Season 6 · Weeks 20–23 · 4 weeks
Season 6: Home.
FRIDAY moves into your laptop: a private model that runs with no internet, is fast, learns your habits and talks out loud.
FRIDAY moves into your laptop: an open model that runs with no internet, numbers squeezed to fit, honest benchmarks, a router that sends easy questions to the small model, and a small model taught your format (Project 6). Then give it a voice you can interrupt.
How you learn
Floats and quantization by hand, the price of a million tokens, a LoRA layer and a latency budget.
What you build
Private mode, a router with fallbacks, a small model taught your format, and a voice you can interrupt.
The weeks
- Week 20Bring it home
You can run an open model on your laptop, explain how bits and floats store numbers, quantize and benchmark it honestly, and check its licence.
- Week 21Fast and cheap
You can explain batching and the KV and prompt caches, price a million tokens, route easy questions to a small model, and fall back when a model is down.
- Week 22Teach it your ways
You can decide when fine-tuning is worth it, explain LoRA, fine-tune a small model on a free GPU, and prove with an eval that it learned your format.
- Week 23Give it a voice
You can build a voice loop that answers fast and lets you interrupt, keep a computer-use agent inside four walls, check whether a model fits a device, and distil a big model into a small one.
Projects
- Project 6: Teach a small model
A small open model fine-tuned with LoRA to answer in your format, with an eval that proves it. Weeks 20–22.
Skills you prove
Tiles on the periodic table of skills that this season lights up.
- FLOPs math
- Compute parameter count, FLOPs and KV-cache memory from a model config (week 21)
- Decide
- Choose between prompt, RAG, fine-tune and agent, and defend it with numbers (week 22)
- Distill
- Run LoRA SFT on a small model with data from your own traces, distill a frontier model into it, and explain every curve (week 22)
- Don't tune
- Know when fine-tuning is the wrong answer (week 22)
- Serving math
- Do serving math: batching, quantization, prefix caching trade-offs (week 21)
Checkpoint test: Home
After week 23. FRIDAY runs offline on your laptop and answers out loud.
- FRIDAY answers with the network off, with a speed and quality benchmark you measured
- Blank file, 45 minutes: a LoRA layer with a merged-weights test
- Your fine-tuned small model beats prompting alone on your eval, with confidence intervals
- Explain to a friend when fine-tuning is the wrong answer