| 일 | 월 | 화 | 수 | 목 | 금 | 토 |
|---|---|---|---|---|---|---|
| 1 | 2 | 3 | 4 | 5 | 6 | 7 |
| 8 | 9 | 10 | 11 | 12 | 13 | 14 |
| 15 | 16 | 17 | 18 | 19 | 20 | 21 |
| 22 | 23 | 24 | 25 | 26 | 27 | 28 |
| 29 | 30 | 31 |
Tags
- model context protocol
- 토크나이저
- langgraph
- chain-of-thought
- MQA
- reinforcement learning from human feedback
- Multi-Head Attention
- Embedding
- catastrophic forgetting
- SK AI SUMMIT 2025
- attention
- Positional Encoding
- rotary position embedding
- MHA
- extended thinking
- RLHF
- CoT
- flashattention
- PEFT
- gqa
- Langchain
- BLEU
- fréchet inception distance
- context engineering
- transformer
- test-time scaling
- self-attention
- re-ranking
- Engineering at Anthropic
- 트랜스포머
Archives
- Today
- Total
목록modular rag (1)
AI Engineer 공간 "사부작 사부작"
고급 RAG 기법: 단순 검색을 넘어 지능적 답변으로대규모 언어 모델(LLM)은 방대한 지식을 학습하지만, 최신 정보 부재나 환각(hallucination) 현상과 같은 한계를 지닙니다. 검색-증강 생성(Retrieval-Augmented Generation, RAG)은 이러한 LLM의 약점을 보완하기 위해 외부 지식 소스에서 관련 정보를 검색하여 답변의 정확성과 신뢰성을 높이는 핵심 기술입니다. 기본적인 RAG 접근법도 효과적이지만, 검색된 정보의 관련성이 떨어지거나 복잡한 질문에 대한 답변 능력이 제한될 수 있습니다. 이러한 한계를 극복하고 RAG 시스템의 성능을 한 차원 끌어올리기 위해 다양한 고급 기법들이 등장했습니다. 이 글에서는 단순 검색을 넘어 보다 정교하고 지능적인 답변 생성을 가능하게 하..
Theory/Retrieval-Augmented Generation
2025. 5. 21. 21:44