Landmark 6 stops to get here · leads to 5

Fine-Tuning

The process of further training a pre-trained model on a specific dataset to adapt it for a particular task or domain.

Your route here

6 stops · basics first

Dataset ✓ understood
A collection of data examples used for training, validating, or testing machine learning models.
Training ✓ understood
The process of fitting a model to data by repeatedly measuring how wrong its outputs are and adjusting its parameters to reduce that error.
Machine Learning ✓ understood
Building systems that learn patterns from data instead of following hand-written rules, getting better at a task as they see more examples.
Unsupervised Learning ✓ understood
Learning from unlabeled data to discover hidden patterns, structures, or relationships without explicit target outputs.
Self-Supervised Learning ✓ understood
Learning representations from unlabeled data by creating supervised tasks from the data itself (masked prediction, contrastive learning).
Pre-training ✓ understood
Training a model on a large dataset (often self-supervised) before fine-tuning on specific tasks, enabling transfer learning.
Fine-Tuning · you are here ✓ understood

Picture it

01 Pre-trained base model General knowledge from broad data
02 Task-specific dataset Smaller, focused examples
03 Continue training Usually a lower learning rate
04 Evaluate on validation Check gains, watch for forgetting
05 Specialized model

Notice that training starts from existing weights, not scratch, so far less data and compute are needed.

Fine-tuning leverages a model’s pre-trained knowledge and adapts it to specific tasks, domains, or behaviors. This is far more efficient than training from scratch.

Process

Start with a pre-trained base model
Prepare task-specific training data
Continue training with a lower learning rate
Evaluate on validation data
Deploy the fine-tuned model

Types

Full Fine-Tuning: Update all model parameters
Parameter-Efficient Fine-Tuning (PEFT): Update only a subset (LoRA, adapters)
Instruction Fine-Tuning: Train on instruction-following data
RLHF: Reinforce learning from human feedback

Benefits

Faster than training from scratch
Requires less data
Improves task-specific performance
Enables domain adaptation

Where it sits

Before this

Fine-Tuning

Leads to

Catastrophic Forgetting Instruction Tuning LoRA RLHF Prompt Tuning

Explore nearby

Training Transfer Learning Leveraging knowledge learned from one task/domain to improve performance on a related task with less data. Training LoRA Low-Rank Adaptation - a parameter-efficient fine-tuning method that updates only small low-rank matrices instead of full weights. Language & LLMs Instruction Tuning Fine-tuning LLMs on diverse instruction-following tasks to improve zero-shot performance on new instructions. Language & LLMs RLHF Reinforcement Learning from Human Feedback - training models using human preferences to align behavior with human values. Training Catastrophic Forgetting The tendency of neural networks to completely forget previously learned information when learning new tasks. Language & LLMs Prompt Tuning Learning continuous prompt embeddings while keeping the LLM frozen, an efficient alternative to fine-tuning.

In the research

7 papers that build on Fine-Tuning ; showing 5, canon first.

Canon · 2018 BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding It established “pre-train once, fine-tune everywhere”, and still powers much of search and classification.

Canon · 2021 LoRA: Low-Rank Adaptation of Large Language Models It made adapting big models cheap enough for everyone, and is why fine-tunes are shared as small adapter files.

Canon · 2022 Training Language Models to Follow Instructions with Human Feedback This recipe, RLHF, is what turned raw language models into assistants like ChatGPT.

Frontier · Sep 2025 · 114 citations Fast-dLLM v2: Efficient Block-Diffusion LLM A cheap path to faster inference from models you already have.

Frontier · Aug 2025 · 112 citations On the Generalization of SFT: A Reinforcement Learning Perspective with Reward Rectification A tiny, theory-backed change to the most common training step in the stack.