티스토리 뷰
8/27/2024
Pytorch Lightning & Pytorch Compatibility
https://lightning.ai/docs/pytorch/stable/versioning.html#compatibility-matrix
자꾸 Pytorch 새로 깔려해서 빡치네. Pip에서 torch version에 맞게 알아서 Pytorch Lightning install 해야하는 거 아닌가? 반대가 아니라.
08/22/2024
Multi-gpu training setting in Pytorch Lightning version 2.4
Default strategy는 DDP인듯. 어찌어찌 하다보니 Pytorch Lightning update했는데 갑자기 SD가 requires_grad = False 인 것 때문에 생기는 문제로 인해 DDP를 그냥 쓰면 안되고 특별한 argument를 줘야했음.
https://lightning.ai/docs/pytorch/stable/common/trainer.html
08/02/2024
Background: ControlNet fine-tuning하려 그러는데 Pytorch Lightning으로 다 wrapping 되어있어서 튜토리얼 좀 보고 있음:
https://lightning.ai/docs/pytorch/stable/starter/introduction.html
다른건 모르겠는데 코드 모듈화해서 유지보수가 편할 것 같고 (솔직히 리서치코드를 누가 유지보수하겠냐만, 이런 코드 틀을 본인이 다른 프로젝트에 계속 가져가서 하면서 일종의 유지보수를 할 수도 있음), mixed precision 같은 최적화 기능을 편리하게 버그없이 넣을 수 있는 게 장점일 것 같음.
Is defining 'training_steps' method enough for the PL module?
Yes. Perhaps.
What about 'configure_optimizers' method? Is it necessary?
Yes. https://lightning.ai/docs/pytorch/stable/common/lightning_module.html
LightningModule — PyTorch Lightning 2.3.3 documentation
LightningModule A LightningModule organizes your PyTorch code into 6 sections: When you convert to use Lightning, the code IS NOT abstracted - just organized. All the other code that’s not in the LightningModule has been automated for you by the Trainer.
lightning.ai
What is the default training epoch?
If both max_epochs and max_steps aren’t specified, max_epochs will default to 1000. To enable infinite training, set max_epochs = -1. https://lightning.ai/docs/pytorch/stable/common/trainer.html
What happens if you pass logger to the call_back argument of PL Trainer?
Just normal call_back... The methods like 'on_train_batch_end' can be overrided. https://lightning.ai/docs/pytorch/stable/extensions/callbacks.html
How does PL handle multi-gpu training?
Default is DDP. You can set any way. https://pytorch-lightning.readthedocs.io/en/0.9.0/multi_gpu.html
'Research (연구 관련)' 카테고리의 다른 글
HHF Part2 - Evolutionary Development and Anatomy of the Hand (0) | 2024.08.05 |
---|---|
HHF Part1 - Historical Overview and General Introduction (0) | 2024.08.05 |
What is Conformal Prediction? (1) | 2024.07.23 |
What are "Spherical harmonics"? (0) | 2024.07.01 |
What is Equivariance in Computer Vision? (0) | 2024.06.28 |
- Total
- Today
- Yesterday
- Transformation
- 피트니스
- Machine Learning
- 인터뷰
- Docker
- 컴퓨터비전
- 에디톨로지
- 헬스
- nerf
- 문경식
- camera coordinate
- densepose
- world coordinate
- VAE
- Generative model
- demo
- 컴퓨터비젼
- pytorch
- 머신러닝
- part segmentation
- nohup
- focal length
- 비전
- spin
- Interview
- Virtual Camera
- pyrender
- Pose2Mesh
- deep learning
- 2d pose
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 | 31 |