Notice
Recent Posts
Recent Comments
Link
| 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 |
Tags
- 사회 네트워크 분석
- python3
- PyTorch
- deepseek
- prompt engineering
- Spatial GNN
- GCP
- Transductive Learning
- DeepLearing.AI
- Google Cloud
- 생성 모델
- SSIM
- Vertex AI
- GNN
- PSNR
- Inductive Learning
- Spectral GNN
- public key
- ARIMA
- session 미종료
- posterior collapse
- 이상 탐지
- 대칭키
- ChatGPT
- cross-entropy
- mode collapse
- Grad-CAM
- zachary’s karate club
- kl divergence
- 생성 모델 평가
Archives
- Today
- Total
데이터 과학 노트
PyTorch 2.0 본문
PyTorch 2.0
torch.compile() 을 이용하여 연산 (operation) 단위를 전체 그래프로 묶어 더 빠르게 연산을 수행할 수 있다.
최적화 예시)
- 전체 그래프를 볼 수 있다면 다음 연산에서 output1 과 output2 를 concat 하는 연산이 있다면 concat 할 때 메모리 값을 옮기는 것이 아니라 처음에 output1 과 output2 를 만들때 concat 위치에 값을 쓴다.
- 전체 그래프를 볼 수 있다면 dependency 가 없는 값들을 funsion, 병렬 연산 등을 수행 할 수 있다.

References
'Data Science > Deep Learning' 카테고리의 다른 글
| mode collapse 와 posterior collapse (1) | 2023.12.29 |
|---|---|
| Inductive Bias (귀납 편향) (0) | 2023.05.28 |