Getting Started
Start with WavesFM
A gentle introduction to the WavesFM site and codebase.
A Gentle Start
Follow these steps in order. Each one explains what the page is for and what to do next.
Quickstart notebook: a zero-setup run
Use the demo notebook for a zero-setup run. Open it, run the cells, and you will get a working WavesFM result in a few minutes without changing anything. It is the fastest way to run before you move to reproducing our results or doing your own experiments.
Benchmarks: pick a target
Start on the Benchmarks page to see which datasets and tasks WavesFM supports and how performance changes across fine-tuning regimes (LP, FT2, LoRA). Pick one dataset you care about, then use it as your "home base" for the rest of the docs.
Datasets: understand the protocol
Before running code, open the matching dataset page. This is the "spec sheet" for downloads, train-test splits, preprocessing steps, and evaluation metrics.
Reproduce: run the baseline recipes
Use the Reproduce page to run the exact recipes (set of commands) that generate the reported numbers. This is the fastest way to validate your environment and establish a baseline before you start making changes.
Detailed Evaluation: go beyond one table cell
If you want to go beyond a single metric (like accuracy or error) and understand how WavesFM behaves in depth, use Detailed Evaluation. It gives you per-metric and per-regime views so you can inspect performance more thoroughly for each dataset and task.
Tutorials: extend WavesFM
Once you have a baseline, move to Tutorials to add a new task, plug in a new dataset, or customize training/evaluation. This is where you start making WavesFM your own.