| 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
- Engineering at Anthropic
- Embedding
- attention
- catastrophic forgetting
- Langchain
- Positional Encoding
- PEFT
- chain-of-thought
- MQA
- reinforcement learning from human feedback
- fréchet inception distance
- Multi-Head Attention
- CoT
- self-attention
- re-ranking
- context engineering
- langgraph
- flashattention
- transformer
- SK AI SUMMIT 2025
- 트랜스포머
- 토크나이저
- BLEU
- extended thinking
- gqa
- test-time scaling
- RLHF
- MHA
- rotary position embedding
- model context protocol
Archives
- Today
- Total
AI Engineer 공간 "사부작 사부작"
Tags
- transformer
- Engineering at Anthropic
- catastrophic forgetting
- RLHF
- attention
- 트랜스포머
- SK AI SUMMIT 2025
- context engineering
- extended thinking
- test-time scaling
- fréchet inception distance
- chain-of-thought
- reinforcement learning from human feedback
- flashattention
- rotary position embedding
- model context protocol
- re-ranking
- gqa
- langgraph
- PEFT
- 토크나이저
- MQA
- Langchain
- CoT
- MHA
- BLEU
- Positional Encoding
- Multi-Head Attention
- Embedding
- self-attention
- tokenizer
- Rag
- reasoning
- prompt
- Lora
- FID
- LLM
- Rope
- context
- MCP
- clip
- benchmark
- Code execution with MCP: Building more efficient agents
- context rot
- attention budget
- Effective context engineering for AI agents
- Writing effective tools for agents
- Writing effective tools for agents — with agents
- 제9회 데보션(DEVOCEAN) 테크 데이
- NAVER Cloud AI DevDay
- 엔지니어링 밋업
- Agentic AI 핸즈온
- 프롬프트 캐시
- Time-to-First-Token
- TTFT
- 프롬프트 캐싱
- Prompt Caching
- Prompt Cache
- Train-time Scaling
- Multi Agents
- Deep Agents
- dynamic prompt
- custom prompt
- reasoning model
- checkpoints
- response metadata
- 스페셜 토큰
- special tokens
- subword tokenizer
- 서브워드 토크나이저
- pre-ln
- post-ln
- 레이어 정규화
- 학습률 예열
- learning rate warm-up
- http+sse
- fastmcp
- streamable http
- best-of-n
- best-of-n distillation
- budget forcing
- misaligned persona
- bad boy persona
- 창발적 오정렬
- 에이전트 오정렬
- 오정렬
- deceptive misalignment
- emergent misalignment
- agentic misalignment
- misalignment
- pad masking
- look-ahead masking
- skip link
- residual link
- swish gated linear unit
- element-wise multiplication
- swiglu
- silu
- implicit gating
- explicit gating
- 게이팅
- gating
- needle-in-a-haystack
- u자형 성능 곡선
- u-shaped performance curve
- reranker
- reranking
- lost in the middle
- mulit head attention
- multi-query attention
- grouped-query attention
- text-to-image retrieval
- recall@k
- coco captions
- visual question answering
- ablation study
- mechanistic interpretability
- inner activation exploration
- feature attribution
- attention visualization
- contrastive language-image pre-training
- humaneval
- hellaswag
- big-bench
- gsm8k
- 연속 배칭
- 정적 배칭
- pagedattention
- open neural network exchange
- io-awareness
- 타겟 모델
- 초안 모델
- 추측 디코딩
- speculative decoding
- halueval
- llm-as-a-judge
- inception score
- human evaluation
- automated metrics
- 에이전트 프레임워크
- ai agent framework
- reason + act
- cot 한계
- tree of thoughts
- 인 컨텍스트 학습
- int8
- bf16
- qat
- ptq
- gptq
- awq
- zero redundancy optimizer
- pytorch fully sharded data parallel
- fsdp
- deepspeed
- roformer
- semantic segmentation map
- human pose
- canny edges
- dpm-solver
- guidance scale
- 클래시파이어-프리 가이던스
- diffusion transformer
- 확산 트랜스포머
- conditional generation
- 잠재 확산 모델
- latent diffusion model
- iterative/recursive retrieval
- query transformation
- modular rag
- ragas
- rag score
- rag ratings
- unanswerability evaluation
- answer correctness
- answer relevancy
- ndcg
- context recall
- context precision
- hybrid search
- 하이브리드 검색
- semantic search
- 의미적 검색
- lexical search
- 어휘적 검색
- 파국적 망각
- 지속적 사전 훈련
- continuous pre-training
- weight averaging
- 가중치 평균
- 구면 선형 보간법
- 태스크 산술
- task arithmetic
- 모델 병합
- model merging
- tiesmerging
- quantized lora
- parameter-efficient fine-tuning
- reinforcement learning from verifiable rewards
- rlvr
- format rewards
- 형식 보상
- group relative policy optimization
- direct preference optimization
- reinforcement learning from ai feedback
- ai 피드백 기반 강화 학습
- 인간 피드백 기반 강화 학습
- rlaif
- 지도 파인튜닝
- 명령어 튜닝
- 거부 샘플링
- byte pair encoding
- 그룹-쿼리 어텐션 (grouped-query attention)
- 멀티-쿼리 어텐션 (multi-query attention)
- kv 캐싱
- kv cashing
- 창발적 능력
- emergent abilities
- 친칠라의 함정
- chinchilla trap
- scaling laws
- 스케일링 법칙
- low-rank approximation
- state space models
- efficient / linearized attention
- decoder-only
- encoder-only
- transformer architectures
- attention with linear biases
- 멀티 헤드 어텐션
- mixture-of-experts
- out-of-vocabulary
- oov
- 연속 학습
- kv cache
- 스토리 포인트
- 전문가 혼합
- mcp client
- grpo
- mcp server
- langchain-mcp-adapters
- claude code
- manus
- story point
- deep research
- textual inversion
- agents
- prompting
- instruction tuning
- deepseek-r1
- agentic ai
- slerp
- ddim
- deepseek
- qlora
- reward hacking
- adamw
- low-rank adaptation
- supervised fine-tuning
- retrieval-augmented generation
- truthfulqa
- mmlu
- vllm
- llm 한계
- batching
- llama3
- 포지셔널 인코딩
- rejection sampling
- 데보션
- mode collapse
- sparse attention
- Retrieval
- LLaVa
- positional embedding
- Multimodal
- HyperClova X
- Benchmarks
- context window
- Chinchilla
- Classifier-Free Guidance
- Diffusion Models
- ddpm
- Layer Normalization
- BPE
- 셀프 어텐션
- claude
- DreamBooth
- controlnet
- Fine-tuning
- Transfomer
- residual
- encoder-decoder
- bge
- DEVOCEAN
- fp16
- contrastive learning
- in-context learning
- Inductive bias
- 동적 프롬프트
- NAVER CLOUD PLATFORM
- challenges
- VQA
- 임베딩
- onnx
- 어텐션
- Observation
- sft
- NCP
- BM25
- Skip Connection
- Residual Connection
- image captioning
- Overfitting
- masking
- continual learning
- dpo
- TensorRT
- QUANTIZATION
- bias
- alibi
- TF-IDF
- depth
- flamingo
- 양자화
- fibonacci
- MOE
- stdio
- Continuous
- react
- T5
- 밋업
- Docker
- Add-on
- 프롬프트
- llama
- Scribble
- 잔차
- KAPLAN
- Dare
- mrr
- LDM
- 마스킹
- thinking
- cfg
- fairness
- Bert
- Cider
- TOT
- 피보나치
- container
- SSE
- Thought
- Training
- bond
- DRAFT
- rouge
- metadata
- Warm-up
- position
- FT
- Fragmentation
- dynamic
- jira
- DROP
- Factory
- Token
- sandbox
- Action
- DIT
- gaming
- PATCH
- padding
- AWS
- math
- ssm
- Spice
- server
- GPT
- bon
- HTTP
- calibration
- target
- FUSION
- ICL
- zero
- hyde
- static
- pattern
- 벤치마크
- retriever
- ARC
- Client
- 패딩
- Think
- IS