LLM Fine-Tuning
A hands-on course covering LLM internals, fine-tuning techniques, on-device deployment for iOS, and production observability. Written for engineers who want to go beyond API calls and understand what’s actually happening under the hood.
Prerequisites
- Comfortable with Python and basic linear algebra (vectors, matrices, dot products)
- A MacBook Pro with Apple Silicon (M1 Pro or later recommended; 16 GB+ unified memory)
- Familiarity with PyTorch is helpful but not required, we introduce what’s needed
Course Outline
Tokens, embeddings, attention, transformers: how language models actually work under the hood.
Model families, parameter counts, quantization, LoRA, QLoRA: the conceptual toolkit for fine-tuning.
End-to-end LoRA fine-tuning with MLX on Apple Silicon. Generic exercise first, then a QuizMe answer-grading model.
Converting fine-tuned models to CoreML and MLX for local inference on Apple Silicon.
Langfuse, LangSmith, LLM-as-judge, evaluation pipelines, and production observability.