subagent-driven-development
OBobra/superpowers298k2,631 new stars in the last 7 days
nlp-alignment
Best practices for LLM alignment techniques including RLHF, DPO, and instruction tuning. Use when working on alignment or safety.
- Star
- 14,635
- Category
- Extending agents
- License
- MIT
- Last commit
- Aug 19, 2026
researchclaw/skills/builtin/domain/nlp-alignment/SKILL.md
View on GitHubLLM Alignment Best Practice
Methods:
- RLHF: Train reward model → PPO fine-tuning (complex but powerful)
- DPO: Direct preference optimization (simpler, no reward model needed)
- GRPO: Group relative policy optimization
- SFT: Supervised fine-tuning as alignment baseline
Training recipe:
- Start with SFT on high-quality instruction data
- DPO: lr=5e-7, beta=0.1, batch_size=64
- PPO: lr=1e-6, clip=0.2, KL coeff=0.02
- Use reference model for KL penalty
- Evaluate on safety benchmarks (TruthfulQA, BBQ, etc.)
Common pitfalls:
- Reward hacking: model finds shortcuts to high reward
- Mode collapse: model generates repetitive outputs
- Catastrophic forgetting: loses general capabilities
Similar Skills
chief-of-staff
MAmattpocock/skills288k10,178 new stars in the last 7 days
setup-matt-pocock-skills
MAmattpocock/skills288k10,178 new stars in the last 7 days
ask-matt
MAmattpocock/skills288k10,178 new stars in the last 7 days
agent-eval
AFaffaan-m/ECC278k4,046 new stars in the last 7 days
autonomous-agent-harness
AFaffaan-m/ECC278k4,046 new stars in the last 7 days