fastai VS pytorch-lightning

Jeremy Howard's YouTube Channel - Jeremy maintains the fastai library, which is an excellent package that will help anyone build complicated ML architectures in minimum time. His YouTube Channel has a number of free courses which do an amazing job of covering a variety of ML topics, and he also maintains a very active forum for people studying ML.

To create the AI we will use fastai. This is a python library, which is build on top of pytorch. No worries, you don't need to know how to code python. We will learn how this stuff works along the way :)

You may know fast.ai as a popular deep learning course. There is also a deep learning library with the same name (https://github.com/fastai/fastai) as well as software development tools like nbdev (https://nbdev.fast.ai/).

fast.ai has been offering education and tools for free for over 7 years, and has been approached by many companies asking for help. This program offers an avenue for business to get relevant professional services and support.

The "fastai" course is free, and does a really nice job walking you through building simple neural nets from the ground up:

https://github.com/fastai/fastai

What's going on here is the exact same thing, just much, much larger.

Development NB: https://github.com/fastai/fastai/blob/master/nbs/09_vision.augment.ipynb

There is already a ton of momentum around automating ML workflows. I would suggest you contribute to a preexisting project like, for instance, PyTorch Lightning or fast.ai.

If you are unaware of what fastai is, its official description is:

Also, not a single docstring to document any code in the library - https://github.com/fastai/fastai/blob/master/fastai/vision/learner.py

def on_validation_end(self, trainer: Trainer, pl_module: LightningModule) -> None: # Trainer calls `on_validation_end` for sanity check. Therefore, it is necessary to avoid # calling `trial.report` multiple times at epoch 0. For more details, see # https://github.com/PyTorchLightning/pytorch-lightning/issues/1391. if trainer.sanity_checking: return

Under the hood, the ModelWrapper object will create a ML model based on the config (so far, an XGBoost model and a PyTorch Lightning model). Each of those will have a wrapper that conducts training and evaluation (since from my understanding of Lightning, Trainers are required to be outside of the class). In lack of a better name, I call these wrappers Fitters. For uniformity, I thought about adding a common interface IFitter, which is inherited by all model wrappers as outlined below.

Join their Slack to ask the community questions and check out the GitHub here.

Pytorch lightning benchmarks against pytorch on every PR (benchmarks to make sure that it is mot slower.

There is already a ton of momentum around automating ML workflows. I would suggest you contribute to a preexisting project like, for instance, PyTorch Lightning or fast.ai.

Is the problem the sheer number of options, or the fact that they are all together in one place? Would it be better if they were organized into the different trainer entrypoints (fit, validate, ...)? If that is the case, there was an RFC proposing this which you might find interesting, feel free to drop by and comment on the issue: https://github.com/PyTorchLightning/pytorch-lightning/issues/10444

I wanted to make a quick performance comparison between the GPU (Tesla K80) and TPU (v2-8) available in Google Colab with PyTorch. To do so quickly, I used an MNIST example from pytorch-lightning that trains a simple CNN.

We've noticed GPU 0 on our 3 GPU system is sometimes idle (which would explain performance differences). However its unclear to us why that may be. Similar to this issue

If you want a good example for reference, take a look at Pytorch Lightning's readme (https://github.com/PyTorchLightning/pytorch-lightning) It answers the 3 questions of "what is this", "why should I care", and "how do i use it" almost instantly