LLM

다양한 LLM에 대한 자료를 수집합니다.

Ⅰ. Reference

다양한 LLM 관련 레퍼런스를 소개합니다.

• 모델학습시 GPU 사용량 계산하는 사이트

  • https://huggingface.co/spaces/Vokturz/can-it-run-llm

• Open Ai : Transformer Debugger

  • https://github.com/openai/transformer-debugger

• Langchain Visualizer

  • https://github.com/amosjyng/langchain-visualizer

• RingAttention

  • https://github.com/lhao499/RingAttention

• Awesome LLM Tabular

  • https://github.com/johnnyhwu/Awesome-LLM-Tabular

• Awesome LLM

  • https://github.com/Hannibal046/Awesome-LLM

• Awesome Korean LLM

  • https://github.com/NomaDamas/awesome-korean-llm

• MiniGPT-4 (opensource)

  • https://github.com/Vision-CAIR/MiniGPT-4

• transformer library tutorial

  • https://huggingface.co/docs/transformers/v4.39.1/ko/index

  • https://github.com/NielsRogge/Transformers-Tutorials

• Py2Mojo

  • https://github.com/msaelices/py2mojo

• Awesome Korean speech recognition

  • https://github.com/rtzr/Awesome-Korean-Speech-Recognition

• Open Korean instructions

  • https://github.com/HeegyuKim/open-korean-instructions

• Practical MLOps

  • https://github.com/paiml/practical-mlops-book

• Embedding (ratsgo 한국어 임베딩 도서 코드)

  • https://github.com/ratsgo/embedding

Ⅱ. LLM Training

LLM 학습에 대한 레퍼런스를 소개합니다.

• WhatsApp : Finetune a LLM to speak like you based on your WhatsApp Conversations

  • https://github.com/Ads-cmu/WhatsApp-Llama
• KB-ALBERT : KB국민은행에서 제공하는 경제/금융 도메인에 특화된 한국어 ALBERT 언어모델

  • https://github.com/KB-AI-Research/KB-ALBERT
• 딥러닝 파이토치 교과서 - 입문부터 파인튜닝까지 wikidocs

  • https://wikidocs.net/book/2788
• 전이학습 기반 NLP

  • https://brunch.co.kr/@learning/12
• 자연어처리 : 토큰화

  • https://real-myeong.tistory.com/41
• 딥러닝을 위한 자연어 처리 입문

  • https://wikidocs.net/book/2155
• SentencePiece를 이용한 효과적인 한국어 토크나이저 만들기 (예제)

  • https://devocean.sk.com/blog/techBoardDetail.do?ID=164570&boardType=techBlog#none
• 언어모델의 말솜씨? NEFTune 한 스푼으로 업그레이드 (예제)

  • https://devocean.sk.com/blog/techBoardDetail.do?page=&boardType=undefined&query=&ID=165452&searchData=&subIndex=#none

Ⅲ. Methods

다양한 학습 방법론에 대한 레퍼런스를 소개합니다. (Pre-Training / Fine-Tuning)

• PPO

  • https://velog.io/@uonmf97/HUFS-RL-%EA%B0%95%ED%99%94%ED%95%99%EC%8A%B5-Reinforcement-Learning-PPO-Proximal-PolicyOptimization
• RLHF (Human Feedback)

  • https://tech.scatterlab.co.kr/luda-rlhf/
• TransformersData Augementation

  • https://github.com/varunkumar-dev/TransformersDataAugmentation
• Domain Adaptation : Generative Pseudo Labeling(GPL)

  • https://github.com/UKPLab/gpl
• X Algorithm

  • https://github.com/twitter/the-algorithm-ml
• MergeKit

  • https://github.com/arcee-ai/mergekit?tab=readme-ov-file#merge-methods
  • https://bigwaveai.tistory.com/79

Ⅳ. Quantizations

LLM 경량화에 대한 레퍼런스를 소개합니다.

• PEFT

• LoRA

• QLoRA

• Knwoledge Distillation

  • https://velog.io/@qtly_u/%EB%AA%A8%EB%8D%B8-%EA%B2%BD%EB%9F%89%ED%99%94-%EA%B8%B0%EB%B2%95-Knowledge-Distillation

Ⅴ. Alignments

LLM Alignment에 대한 레퍼런스를 소개합니다.

• Offline-DPO

• Online-DPO

  • https://jackyoung96.github.io/2024/03/24/DAP-OnlineDAP/
• OpenAi Alignment handbook

  • https://github.com/huggingface/alignment-handbook
• LoRA Instruct

  • https://github.com/leehanchung/lora-instruct

Ⅵ. Etc

• FlashAttention : Fast and memory-efficient exact attention

  • https://github.com/Dao-AILab/flash-attention
• Speculative Decoding : Fast Inference from Transformers via Speculative Decoding

  • https://arxiv.org/pdf/2211.17192