Results for “alignment-handbook”

Plans and coordinates the deployment of a model to Amazon SageMaker AI, selecting the appropriate pathway (real-time, serverless, async, batch, or Bedrock CMI) based on model type, traffic, latency, and cost constraints.

Sets up and audits the Claude Code harness for a project: CLAUDE.md, .claude/rules, skills, subagents, settings.json permissions, hooks, verification loop. Commands: kickoff (scaffold new setup), check-playbook (score an existing one), scouting-report (last scorecard), adjust (fix bloat/misplaced instructions), drill (turn advisory prose into real hooks/permissions/CI), decline (record an accepted risk), gameplan (work order with acceptance criteria before building), watch-film (check a diff against that order for scope creep/weakened tests/false claims). Use for setting up Claude Code, or on: "Claude ignores my CLAUDE.md", "it's huge and still misses things", "it said done but ran nothing", "it changed files I didn't ask about", "it weakened a test to pass", "rule, skill, or hook?", "is my setup any good". Not for code quality (code-audit), test coverage (test-assessment), one-off prompt wording (genie-proof-prompts), new skill authoring (skill-creator), or compacting a conversation (handoff).

Rewrites free-form draft prompts into the conventions Claude Opus 4.6 or 4.8 responds best to, either inline or by overwriting a draft file.

Provides an adaptive methodology for adding machine learning models to existing codebases, covering problem framing, data readiness, architectural decoupling, and baseline model integration.

Fine-tune and align language models using reinforcement learning with TRL, including SFT, DPO, PPO, GRPO, and reward model training.

Surfaces Han's shared Human-Readable Output Standard — the readability rule and the writing-voice profile — into the calling skill's own context, so the caller drafts in voice and runs its self-check against the current standard sourced from one canonical copy. Use when a prose-producing skill needs the shared readability standard available in context before it drafts. Governs the shape of a written deliverable, where explanation-guidance governs what a run says to a person in a turn. Runs in the caller's context and hands control straight back; it does not produce a deliverable of its own, rewrite anything, or judge the caller's work. Does not run the adversarial rewrite pass — dispatch the readability-editor agent for that, or use edit-for-readability to rewrite an existing target. Does not cover explaining technical work to a reader who will not implement it — use explanation-guidance for that.

End-of-session context handoff. Saves session state so the next session or a parallel agent can pick up exactly where the left off. Run this when context is getting full (~80%) or before stopping work.

Package an in-progress task on this cluster so a Claude agent on another cluster (MMM Young, CX3, HX1, Isambard) can pick it up, or resume a task handed off to this cluster. Use when the user says hand off / migrate / continue this on <cluster>, or asks you to pick up an existing handoff.

Discovers, audits, repairs, adapts, and designs bounded AI-agent loops with explicit triggers, actions, verification, stopping conditions, guardrails, and handoffs.

Meta-adder for toolkit artifacts. Single entry point to add a skill, pattern, best-practice, hook, agent, or guide with correct frontmatter + sync to 4 publish targets.

Author learning targets for a competency across developmental bands with precise, observable progression language. Use when writing 'I can' statements for competency-based programmes.

Alpaca: A Strong, Replicable Instruction-Following Model

Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.

Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.

Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.

Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.

Trains LLMs with SimPO, a reference-free preference optimization method that outperforms DPO, using configurable hyperparameters and workflows for various models and tasks.