# ludwig-ai/ludwig

Low-code framework for building custom LLMs, neural networks, and other AI models

Repository: https://github.com/ludwig-ai/ludwig
Canonical: https://ross.abutalabs.com/products/ludwig
Homepage: http://ludwig.ai
Language: Python
License: Apache-2.0
License Family: permissive
Topics: deep-learning, deeplearning, deep, learning, machine-learning, machinelearning, natural-language-processing, natural-language, computer-vision, data-centric, data-science, pytorch, neural-network, ml, llm, llm-training, fine-tuning, llama, mistral, llama2
Last push: 2026-08-24T21:24:13+00:00

## Health v2 (maintenance only)
Score: 99/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 98, longevity 100
- inputs: {"age_days": 2806, "days_push": 9, "days_rel": 17, "gap_med": 3.0, "n_releases_24m": 23}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 11745, forks 1217 (observed 2026-08-28T04:10:49.640514+00:00)

## What it is
Ludwig is a declarative, low-code deep learning framework for training, fine-tuning, and deploying AI models — from LLMs to tabular, image, audio, and time-series models — using YAML configuration files instead of boilerplate Python. Built on PyTorch with Ray-based distributed training, it supports PEFT/LoRA fine-tuning, multi-task and multimodal learning, and production export to TorchScript and Triton.

## Use cases
- fine-tune llama or mistral on my own data with lora
- train a classifier from a csv without writing training code
- build custom llms with a yaml config
- fine-tune a vision-language model like llava
- run distributed deep learning training on ray or kubernetes
- do hyperparameter optimization for a neural network
- train a model on tabular, text, and image features together
- export a trained model to torchscript or triton for serving

## When to choose
- you want to train or fine-tune models without writing training loops
- you need LLM fine-tuning with PEFT adapters like LoRA or QLoRA
- you want declarative, reproducible ML pipelines validated by config schemas
- you need multi-task or multimodal models mixing tabular, text, image, and audio features
- you want to scale training across GPUs with Ray, DDP, or DeepSpeed

## When to avoid
- you need full low-level control over every training detail and prefer raw PyTorch
- you want a lightweight inference-only library rather than a training framework
- your project requires Python versions below 3.12
- you need a no-code GUI tool rather than config-driven workflows

## Facets
- artifact type: framework
- maturity: active
- function: machine-learning, deep-learning, llm-training, llm-inference, nlp, computer-vision, data-science, cli
- domain: machine-learning, deep-learning, large-language-models, computer-vision, data-science
- platform: python, cli
- tags: declarative-ml, yaml-config, fine-tuning, lora, peft, qlora, pytorch, ray, distributed-training, multimodal, low-code, huggingface, natural-language-processing, linux, macos, docker, kubernetes, gpu

## Member repositories
- ludwig-ai/ludwig (main) score 99

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:10:49.640514+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:15:21.407552+00:00, confidence not recorded.
  - readme: https://github.com/ludwig-ai/ludwig (fetched 2026-08-28T04:10:49.640514+00:00, sha 046a073f7e12)
  - homepage: http://ludwig.ai (fetched 2026-08-29T08:13:33.017713+00:00, sha e80e988ff4d4)
  - registry_pypi: https://pypi.org/pypi/ludwig/json (fetched 2026-08-29T08:13:33.027178+00:00, sha 814ef46a6beb)
  - site_page: https://ludwig.ai/latest (fetched 2026-08-29T08:13:33.021798+00:00, sha ae35d413133d)
  - site_page: https://ludwig.ai/latest/getting_started (fetched 2026-08-29T08:13:33.024904+00:00, sha 1f3acc836d1b)
- Data as of 2026-08-30T08:39:29.467469+00:00.