update representation for F.
Fine-tuning success does not demonstrate self-directed updates.
Source record
- Reference ID
- dora
- Archived version
- 2402.09353v6
- Summary provenance
- reference-library/papers/dora/metadata.json · summary
- Review status
- initial orientation summary; not a critical full-paper review
- Reading scope
- Selected abstract/opening and available closing discussion/conclusion passages in the preserved text extract; broader project notes provide context. No reproduction of experiments, exhaustive methods review, or equation/figure audit.