# microsoft/LoRA

Code for loralib, an implementation of "LoRA: Low-Rank Adaptation of Large Language Models"

Repository: https://github.com/microsoft/LoRA
Canonical: https://ross.abutalabs.com/products/lora
Homepage: https://arxiv.org/abs/2106.09685
Language: Python
License: MIT
License Family: permissive
Topics: gpt-2, adaptation, language-model, gpt-3, low-rank, pytorch, deep-learning, roberta, deberta, lora
Last push: 2024-12-17T03:55:52+00:00

## Health v2 (maintenance only)
Score: 23/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 8, longevity 100
- inputs: {"age_days": 1903, "days_push": 624, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 13767, forks 923 (observed 2026-08-28T04:11:04.754835+00:00)

## What it is
loralib is the official PyTorch implementation of LoRA (Low-Rank Adaptation), which fine-tunes large language models by injecting trainable rank-decomposition matrices while freezing original weights. It drastically reduces trainable parameters and storage for task adaptation without adding inference latency.

## Use cases
- fine-tune large language models with minimal trainable parameters
- adapt GPT-2, GPT-3, RoBERTa, or DeBERTa to downstream tasks
- reduce GPU memory usage during model fine-tuning
- store many task-specific model adaptations cheaply
- switch between adapted tasks at inference without latency
- integrate LoRA layers into existing PyTorch or Hugging Face models

## When to choose
- you want the original reference implementation of LoRA in PyTorch
- you need parameter-efficient fine-tuning of transformer models
- you must deploy many task adaptations of one large model cheaply
- you want fine-tuning quality comparable to full fine-tuning with a fraction of the parameters

## When to avoid
- you need broad model support and active development - Hugging Face PEFT is now the maintained ecosystem standard
- you are not working with PyTorch
- you need full fine-tuning rather than parameter-efficient adaptation

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, llm-training, deep-learning
- domain: large-language-models, machine-learning, deep-learning
- platform: python
- tags: lora, parameter-efficient-fine-tuning, pytorch, low-rank-adaptation, peft, transformers, fine-tuning, natural-language-processing, gpu

## Member repositories
- microsoft/LoRA (main) score 23

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:11:04.754835+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:13:03.164099+00:00, confidence not recorded.
  - readme: https://github.com/microsoft/LoRA (fetched 2026-08-28T04:11:04.754835+00:00, sha 2255a547ce62)
  - homepage: https://arxiv.org/abs/2106.09685 (fetched 2026-08-29T08:07:23.794100+00:00, sha c197a338f969)
  - site_page: https://info.arxiv.org/about/donate.html (fetched 2026-08-29T08:07:23.803736+00:00, sha cca9c3a11c56)
  - site_page: https://info.arxiv.org/about/ourmembers.html (fetched 2026-08-29T08:07:23.807531+00:00, sha 47cbc55ff1de)
  - site_page: https://info.arxiv.org/about (fetched 2026-08-29T08:07:23.809566+00:00, sha a1f16f915a9a)
  - site_page: https://info.arxiv.org/labs/index.html (fetched 2026-08-29T08:07:23.805764+00:00, sha b14a8d05a0ec)
- Data as of 2026-08-30T08:39:29.467469+00:00.