Llama Guard and Prompt Guard: Local Safety Models

Llama safety models is a useful local-model lesson because it makes one trade-off visible: teaching guardrails, prompt-injection detection, local moderation, and defense-in-depth around open-weight assistants. The point is not to crown a permanent winner. The point is to learn how to match a model family to hardware, task, license, and risk.

Current source signal

Build the small version

Build a two-step local pipeline: classify the prompt, then either answer, refuse, or ask for safer framing.

1. 1Pick one exact model file or runtime tag from the current model card.
2. 2Run three short prompts: one easy, one task-specific, and one likely failure case.
3. 3Record load time, response speed, memory pressure, answer quality, and one surprising failure.
4. 4Write a one-paragraph recommendation: use it, do not use it, or use it only for a narrow job.

local_guardrail_pipeline:
  input -> prompt_guard
  if injection_risk == high: stop_and_explain
  input -> safety_classifier
  if unsafe == true: safe_refusal
  else: main_local_model

log: category, confidence, decision, no private text

The big idea: remember local guardrail. Local model work is product design under constraints, not just downloading the model with the loudest leaderboard score.