LyCORIS Documentation

Start here. Every document in this tree is listed below with what it answers, so you can jump straight to the one you need.

I want to …

Goal	Read
Pick an algorithm	algorithms/README.md, then algorithms/guidelines.md
Understand the math behind an algorithm	algorithms/details.md
Configure a training run	usage/network-args.md and usage/presets.md
Call LyCORIS from my own code	api/README.md
Make training and merging faster	kernels/README.md
Know which backend is running	kernels/backends.md
Check numerical accuracy	kernels/precision.md
Reproduce the speed numbers	kernels/benchmarks.md
Convert between formats	usage/conversion-scripts.md
Work on LyCORIS itself, or cut a release	development/README.md
See sample results	resources/demo.md
Learn more about fine-tuning in general	resources/README.md

Tree

docs/
├── algorithms/     what LyCORIS implements
│   ├── README.md   list of implemented algorithms, with shapes and parameters
│   ├── details.md  the math of each algorithm
│   └── guidelines.md  which algorithm to pick, and with what hyperparameters
├── usage/          how to drive it
│   ├── network-args.md       every network argument, per algorithm
│   ├── presets.md            preset files and layer selection
│   └── conversion-scripts.md format conversion tools
├── api/
│   └── README.md   module, functional and wrapper APIs
├── kernels/        the fused kernel backends (experimental)
│   ├── README.md      what is fused, what it buys, how to switch it off
│   ├── backends.md    selection order, per-op scope, environment variables
│   ├── precision.md   dtype policy and measured accuracy
│   └── benchmarks.md  the measurement harness and current results
├── development/
│   └── README.md   checks, versioning, release channels, workflows
└── resources/
    ├── README.md   external reading on fine-tuning
    └── demo.md     example outputs

Source map

Path	Contents
lycoris/modules/	one nn.Module wrapper per algorithm, plus base.py
lycoris/functional/	the same algorithms as plain functions (weight_gen, diff_weight, bypass_forward_diff)
lycoris/kernels/	fused Triton and TileLang kernels, the autograd layer, and backend selection
lycoris/kernels/plans/	the analytic tile planner and the measured tuner
lycoris/wrapper.py	LycorisNetwork, create_lycoris
lycoris/kohya.py	the kohya-ss/sd-scripts entry point
scripts/bench/kernels/	the kernel benchmark suite (measure scripts write JSON, plot scripts draw it)
scripts/ci/	the checks and version helpers the workflows call
test/	functional, module, wrapper and kernel test suites