QLoRA
Also known as: Quantized LoRA
LoRA combined with 4-bit base-model quantization, letting you fine-tune huge models on a single consumer GPU.
A fine-tuning method that combines 4-bit quantization of the frozen base model with LoRA adapters. Cuts memory by ~3x vs LoRA, making it feasible to fine-tune 65B-parameter models on a single 48GB GPU.
In practice
How most hobbyists and small teams fine-tune large open models. Expect questions on NF4 vs INT4, double quantization, and quality vs full LoRA.
How it compares
LoRA leaves the base model in FP16; QLoRA additionally quantizes the base model to 4-bit.
Related topics
Questions that mention this term
- Misconception: 'LoRA is just…
- Which statements about QLoRA are true?
- Adam stores two fp32 buffers per trainable parameter: what does that imply for LoRA vs full FT on a 7B?
- Match each 4-bit format to its niche in 2026 inference and fine-tuning.
- Why does QLoRA use…
- What does bnb_4bit_compute_dtype actually control in a QLoRA config?
Related terms
DPO (Direct Preference Optimization)
Skip the reward model and PPO: fine-tune directly on preferred-vs-rejected response pairs.
Fine-tuning
Continue training a pre-trained model on task-specific data to specialize its behavior.
FSDP (Fully Sharded Data Parallel)
Shard weights+grads+optimizer states across GPUs and gather just-in-time; train huge models with limited per-GPU memory.
Instruction Tuning
Train a base model on many (instruction, response) pairs to make it follow natural-language commands.
Knowledge Distillation
Train a small student model to match a big teacher's outputs: cheap, fast inference with most of the quality.
LoRA (Low-Rank Adaptation)
Fine-tune by training small low-rank update matrices instead of every weight: same quality, ~1% of the cost.