Wiring Language Models

Create a free account to open this chapter.

• Characters, words, and subwords ~16 min
• Byte-pair encoding by hand ~24 min
• Vocabulary, special tokens, and the cost of a token ~18 min

Create a free account to open this chapter.

• Word, sentence, and document embeddings ~18 min
• Measuring similarity: cosine vs. Euclidean ~20 min
• Dense retrieval: why embeddings power search ~18 min

Create a free account to open this chapter.

• Logits, softmax, and the next-token distribution ~16 min
• Temperature, top-k, and top-p ~18 min
• Context window and KV cache ~15 min

Create a free account to open this chapter.

• The pretraining objective, at scale ~16 min
• Data, compute, and scaling laws ~17 min
• What pretraining alone produces -- a base model ~15 min

Create a free account to open this chapter.

• Base, instruct, chat, and reasoning models ~15 min
• Parameters, context length, and quantization ~17 min
• Cost, latency, and throughput trade-offs ~16 min

Create a free account to open this chapter.

• The message abstraction -- roles and parameters ~18 min
• Streaming, errors, retries, and rate limits ~20 min
• Cost logging and the running ledger ~18 min

Create a free account to open this chapter.

• Instructions, context, examples, and constraints ~16 min
• Few-shot prompting and templates ~18 min
• Failure modes, and a first look at prompt injection ~18 min

Create a free account to open this chapter.

• JSON schemas and constrained decoding ~20 min
• Function-call argument shapes ~18 min
• Validation with Pydantic, retries, and guardrails ~20 min

Create a free account to open this chapter.

• Hallucination and the knowledge cutoff ~16 min
• The RAG shape: chunk, embed, search, inject, generate ~20 min
• A worked RAG example: end to end ~20 min

Create a free account to open this chapter.

• Exact and approximate nearest-neighbor search ~20 min
• Metadata filtering and hybrid search ~22 min
• Reranking ~18 min

Create a free account to open this chapter.

• Query rewriting and HyDE ~18 min
• Parent-child chunking and context compression ~18 min
• The context budget: retrieval vs. stuffing vs. summarizing ~20 min

Create a free account to open this chapter.

• Tool schemas and the tool loop ~20 min
• Permissions, confirmation, and tool security ~20 min
• Error handling and tool results ~18 min

Create a free account to open this chapter.

• The agent loop: plan, act, observe, reflect ~22 min
• State and memory ~16 min
• When a workflow beats an agent ~18 min

Create a free account to open this chapter.

• ReAct and planner/executor patterns ~18 min
• Router and multi-agent/supervisor patterns ~17 min
• Stateful graphs, human-in-the-loop, and framework vs. native ~18 min

Create a free account to open this chapter.

• Resources, tools, and prompts ~16 min
• Servers, clients, and permissions ~17 min
• A small MCP example in Python ~15 min

Create a free account to open this chapter.

• From single-purpose calls to composed systems ~16 min
• Seven project patterns: an architecture walkthrough ~20 min
• Eval before you build: the course-wide harness ~15 min

Create a free account to open this chapter.

• The adaptation ladder as a decision chain ~18 min
• When prompting and RAG stop being enough ~16 min
• Cost, latency, and maintenance once you fine-tune ~16 min

Create a free account to open this chapter.

• Supervised fine-tuning, the basics ~16 min
• LoRA and QLoRA, by the numbers ~22 min
• Data prep, overfitting, forgetting, and merging ~18 min

Create a free account to open this chapter.

• Instruction tuning and preference data ~18 min
• Reward modeling and RLHF ~20 min
• DPO, KTO, GRPO, and distillation ~20 min

Create a free account to open this chapter.

• Exact match, F1, BLEU, and ROUGE ~20 min
• Perplexity, recalled, and human evaluation ~18 min
• LLM-as-judge: pairwise comparison and rubrics ~20 min

Create a free account to open this chapter.

• Retrieval metrics: recall, precision, and context relevance ~20 min
• Generation metrics: faithfulness, groundedness, and answer relevance ~20 min
• Agent metrics, and the reusable eval harness ~22 min

Create a free account to open this chapter.

• Prompt injection and jailbreaking, a deeper look ~22 min
• Tool abuse, excessive agency, and data leakage ~22 min
• Context and retrieval poisoning ~22 min

Create a free account to open this chapter.

• Input and output validation ~20 min
• Permissions, sandboxing, and least privilege ~20 min
• Audit logging and defense in depth ~22 min

Create a free account to open this chapter.

• Caching, batching, and quantization, recalled ~18 min
• Prompt compression and context management ~17 min
• Routing and cascades: the cost ledger's payoff ~22 min

Create a free account to open this chapter.

• The application server's shape ~16 min
• Async processing, streaming, and caching at the systems level ~17 min
• What this course doesn't own: generic MLOps, and where it lives ~12 min

Create a free account to open this chapter.

• Vision-language models and image understanding ~22 min
• Document understanding, OCR, and speech ~20 min
• Multimodal embeddings and multimodal RAG ~22 min

Create a free account to open this chapter.

• What "reasoning" means here -- observable behavior, not extraction ~16 min
• Self-consistency as sampling plus verification ~20 min
• Search-augmented test-time compute ~16 min

Create a free account to open this chapter.

• The capstone brief and the 11-point architecture document ~24 min
• Composing your system from this course's pieces ~20 min
• What good documentation and a real demo look like ~18 min

Create a free account to open this chapter.

• Colab and browser-based cloud environments ~12 min
• Running models locally: CPU and GPU ~12 min
• Hosted API access: keys and environment variables ~14 min

Create a free account to open this chapter.

• The Hub: models, datasets, and cards ~16 min
• transformers and tokenizers ~18 min
• Datasets, Evaluate, Accelerate, PEFT, and TRL ~18 min

Create a free account to open this chapter.

• Ollama and llama.cpp: two local-inference runners ~12 min
• GGUF, quantized models, and VRAM budgets ~15 min
• vLLM: local batching and throughput ~12 min

Create a free account to open this chapter.

• How to read a model card or a new tool's documentation ~14 min
• How to evaluate whether a new tool is worth adopting ~14 min
• A dated snapshot -- September 2026 ~12 min