Living systematic review

Parameter-efficient fine-tuning (LoRA family)

Adapting large models by training a small number of added/low-rank parameters instead of full weights.

721 papers1,563 critique receipts6,581 benchmark resultsupdated Jun 18, 2026

Most-superseded baselines

Ranked by how many distinct papers critique or beat each method — the standard baselines newer work routinely measures against.

  1. 1
    LoRA

    LoRA: Low-Rank Adaptation of Large Language Models

    194 critique · 201 beaten on benchmarks

  2. 2
    AdaLoRAin LoRA

    AdaLoRA: Adaptive Budget Allocation for Parameter-Efficient Fine-Tuning

    29 critique · 54 beaten on benchmarks

  3. 3
    DoRAin LoRA

    DoRA: Weight-Decomposed Low-Rank Adaptation

    16 critique · 66 beaten on benchmarks

  4. 4
    PiSSAin LoRA

    PiSSA: Principal Singular Values and Singular Vectors Adaptation of Large Language Models

    14 critique · 30 beaten on benchmarks

  5. 5
    BitFitin LoRA

    BitFit: Simple Parameter-efficient Fine-tuning for Transformer-based Masked Language-models

    7 critique · 28 beaten on benchmarks

  6. 6
    Adapterin LoRA

    Adapters: A Unified Library for Parameter-Efficient and Modular Transfer Learning

    10 critique · 23 beaten on benchmarks

  7. 7
    FFA-LoRA

    9 critique · 17 beaten on benchmarks

  8. 8
    MoELoRAin LoRA

    MoELoRA: Contrastive Learning Guided Mixture of Experts on Parameter-Efficient Fine-Tuning for Large Language Models

    11 critique · 10 beaten on benchmarks

  9. 9
    QLoRAin LoRA

    QLoRA: Efficient Finetuning of Quantized LLMs

    5 critique · 14 beaten on benchmarks

  10. 10
    Prefix-Tuningin LoRA

    Prefix-Tuning: Optimizing Continuous Prompts for Generation

    7 critique · 8 beaten on benchmarks

  11. 11
    Prompt Tuningin LoRA

    9 critique · 6 beaten on benchmarks

  12. 12
    LoRA+in LoRA

    LoRA: Low-Rank Adaptation of Large Language Models

    2 critique · 12 beaten on benchmarks

The competition

Methods that fight on the same benchmarks cluster into distinct sub-problems.

LoRA447 methods

LoRA · AdaLoRA · DoRA · PiSSA · BitFit · Adapter

InfLoRA133 methods

InfLoRA · O-LoRA · ZipLoRA · L2P · B-LoRA · DB-LoRA

FFA-LoRA76 methods

FFA-LoRA · FedIT · FlexLoRA · HetLoRA · FedEx-LoRA · FedSA

BLoB31 methods

BLoB · MAP · LAP · Random Forest · BLOB · FLare

CNN30 methods

CNN · AIDE · Effort · LFT · frequency-based detectors · domain-specific fine-tuning

MoLoRA25 methods

MoLoRA · ABot-M0 · AdaMoLE · CogVLA · Dita · FLOWER

CLIP-Adapter19 methods

CLIP-Adapter · CoOp · TaskRes · MaPLe · MMRL · PMC-CLIP

MC Dropout18 methods

MC Dropout · SNGP · Laplace-LoRA · SWAG-LoRA · Batch-Ensemble · SWAG

LISA19 methods

LISA · Cascade R-CNN · greedy layer selection (kaplun2023less) · LIFT · raw gradient norm · Sequential adapters

TimeSformer17 methods

TimeSformer · SAFE · MTLoRA · ORAA-net · SVFormer (Rizvi et al. 2024) · ZSC-CO

The frontier

Recent methods not yet superseded in the knowledge base.