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 |
- AI Fairness
- trustworthiness
- Transformer
- Ai
- GPT
- 인공지능
- 지피티
- MLOps
- 챗지피티
- DevOps
- 머신러닝
- 케라스
- word2vec
- fairness
- nlp
- 트랜스포머
- 챗GPT
- 자연어
- 설명가능성
- XAI
- Tokenization
- ML
- Bert
- 딥러닝
- gpt2
- 신뢰성
- cnn
- 인공지능 신뢰성
- ChatGPT
- GPT-3
- Today
- Total
목록gpt2 (2)
research notes

*** Jay Alammar blog 필요 부분 발췌 내용 *** https://jalammar.github.io/illustrated-gpt2/ https://jalammar.github.io/how-gpt3-works-visualizations-animations/ The illustrated GPT-2 □ Looking Inside GPT-2 The simplest way to run a trained GPT-2 is to allow it to ramble on its own (which is technically called generating unconditional samples) – alternatively, we can give it a prompt to have it speak about..

GPT-1: Generative Pre-Training of a Language Model 1. 연구배경 보통 NLP 관련 데이터 셋은 Labeled 데이터 세트 보다 Unlabeled 형태의 데이터 셋이 훨씬 많다. 즉, 언어 모델을 학습하는데 필요한 데이터의 활용측면에 있어 레이블 된 데이터와 그렇지 않은 학습데이터 수의 차이가 상당하다. 따라서, Unlabeled 데이터 셋을 이용해 학습을 한 후 이를(Pre-trained model) 활용해 특정 태스크에 대한 학습을 수행하면(레이블 된 데이터를 활용해) 더 좋은 성능을 가진 모델을 만들 수 있을 것이다라는 가정을 가지고 본 연구가 수행되었다. 2. GPT: Unsupervised pre-training GPT는 기존 트랜스포머 구조에서 인코더를 ..