lightning-hydra-template

lightning-hydra-template copied to clipboard

Bump pytorch-lightning from 1.6.5 to 1.7.0

2

Bumps [pytorch-lightning](https://github.com/Lightning-AI/lightning) from 1.6.5 to 1.7.0. Release notes Sourced from pytorch-lightning's releases. PyTorch Lightning 1.7: Apple Silicon support, Native FSDP, Collaborative training, and multi-GPU support with Jupyter notebooks The core...

dependabot[bot]

Hi, I saw that wandb callbacks can't be used when running using the trainer flag `++trainer.fast_dev_run=true`. This causes a problem when running test cases on experiments that use wandb callbacks...

RafayAK

I need to load a checkpoint trained from other folder. When I use 'torch.load', it throw an error since I change the working directory. I search the solution from internet...

yangze68

how can we save `python run.py ...` command line into output directory?

1

I wonder how we automatically save the following command line into the current experiment output directory. ``` python run.py trainer.max_epochs=100 data.batch_size=23 optimizer.lr=0.01 optimizer.moment=0.9 a_lot_of_other_settings ``` In this way, it would...

marsggbo

how do we resume from a previous experiment?

1

Suppose the previous exp results are saved in `./logs/runs/exp1`, if we want to resume from `exp1`, we can set `trainer.resume_from_checkpoint=./logs/runs/exp1/checkpoint/best.ckpt` to load the checkpoint. However, my question is: 1. can...

marsggbo

Hi ashleve, I find this [line](https://github.com/ashleve/lightning-hydra-template/blob/9e7089ce70dac66096576c6596e966d4c38cbdce/src/models/mnist_module.py#L70) logs acc without setting `sync_dist=True` and may log unexcepted values in DDP mode. The correct way in my understanding is: `self.log("train/acc", self.train_acc, on_step=False, on_epoch=True,...

zhiyuanpeng

Unify the redundant running mode

3

AFAIK, In Pytorch Lightning, there are 4 running modes including [train/val/test/predict] (or maybe 5? including tune). But now in lightning-hydra-template, for using these different running modes, different *entry files* [train.py/val.py/test.py(eval.py)/predict.py]...

genghisun

Adapting the template to reinforcement learning applications

9

Hello @ashleve thanks so much for creating this template. I am finding it so helpful in a lot of my development. I can't thank you enough. I had a question...

00krishna

Add example notebook showcasing how to aggregate logs generated by CSVLogger and plot the results of multiple runs.

ashleve

Proposition: replace `Makefile` with `Snakefile`. Snakemake is a workflow management system for reproducible and scalable data analyses. https://github.com/snakemake/snakemake https://snakemake.readthedocs.io/

ashleve