LLM Models and Serving

Step past calling a model and open it up. Write a transformer yourself, train a small language model from scratch, cut costs with batching, the KV cache and quantization in serving engines, and make models reproducible with experiment tracking.

코스

• Transformers — Compute Attention By Hand — Confirm with numbers why you divide by √d
• MiniMind — Train a Small Language Model Yourself, End to End — Bake a small model from tokenizer to DPO on two CPUs
• LLM Serving — Learn the machinery of serving without a model.
• The AI Diet Gone Wrong — Build a real INT8 model and measure accuracy and performance
• Last Week Had a Better Model. Nobody Can Find It — Build the run ledger, registry and promotion gate by hand