Technique (Paper Review) — MOC
상위: VAULT-INDEX · 규칙: 설계 규약
LLM 학습·추론 기법(논문 리뷰 중심).
목차
- 스케일링 법칙: Scaling Laws · Chinchilla
- RAG 계열: RAG · REALM · Self-RAG · In-Context RALM · ARES
- 정렬: InstructGPT(RLHF) · Constitutional AI · DPO
- Instruction Tuning: FLAN · T0 · Flan-PaLM
- 위치 인코딩: RoPE
- MoE: Switch Transformer · Sparse Expert Models
- 추론 최적화: Speculative Decoding · PagedAttention
- 벤치마크: Chatbot Arena · AgentBench · MegaVerse
키워드: #technique RAG Chinchilla RLHF DPO RoPE scaling-laws
현재 노트 (25)
- agentbench
- ares-rag-eval
- chain-of-thought
- chatbot-arena
- chinchilla
- constitutional-ai
- dpo
- flan
- in-context-ralm
- instructgpt
- megaverse
- multitask-prompted-training
- paged-attention
- rag
- realm
- rethinking-demonstrations
- roformer-rope
- scaling-data-constrained
- scaling-instruction-finetuning
- scaling-laws
- self-rag
- sparse-expert-models
- speculative-decoding
- switch-transformers
- training-helpful-harmless