lightning-hydra-template

lightning-hydra-template copied to clipboard

What happended to the WatchModel callback?

1

I have been using this template for quite some time but haven't been catching up the latest update. I just saw that this thing is gone? Just want to know...

2533245542

Confused about reduce_fx in self.log

5

Hi, I was confused when I read about the way [reduce_fx](https://pytorch-lightning.readthedocs.io/en/stable/extensions/logging.html) is used. > Reduction function over step values for the end of an epoch. Uses torch.mean() by default and...

yipliu

According to hydra documentation config names should not have .yaml extension

1

Hello. Taken from the official [documentation](https://hydra.cc/docs/advanced/defaults_list/) of Hydra 1.2: > CONFIG_NAME: The name of a config, without the file system extension. e.g. mysql and not mysql.yaml. https://github.com/ashleve/lightning-hydra-template/blob/5034fca4e59dd3b1dff4797cb313886ee4daba54/configs/train.yaml#L5-L14 There are more...

yermandy

Are there any codes for training GANs?

1

Great job! Are there some demos or code snippets for training a GAN? Many thanks

DreamtaleCore

Multi-GPUs without DDP for hparams_search

2

It seems that current template only supports using multi-GPUs with DDP (Distributed Data Parallel), which means that multi-GPUs run with one common set of hyper-parameters. Is it possible to use...

ker2xu

Support for lightning light?

1

@williamFalcon wrote "if your workflow is complex and you want to own your full loop, then i would look at lightning Lite": https://pytorch-lightning.readthedocs.io/en/stable/starter/lightning_lite.html This applies to me, and many people....

turian

I want to use the function of dynamically selecting config_name. Can this function be added in future versions?

YChienHung

Override hydra config group with none/null

2

Hi everyone, is it possible to achive this from the command line? I.e. I want to remove all callbacks for one run. I tried `python3 src/train.py callbacks=null` `python3 src/train.py callbacks=none`...

Tomakko

Training stuck when submitting job to slurm with multigpu and ddp

1

The training is stuck and I get the error The client socket has failed to connect to [ip6-localhost]:24355 (errno: 99 - Cannot assign requested address) Need help with this.

sri9s

Added mdformat-myst

1

## What does this PR do? This PR adds mdformat-myst to pre-commit config for mdformat, to ensure scientific functions are checked correctly. No breaking changes. Fixes #625 ## Before submitting...

alexkrz