The Five Minute Course
Learn Weights & Biases in five minutes.
Weights & Biases tracks machine learning experiments and, with Weave, traces and evaluates LLM applications. Once an AI system matters, you need to know whether a change made it better. Your five minutes go to logging a run, tracing a call, and reading an evaluation honestly.
Plate I The lesson, live
Open full screen ↗I. The idea
Runs, traces, and evals
Three objects cover most of it: a run is one training or experiment, a trace is one call through an LLM app, and an eval is a set of examples scored the same way every time.
One experiment, logged
Metrics, config, and artifacts for one attempt. Compare runs side by side and the chart shows whether the change helped.
What the app actually did
Weave records each call in an LLM app: inputs, outputs, latency, cost, and nested steps. When an answer is wrong, the trace shows where.
Same test, every version
A dataset of examples and a scorer. Run it on each version and the number moves for a reason you can inspect, not a vibe.
II. The practice
What you will do in the lesson
You answer and try things. Nobody demonstrates at you for five minutes.
- Log a run. You record a config and a few metrics and see them appear in a project.
- Compare two runs. You put two runs on one chart and say which is better and on what.
- Trace one call. You inspect a traced LLM call and find the step where the output went wrong.
- Build a tiny eval. You write five examples with expected answers and score a model on them.
- Judge an improvement. You end with a change that scores higher and decide whether the gain is real or the eval is too small.
III. Notes
Before you start
Is Weights & Biases free?
Free for personal use with generous limits; teams pay. Weave, the LLM tracing product, has a free tier as well.
I am not training models. Is this for me?
If you build anything on an LLM, yes: traces and evals are how you know a prompt change did not break something.
How is it different from PostHog?
PostHog measures users in a product. Weights & Biases measures the model and the AI pipeline behind it. Serious teams use both.
Does it need code changes?
A few lines: initialise a run, log metrics, or decorate a function to trace it. The lesson shows the shape without requiring an install.
IV. Further reading
Other tools people learn next
Same format, same five minutes, still free.
Learners from these organizations are already learning on 
