Fabrice Normandin
Fabrice Normandin
https://github.com/libffcv/ffcv-imagenet/blob/e97289fdacb4b049de8dfefefb250cc35abb6550/train_imagenet.py#L124 Not to be nitpicky, but this could actually be replaced with two "1d" convolutions, one for width and one for height, which would use `~2K` operations instead of `~K^2`:...
Fixes #288 TODOs: - [ ] Figure out how to best make this available to the user.
Fixes #276 - Raises a warning whenever the config default is a dict that isn't one of the options in the subgroup values - Tries to use the closest matching...
It would be nice to somehow get a cleaner / more minimalist help output for programs that use the `subparsers` feature.
**Is your feature request related to a problem? Please describe.** A clear and concise description of what the problem is. Ex. I'm always frustrated when [...] **Describe the solution you'd...
**Is your feature request related to a problem? Please describe.** I'm thinking of deprecating this option. Rest assured, I won't do anything backward-incompatible in the short term. I'd like to...
**Describe the bug** Getting some weird conflict issues when trying to install Orion using conda. Here's the script I'm using: ```bash #!/bin/bash #SBATCH --gres=gpu:1 #SBATCH --cpus-per-task=1 #SBATCH --mem=16G #SBATCH --time=00:30:00...
Adds an example of how to use the new warm-start feature. Currently uses the Profet tasks, which require quite a bit of compute to run the first time (GPUS also...