মিডিয়া
- টুইট
- টুইট এবং উত্তর
- মিডিয়া, বর্তমান পৃষ্ঠা।
-
Thank you very much again for noticing this Sylvain
New version pushed.
This is what the learning rate and the momentum look like now:pic.twitter.com/IOwnVXyNBL
-
Fastest way to train on CIFAR10 - now available on a PC near you!
This is a
included docker setup featuring recent work by @GuggerSylvain &@jeremyphoward (training with AdamW and the 1 cycle policy). Very excited by the development! https://github.com/radekosmulski/cifar10_docker …pic.twitter.com/JA2NkeL212
-
A great into to pipelines! Learning this is one of my mini-projects. Gives you superpowers when working with structured data and is a very nice way to reason about calculations in general. Did you know sklearn had FunctionTransformer and Imputer?

https://youtu.be/BFaadIqWlAg -
This is from the first lecture of the Computational Linear Algebra course:https://www.youtube.com/watch?v=8iGzBMboA0I&index=2&list=PLtmWHNX-gukIc92m1K0P6bIOnZb-mg0hY …
-
But enough of making statements based on perceptions! Time to look at some data
Here is model performance on the val set. Not indicative of arch performance in general but carries some info on the ability to train said models using the tools I used on a single 1080tipic.twitter.com/VKWVxbmKih
এই থ্রেডটি দেখান -
1yr ago I gave up on ML. I didn't know what to learn nor how After a 5 mths break I decided to give ML one last try. If it would not work out I would need to let it go to not continue to waste my time - maybe I am unable to learn this I then signed up for the
@fastdotai coursepic.twitter.com/wjORNbkctx
এই থ্রেডটি দেখান -
Just how fast is a 1080ti? Training on cifar10 to 94% accuracy on a p2.8xlarge takes 4 min 10 sec (30%) longer than locally on a 1080ti! On a p2.xlarge that's nearly 1hr. The hyperparams have been tweaked for squeezing out every ounce of performance out of 1080ti but still...pic.twitter.com/C52YJxp689
এই থ্রেডটি দেখান -
`ssh -A` so good and yet I didn't know about it until recently! Effectively allows you to take your identity with you to the machines you ssh into. Committing to github / bitbucket without having to generate / add key on remote machine? No problem!pic.twitter.com/cbmY8j1AJ2
-
seemed like a good moment to fly the
@fastdotai banner
(this is from the @kaggle iMaterialist Challenge (Fashion) at FGVC5 competition that is still under way)pic.twitter.com/uSmKPHOAMf
-
-
A couple of pictures from the training. This is the groundbreaking 1cycle policy by Leslie Smith implemented beautifully in
@fastdotai by@GuggerSylvain. You can read more about it on his blog here: https://sgugger.github.io/the-1cycle-policy.html#the-1cycle-policy ….pic.twitter.com/uPNXWefo8U
এই থ্রেডটি দেখান -
Train on cifar10 to over 94% accuracy on a single
@nvidia 1080ti while having a
? Why not
train time < 13 min 30 sec, 20 epochs, lr = 1.5, bs = 128
(this is essentially @fastdotai DAWNbench submission with minor tweaks for running locally) https://github.com/radekosmulski/machine_learning_notebooks/blob/master/cifar10_fastai_dawnbench.ipynb …pic.twitter.com/xpzi71KnkG
এই থ্রেডটি দেখান -
I know this has been done a trillion of times before and is not very complicated... but this is fun and wanted to share! (CAM is covered at the end of
@fastdotai v2 lesson 7 and is a nice intro to FCNs and using hooks in@PyTorch)pic.twitter.com/FOi6bj6w45
-
-
NVME vs HDD: training on 10k train, 5k val, single epoch (no opportunity for caching on val to kick in), small arch, files a subset of 76GB dataset, highly optimized
@fastdotai dataloader cc@Ml101Freakpic.twitter.com/h4iA2oDCWL
-
I finally understand why git. Great talk to watch if you have been using git for a while but are still relatively new. Has also other great thoughts. My favorite one was that speed doesn't only mean we can do smth faster, but that it changes how we do it.https://youtu.be/4XpnKHJAok8
-
nvmes are an amazing piece of tech! random reads are so important for many applications - so looking forward to exploring the difference they can make for ML Below what iotop reports when reading from the hdd and the nvmepic.twitter.com/vjWVxVQw8b
এই থ্রেডটি দেখান -
How long does it take to read 1 mln (76 GB) worth of jpegs? A lot depends on the drive (below comparison of reading from a hdd and an nvme)pic.twitter.com/qSnd4aMnSg
এই থ্রেডটি দেখান -
You can now make
@kaggle submissions directly from a@ProjectJupyter notebook with ease
pic.twitter.com/yaUtOzupEm
-
Outstanding work! :) I can barely keep up with all the things being added to the library given the pace nowadays BTW since not too long ago the black border is also supportedpic.twitter.com/xeaSSBU2Kd
লোড হতে বেশ কিছুক্ষণ সময় নিচ্ছে।
টুইটার তার ক্ষমতার বাইরে চলে গেছে বা কোনো সাময়িক সমস্যার সম্মুখীন হয়েছে আবার চেষ্টা করুন বা আরও তথ্যের জন্য টুইটারের স্থিতি দেখুন।
ML / DL ideas - I tweet about them / write about them / implement them. Self-taught RoR developer by trade.