microsoft/LoRA

Health v2 · maintenance only

Adoption not part of the score

13767 stars · 923 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

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

library · maturity maintenance

machine-learning llm-training deep-learning large-language-models machine-learning deep-learning python lora parameter-efficient-fine-tuning pytorch low-rank-adaptation peft transformers fine-tuning natural-language-processing gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="microsoft/LoRA")