Module 10 · Neural and language models

Module 10: LLMs and evaluated generative workflows

Read the overview and learning outcomes here, then follow the learning route. Return to this page’s run instructions for the complete study.

Run an actual small local language model and evaluate structured outputs without repairing failures into apparent success.

This module works independently: no earlier notebook state, Canvas assignment, paid AI account, or GPU is required. Allow approximately 2–3 hours plus first installation/downloads. Small CPU runs demonstrate mechanisms, not production readiness.

Learning outcomes

• M10.1: Explain token generation and distinguish instructions from untrusted message data.
• M10.2: Run a recorded public model revision with bounded greedy decoding.
• M10.3: Validate an exact JSON schema and count failures.
• M10.4: Compare task fit, latency, and limitations with simpler NLP.
• M10.5: Write a source-grounded model-use note with privacy and cost constraints.

Learning route

Follow foundation → concepts and small cases → guided construction → independent practice. Use solutions after your attempt, then connect your work to the complete reference below.

Entry support and topics

Use the foundation bridge if these concepts are new. Basic Python/functions/files and the train/validation/test distinction are assumed with support from Module 2 and Module 5.

A language model predicts token continuations, not guaranteed facts or valid JSON. Temperature/sampling change generation behavior; this core uses greedy decoding and a token cap. Format compliance, label correctness, and latency are separate measurements.

Real data and its limits

UCI SMS Spam Collection; real human-labeled messages. Public messages may contain personal content; processing stays local and reports use hashes/contact masking. See the source register.

Sources, licenses, hashes, and model provenance distinguish external records from original teaching code. No generated dataset or invented labels are used in this module. Public files still require responsible handling. Source versions are fixed after verified acquisition; changed cached/upstream bytes cause an error. The first data/model download requires internet; retain the cache for offline replay.

The complete workflow

Compare conditional label likelihood, strict free generation and a TF-IDF/logistic baseline on the same 100 final SMS messages. The four few-shot examples are real training records. A separate seeded 100-message validation subset selects the likelihood decision threshold; the lexical threshold uses its validation predictions. Both LLM methods see the same contact-masked first 240 characters and demonstrations. Free-generated JSON is parsed without repair. Report format validity, accuracy, confusion counts, latency and Wilson intervals separately. The final sample preserves the historical split prevalence approximately; it is not a current deployment population.

The reported likelihood score is relative to the two candidate completions, not calibrated confidence. A legal label is guaranteed by the candidate set; a correct label is not. Compare against the training-majority classifier as well as the learned lexical baseline.

Run and inspect

Download the complete source or use a checkout. Work from its ml_ai_it folder. Use the isolated CPU framework environment. Run framework commands in the Linux CPU reference environment; see the support matrix for platform coverage. Reading the reports needs no framework installation. In commands below, python means your environment executable; use its full path if not activated. If a download link does not open, use the offline access route.

python examples/sms-llm/scripts/run.py --output runs/module10 --examples 100

Use a fresh output directory for a new run; completed studies are not silently overwritten. Open runs/module10/index.html or read summary.txt/summary.json and the companion CSV tables. Read the verified worked report without installing an execution interface.

Key artifacts:

• generation_evaluation.csv
• split_manifest.json

The shared implementation is under src/practical_ml_ai; scripts and commented notebook call the same functions. Keep the notebook inside the complete source tree. Named results and captioned/scoped tables support interpretation without a chart or screenshot. For interface choices and test coverage, see the accessibility guide.

Read selectively

• Core concept/reference: Transformers generation strategies — greedy decoding, token limits, and inference. Read these named sections, not an entire external course.
• Source/model context: central source register, including the relevant official card and license.
• Optional depth: Declare a prompt/template or decoding comparison on validation examples. Freeze it before a new final evaluation. Larger models or hosted providers are optional and need separate rights, privacy, compute, and evaluation decisions.

Practice and continue

Complete original practice, then compare worked feedback. Instructor notes map evidence and explain adaptation.

Next: Module 11.

Build understanding step by step

Use the guided construction and mastery check before the full study. New programmers should start with Python Foundation and readiness; use the glossary for notation.