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
- 티스토리챌린지
- covolution
- C++
- 최대 사후 확률
- Unreal Engine
- Focal loss
- 딥러닝
- 모수 추정 방법
- CNN
- 오블완
- Diffusion
- convolution
- Action Recognition
- 포인터
- computer graphics
- 최대 가능도 추정
- Deep Learning
- Swin Transformer
- depthwise convolution
- OpenGL
- Image denoising
- Vision Transformer
- posec3d
- mobilenet
- DDPM
- deformable covolution
Archives
- Today
- Total
심드렁하게 저장
Error Fix List 본문
RuntimeError: CUDA error: device-side assert triggered
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
이 에러는 GPU에서 수행 중인 연산에서 잘못된 인덱스/값이 들어가서 assert가 실패했을 때 발생
대표적인 원인
- 잘못된 라벨 인덱스
- 예: CrossEntropyLoss 사용 중인데 target 값이 클래스 개수 이상 (>= num_classes)일 때
- 음수 인덱스
- 예: label = -1이거나 None 상태로 학습에 들어가는 경우
- Tensor 크기 불일치
- 예: input.shape = [B, 101], target = 101 (정수값이 클래스 개수보다 큼)