Engineering / advanced
Build and evaluate language model systems
Follow tokens through inference, retrieval, tools, adaptation, and evaluation with explicit quality and resource budgets.
Your route through engineering
Follow the path or begin with a question. Every lesson is open.
Your place, at your pace. Progress is optional and saved only in this browser when available.
Read / explore / reason
Inside this module
Module 01 advanced
Build and evaluate language model systems
Follow tokens through inference, retrieval, tools, adaptation, and evaluation with explicit quality and resource budgets.
- 01
Tokenization and vocabulary design
Connect subword representations to context use, embedding memory, language coverage, and reproducible model inputs.
29 min - 02
Autoregressive decoding and sampling
Derive temperature, top-k, and nucleus sampling while separating likelihood, diversity, correctness, and reproducibility.
30 min - 03
KV caches and inference memory
Account for attention state by layers, key-value heads, sequence length, precision, and concurrent requests.
32 min - 04
Prefill, decoding, and phase disaggregation
Model first-token and streaming latency before separating prompt processing and token generation across workers.
32 min - 05
Quantization and calibration
Derive a uniform quantizer and evaluate precision changes through payload size, clipping error, representative calibration, and kernel behavior.
31 min - 06
Retrieval-augmented generation
Design evidence retrieval with separate recall, grounding, freshness, authorization, and context-budget measurements.
33 min - 07
Tool use and structured generation
Separate grammar-constrained output, schema validation, semantic checks, authorization, and side-effect execution.
32 min - 08
Preference optimization and alignment
Derive a pairwise preference objective relative to a reference policy and evaluate what preference labels can establish.
34 min - 09
Language model evaluation
Build reproducible, paired evaluations that separate task success, uncertainty, contamination, judgment bias, and serving cost.
34 min - 10
Agent orchestration and resource budgets
Model a bounded agent as a state machine with explicit observations, permissions, verification, retries, and token-cost limits.
34 min