Ludwig is a declarative, YAML-config driven framework for training, fine-tuning, and deploying AI models, including LLMs, multimodal, and tabular tasks. It supports PEFT, quantization, distributed training, and multiple deployment options.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Ludwig is a declarative deep learning framework that lets you train, fine-tune, and deploy AI models — from LLM fine-tuning to tabular classification — using a YAML config file and zero boilerplate Python.
How it works
The project provides capabilities across LLM fine-tuning, multimodal and tabular models, training infrastructure, hyperparameter optimization, and deployment tooling. It supports input modalities (text, numbers, images, etc.), multiple encoders and decoders, PEFT adapters (e.g., LoRA, PiSSA, EVA), quantization, and distributed training with Accelerate, Ray, and other backends. It includes REST API deployment via FastAPI, vLLM serving, Ray Serve, and KServe for deployment, plus model export options (SafeTensors, .pt2 bundles, ONNX).
Getting started
Installation is via pip with optional extras:
pip install ludwig # core
pip install ludwig[full] # all optional dependencies
pip install ludwig[llm] # LLM fine-tuning only
Requires Python 3.12+. Then you can train with a YAML config, for example:
model_type: llm
base_model: meta-llama/Llama-3.1-8B
quantization:
bits: 4
adapter:
type: lora
prompt:
template: |
### Instruction: {instruction}
### Input: {input}
### Response:
input_features:
- name: prompt
type: text
output_features:
- name: output
type: text
trainer:
type: finetune
learning_rate: 0.0001
batch_size: 1
gradient_accumulation_steps: 16
epochs: 3
learning_rate_scheduler:
decay: cosine
warmup_fraction: 0.01
backend:
type: local
Then run:
export HUGGING_FACE_HUB_TOKEN="<your_token>"
ludwig train --config model.yaml --dataset "ludwig://alpaca"
Recent releases
- v0.17.8 (2026-07-27): path traversal fix in dataset archive extraction; security fix.
- v0.17.7 (2026-07-04): Bug fixes (Arrow/Ray compatibility).
- v0.17.6 (2026-06-26): Preprocessing progress callback introduced (live updates).
- v0.17.5 (2026-05-29): Bug fix: GPU Docker images install CUDA build of PyTorch.
- v0.17.3 (2026-05-24): Bug fixes including LLM fine-tuning crash with torchao>=0.17.
Traction
Stars: 11746; Forks: 1216; Open issues: 2
Behind the repo
Not provided in the provided material.
Caveats
License: Apache-2.0. Requires Python 3.12+. Release notes mention security fixes and bug fixes related to dependencies and GPU builds.