ludwig
Low-code framework for building custom LLMs, neural networks, and other AI models
What is ludwig?
Low-code framework for building custom LLMs, neural networks, and other AI models
Overview of ludwig
What it is
Ludwig is a declarative deep learning framework for LLMs, multimodal models, and tabular AI. It lets you train, fine-tune, and deploy AI models using a YAML config file and zero boilerplate Python. The tech stack includes Python 3.12, PyTorch 2.7+, Pydantic 2, Transformers 5, and Ray 2.54. Ludwig is hosted by the Linux Foundation AI & Data.
Key features
- Declarative YAML configuration for training, fine-tuning, and deploying AI models - Supports LLM fine-tuning (SFT, DPO, KTO, ORPO, GRPO) with various PEFT adapters (LoRA, DoRA, VeRA, PiSSA, etc.) and quantization (bitsandbytes, torchao)
- Multimodal and tabular models: input modalities include text, numbers, categories, images, audio, timeseries, vectors, and dates - Timeseries forecasting with PatchTST and N-BEATS encoders, MASE/sMAPE metrics, and a
model.forecast()API - Multi-task learning with Nash-MTL and Pareto-MTL loss balancing - Image segmentation decoders (UNet, SegFormer, FPN) for semantic segmentation
Use cases
- Fine-tune an LLM for instruction tuning (e.g., Llama-3.1 with LoRA)
- Train a multimodal classifier combining text, numeric, and image inputs - Generate a model configuration from natural language using
ludwig generate_config - Make predictions on new data with
ludwig predict - Launch a REST API for model serving with
ludwig serve - Train tabular classifiers, timeseries forecasters, and multi-task models
Getting started
bash pip install ludwig # core pip install ludwig[full] # all optional dependencies pip install ludwig[llm] # LLM fine-tuning only # Example: fine-tune an LLM ludwig train --config model.yaml --dataset my_data.csv
License and pricing
The license is Apache-2.0. The project is open source.
Based on the project's public README and repository metadata.