Pith. sign in

REVIEW 33 cited by

LoRA Learns Less and Forgets Less

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2405.09673 v2 pith:5QFDM4HM submitted 2024-05-15 cs.LG cs.AIcs.CL

LoRA Learns Less and Forgets Less

classification cs.LG cs.AIcs.CL
keywords lorafinetuningfulllearnslesslow-rankperformanceperturbations
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original abstract

Low-Rank Adaptation (LoRA) is a widely-used parameter-efficient finetuning method for large language models. LoRA saves memory by training only low rank perturbations to selected weight matrices. In this work, we compare the performance of LoRA and full finetuning on two target domains, programming and mathematics. We consider both the instruction finetuning (approximately 100K prompt-response pairs) and continued pretraining (20B unstructured tokens) data regimes. Our results show that, in the standard low-rank settings, LoRA substantially underperforms full finetuning. Nevertheless, LoRA better maintains the base model's performance on tasks outside the target domain. We show that LoRA mitigates forgetting more than common regularization techniques such as weight decay and dropout; it also helps maintain more diverse generations. Finally, we show that full finetuning learns perturbations with a rank that is 10-100X greater than typical LoRA configurations, possibly explaining some of the reported gaps. We conclude by proposing best practices for finetuning with LoRA.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 33 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Transplanting, inverting, and preventing a misalignment persona: method-conditional emergent misalignment in Qwen2.5

    cs.CL 2026-07 conditional novelty 7.0

    Emergent misalignment in Qwen2.5 is mediated by a causal persona direction that low-rank LoRA recruits from covert code while full SFT does not and moves against it.

  2. Beyond LoRA vs. Full Fine-Tuning: Gradient-Guided Optimizer Routing for LLM Adaptation

    cs.CL 2026-05 unverdicted novelty 7.0

    MoLF routes updates between full fine-tuning and LoRA at the optimizer level to match or exceed the better of either static method, with an efficient LoRA-only variant outperforming prior adaptive approaches.

  3. InstructMoLE: Instruction-Guided Mixture of Low-rank Experts for Multi-Conditional Image Generation

    cs.CV 2025-12 unverdicted novelty 7.0

    InstructMoLE replaces per-token routing with instruction-guided global routing for mixture-of-low-rank-experts in diffusion transformers and adds an output-space orthogonality loss to improve multi-conditional image g...

  4. Continual Learning for VLMs: A Survey and Taxonomy Beyond Forgetting

    cs.CV 2025-08 unverdicted novelty 7.0

    The paper offers a comprehensive survey and proposes a new taxonomy for continual learning strategies in VLMs and MLLMs to combat catastrophic forgetting beyond traditional methods.

  5. MemSFT: Mitigating Alignment Tax with an External Parametric Memory

    cs.LG 2026-07 conditional novelty 6.0

    MemSFT attaches a retriever-imitating 8B memory plus a word-level router to frozen Qwen3 backbones, boosting domain scores by ~36 points while holding general-benchmark averages essentially flat, where full SFT loses ...

  6. How Meta-Learning Shapes LoRA Adapter Geometry in Speech Deepfake Detection

    eess.AS 2026-07 conditional novelty 6.0

    Meta-learning training concentrates loss-relevant LoRA updates in query/key projections and spreads them in output projections, relative to standard empirical-risk training.

  7. MoE$^2$-LoRA: When MoE Models Meet MoE-style Low-Rank Adaptation

    cs.CL 2026-07 conditional novelty 6.0

    Routing LoRA adapters with the frozen base router's logits plus a shared cross-layer adapter pool gives the best PEFT accuracy and retention on three MoE backbones.

  8. PoLoRA: A Preconditioned Orthogonalized LoRA Optimizer

    cs.LG 2026-07 conditional novelty 6.0

    A preconditioned, product-aware spectral optimizer for LoRA reaches Adam's final loss in 1.2–1.7x fewer steps on instruction-tuning, with ~3% per-step overhead.

  9. After the Euclidean Highway: Hyperbolic Expert AI as the Next Innovation

    cs.LG 2026-07 conditional novelty 6.0

    Hyperbolic losses at the loss layer alone train stably at 18M-sample scale, while the same geometry inside a trainable adapter is claimed to collapse training.

  10. Early Data Exposure Improves Robustness to Subsequent Fine-Tuning

    cs.LG 2026-05 conditional novelty 6.0

    Early mixing of post-training data into pretraining improves retention of acquired capabilities after subsequent fine-tuning in language models.

  11. Beyond LoRA vs. Full Fine-Tuning: Gradient-Guided Optimizer Routing for LLM Adaptation

    cs.CL 2026-05 unverdicted novelty 6.0

    MoLF routes updates between full fine-tuning and LoRA at the optimizer level to match or exceed the better of the two static methods on SQL, medical QA, and counterfactual tasks while an efficient variant outperforms ...

  12. Optimizer-Model Consistency: Full Finetuning with the Same Optimizer as Pretraining Forgets Less

    cs.LG 2026-05 unverdicted novelty 6.0

    Full finetuning with the pretraining optimizer reduces forgetting compared to other optimizers or LoRA while achieving comparable new-task performance.

  13. COMPASS: COntinual Multilingual PEFT with Adaptive Semantic Sampling

    cs.LG 2026-04 unverdicted novelty 6.0

    COMPASS uses semantic clustering on multilingual embeddings to select auxiliary data for PEFT adapters, outperforming linguistic-similarity baselines on multilingual benchmarks while supporting continual adaptation.

  14. TLoRA: Task-aware Low Rank Adaptation of Large Language Models

    cs.CL 2026-04 unverdicted novelty 6.0

    TLoRA jointly optimizes LoRA initialization via task-data SVD and sensitivity-driven rank allocation, delivering stronger results than standard LoRA across NLU, reasoning, math, code, and chat tasks while using fewer ...

  15. Do Transformers Use their Depth Adaptively? Evidence from a Relational Reasoning Task

    cs.LG 2026-04 unverdicted novelty 6.0

    Transformers show limited adaptive depth use on relational reasoning, with clearer evidence after finetuning on the task.

  16. Robust Policy Optimization to Prevent Catastrophic Forgetting

    cs.LG 2026-02 unverdicted novelty 6.0

    FRPO applies a max-min robust optimization over KL-bounded policy neighborhoods during RLHF to reduce catastrophic forgetting of safety and accuracy under subsequent SFT or RL fine-tuning.

  17. CR-Net: Scaling Parameter-Efficient Training with Cross-Layer Low-Rank Structure

    cs.LG 2025-09 unverdicted novelty 6.0

    CR-Net uses cross-layer low-rank residuals in a dual-path network plus specialized recomputation to outperform prior low-rank methods on 60M-7B model pre-training while using less compute and memory.

  18. Little by Little: Continual Learning via Incremental Mixture of Rank-1 Associative Memory Experts

    cs.LG 2025-06 unverdicted novelty 6.0

    MoRAM frames continual learning as incremental addition of rank-1 adapters viewed as self-activating key-value associative memory units in a mixture-of-experts setup.

  19. LoRA-Mixer: Coordinate Modular LoRA Experts Through Serial Attention Routing

    cs.LG 2025-06 unverdicted novelty 6.0

    LoRA-Mixer routes modular LoRA experts into attention projection matrices with an adaptive Routing Specialization Loss to improve multi-task performance while using fewer trainable parameters than prior LoRA-MoE methods.

  20. Scaling Point-in-Time Language Models

    cs.CL 2026-04 conditional novelty 5.5

    Scaling point-in-time decoder-only LMs to 4B parameters on 1T chronologically filtered FineWeb tokens largely closes the quality gap to unrestricted open models while preserving temporal validity.

  21. One Student, Many Teachers: Multi-Task On-Policy Distillation via Soft-Prompt Privileged Context

    cs.LG 2026-06 conditional novelty 5.0

    Soft-prompt teachers, each a small set of learnable tokens on a frozen backbone, can replace full fine-tuning as on-policy distillation supervisors and compose across tasks into a single student.

  22. SOS-LoRA: Static Orthogonal-Subspace Low-Rank Adaptation with Fixed Multi-Scale Scaling

    cs.LG 2026-06 conditional novelty 5.0

    A LoRA update split into several fixed, differently-scaled low-rank experts with orthogonal input directions improves fine-tuning accuracy at the same parameter count.

  23. On the Scaling of PEFT: Towards Million Personal Models of Trillion Parameters

    cs.LG 2026-06 unverdicted novelty 5.0

    PEFT adapters are positioned as persistent personal state on foundation models, organized via Scale Up, Scale Down, and Scale Out axes, with MinT as an infrastructure example for managing them.

  24. Mask the Target: A Plug-and-Play Regularizer Against LoRA Forgetting

    cs.CL 2026-05 unverdicted novelty 5.0

    A plug-and-play KL regularizer that masks the target token and renormalizes probabilities to improve the learning-forgetting trade-off in LoRA adaptation of LLMs.

  25. Fine-Tuning Without Forgetting via Loss-Adaptive Learning Rates

    cs.LG 2026-05 unverdicted novelty 5.0

    FINCH is a loss-adaptive learning-rate schedule that reduces forgetting by 93% on average during LLM fine-tuning while matching standard task performance across several benchmarks.

  26. Pion: A Spectrum-Preserving Optimizer via Orthogonal Equivalence Transformation

    cs.LG 2026-05 unverdicted novelty 5.0

    Pion is an optimizer that preserves the singular values of weight matrices in LLM training by applying orthogonal equivalence transformations.

  27. Scaling Point-in-Time Language Models

    cs.CL 2026-04 conditional novelty 5.0

    Scaling point-in-time LLMs to 4B parameters and 1T temporally filtered tokens narrows the gap to unrestricted models to about 8–11 average points and yields positive out-of-sample Sharpe ratios from news embeddings.

  28. A Language-Guided Bayesian Optimization for Efficient LoRA Hyperparameter Search

    cs.CL 2026-01 conditional novelty 5.0

    LLM embeddings plus Bayesian optimization find better LoRA hyperparameters in ~30 proxy trials than standard published settings.

  29. CLaaS: Continual learning as a service for sample efficient online learning

    cs.LG 2026-06 unverdicted novelty 4.0

    CLaaS enables sample-efficient online continual learning for agents via replay-buffered parametric updates, outperforming in-context learning in forward transfer and retention on an adversarial task.

  30. LLiMba: Sardinian on a Single GPU -- Adapting a 3B Language Model to a Vanishing Romance Language

    cs.CL 2026-05 conditional novelty 4.0

    Qwen2.5-3B was continued-pretrained and then fine-tuned with rsLoRA r256 on Sardinian data to reach 28.5 BLEU into the language, outperforming full fine-tuning and other LoRA variants.

  31. ScatterPrism: convergence for generative simulation and inverse problems in particle and nuclear physics

    cs.LG 2026-04 unverdicted novelty 4.0

    Standard Conditional Flow Matching loss is a misleading early plateau; physics-informed metrics keep improving, so ScatterPrism and multi-metric diagnostics are needed for kinematic fidelity.

  32. ScatterPrism: convergence for generative simulation and inverse problems in particle and nuclear physics

    cs.LG 2026-04 unverdicted novelty 4.0

    CFM training loss plateaus while physics-informed metrics continue improving; ScatterPrism and a multi-metric protocol are proposed to restore kinematic fidelity without memorization.

  33. Lifelong In-Context Learning with Transformers Requires Parametric Forms of Attention

    cs.LG 2026-06 unverdicted novelty 3.0

    Argues that parametric attention forms are necessary for lifelong in-context learning in transformers to maintain constant memory footprint over arbitrary sequence lengths.