๐ง Fine-tuning
Fine-tuning AI models involves adjusting the parameters of a pre-trained model to perform better on a specific task or dataset. This process allows the model to adapt its learned knowledge to new, related problems, enhancing its accuracy and effectiveness for specialized applications.
21 tools
Hugging Face library for training language models with RLHF, SFT, and DPO
Declarative deep learning framework for building and fine-tuning models with YAML configuration
Open-source toolkit for fine-tuning LLMs with a single YAML config across the full training pipeline
Managed API access to foundation models on AWS with built-in fine-tuning and agent tooling
The open-source AI platform with 500K+ models, inference endpoints, and fine-tuning tools
Run and fine-tune open-source models. Deploy custom models at scale. All with one line of code.
Fine-tuning overview
Fine-tuning platforms enable teams to customize pre-trained language models on their own data, creating specialized models that outperform general-purpose ones for specific tasks. These tools handle the end-to-end workflow: dataset preparation, training job management, hyperparameter tuning, evaluation, and deployment.
Modern fine-tuning approaches like LoRA and QLoRA have dramatically reduced the cost and complexity of model customization. What once required multi-GPU clusters can now run on a single GPU in hours. The platforms listed here make these techniques accessible through managed infrastructure and intuitive interfaces.
Fine-tuning is particularly valuable when you need consistent output formatting, domain-specific knowledge, reduced hallucination rates, or lower inference costs by using a smaller, specialized model instead of a large general-purpose one.
Browse other categories
Related stacks
See how fine-tuning tools fit into a full infrastructure stack.
Frequently Asked Questions
When should I fine-tune instead of using prompt engineering?
Fine-tune when prompt engineering hits its limits: when you need consistent output formatting at scale, domain-specific behavior that few-shot examples cannot achieve, lower latency from a smaller model, or cost reduction by replacing a large model with a specialized smaller one. Start with prompting and move to fine-tuning when you have clear training data and measurable quality gaps.
What is LoRA and why does it matter?
LoRA (Low-Rank Adaptation) is a parameter-efficient fine-tuning technique that trains only a small set of additional weights rather than the full model. This reduces GPU memory requirements by 10-100x and training time significantly, making fine-tuning practical for teams without large compute budgets.
How much data do I need to fine-tune a model?
It depends on the task. For format and style changes, 50-100 high-quality examples can be enough. For domain knowledge, you typically need 500-5,000 examples. Quality matters more than quantity. Poorly labeled or inconsistent training data will degrade model performance regardless of dataset size.
Is your product missing?