artidoro/qlora

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

10998 stars · 876 forks observed · 2026-08-28

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

QLoRA is the official implementation of the QLoRA paper, an efficient finetuning approach that backpropagates through a frozen 4-bit quantized LLM into Low Rank Adapters, enabling finetuning of a 65B model on a single 48GB GPU. It integrates bitsandbytes quantization with Hugging Face PEFT and transformers, and includes the Guanaco model family and training scripts.

Use cases

• finetune a large language model on a single GPU
• run 4-bit quantized LoRA finetuning of LLaMA models
• train an instruction-following chatbot cheaply
• reproduce the Guanaco models from the QLoRA paper
• reduce memory usage when finetuning 33B or 65B parameter models
• create LoRA adapter weights for a quantized LLM

When to choose

• you need to finetune very large LLMs with limited GPU memory
• you want the reference implementation of the QLoRA/NF4 method
• you want to train LoRA adapters on quantized models with bitsandbytes and PEFT

When to avoid

• you need a maintained, general-purpose finetuning framework (the repo is a research artifact with infrequent updates)
• you want full 16-bit finetuning or non-LoRA training methods
• you need multi-GPU or production training pipelines beyond the provided scripts

Facets

library · maturity maintenance

llm-training machine-learning deep-learning large-language-models machine-learning deep-learning python qlora lora quantization 4-bit finetuning bitsandbytes peft guanaco instruction-tuning gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="artidoro/qlora")