LLM & GenAI — MOC
상위: VAULT-INDEX · 규칙: 설계 규약
디코더 계열 생성형 LLM. 계보별·연대순 정독.
목차 (계보)
- GPT / o-series — OpenAI (GPT-1→5, o1/o3/o4)
- Claude — Anthropic
- Gemini — Google DeepMind
- Llama — Meta
- Qwen — Alibaba
- DeepSeek (V2/V3/R1)
- Mistral / Mixtral
- Gemma / Phi (소형·오픈)
- 오픈소스·스케일링기 — GPT-J·PaLM·BLOOM·Falcon·OPT
- 프론티어·리저널 — Grok·Kimi·Yi·OLMo·Command
- 하이브리드 SSM-LLM — Jamba (→ 22.SSM)
- 효율화·어텐션 — LoRA·QLoRA·FlashAttention·GQA·MoE
키워드: #llm GPT Claude Gemini Llama Qwen DeepSeek MoE reasoning-model
현재 노트 (81)
- PrompterDay(SKT,OPENAI)
- albert
- architecture-pretraining-objectives
- attention-is-all-you-need
- bart
- bert
- bloom
- claude-4-5
- claude-4
- claude
- cohere-command-a
- deberta
- deepseek-r1-zero
- deepseek-r1
- deepseek-v2
- deepseek-v3
- distilbert
- electra
- elmo
- ernie
- falcon
- flan-t5
- flash-attention-2
- flash-attention
- gemini-1-5
- gemini-2-5
- gemini-3
- gemini
- gemma-3
- gemma
- gopher
- gpt-1
- gpt-2
- gpt-3
- gpt-4-1
- gpt-4
- gpt-5-2
- gpt-5
- gpt-j
- gpt-neo
- gqa
- grok-3
- jamba-1-6
- jamba
- kimi-k2-5
- kimi-k2
- layer-norm-transformer
- llama-2
- llama-3
- llama-4
- llama
- llava
- lora
- mamba
- mistral-7b
- mistral-large-3
- mixtral
- mt5
- o1
- o3-pro
- o3
- o4-mini
- olmo
- opt
- palm
- phi-3
- phi-4-reasoning
- phi
- qlora
- qwen2-5
- qwen2
- qwen3-5
- qwen3
- roberta
- sheared-llama
- switch-transformer
- t5
- ul2
- vit
- xlnet
- yi