Module 12: Foundation Models
1 min read·
What you need first
- The message intents (Module 4 lesson 3), quantisation and distillation (Module 8 lesson 1), KV caching, batching and the benchmark harness (Module 8 lessons 2 and 3). The lab's Step 1 rebuilds Modules 1 to 11.
- A CPU (a T4 is optional in lesson 1); the lab downloads SmolLM2-135M-Instruct (257 MiB). Hosted paths go through
ask(), checked with scripted fakes; a key is optional. Ollama and vLLM are local, not run. - If you already fine-tune with LoRA and gate models on a task eval set, go to the lab.