Hacker News-首页· laxmena·· 3 小时前AI 评分62
Interactive notes on RLHF, attention, LoRA and LLM evals
Show HN: Interactive notes on RLHF, attention, LoRA and LLM evals
AI 导读
作者推出了一个关于 RLHF、attention、LoRA 和 LLM 评测等核心概念的交互式笔记与学习资料库。内容涵盖从基础的准确率、F1、nDCG@k、pass@k 到 LLM-as-a-judge、RAG 和 Agent 评测，以及机器学习中的 attention、强化学习、LoRA 和搜索排序等工程实践。
正文 · 原文
Concepts
Engineering ideas you understand by changing the inputs.
Evals
The map of evalsStart here
- Is 82% better than 80%?
- Exact match and F1
- Precision, recall and reciprocal rank
- nDCG@k
- pass@k
- LLM as judge
- Pairwise preference and Elo
- Human agreement and kappa
- RAG evals
- Agent evals
- Online evals and A/B tests
- Contamination and Goodhart
Machine learning
Search and ranking
Notes I keep for myself, added as I learn. Found a mistake? · Source · Analytics choice
Allow Google Analytics to count page views? It sets cookies. Nothing else on this site tracks you.
Allow No thanks
来源：Hacker News-首页 · concepts.laxmena.com