Hacker News Reader

vpjon PyTorch 1.10

Agreed. But it makes implementing and testing a lot easier for us. We try to minimize the dependencies as much as possible.

Do you think it also comes in the way of understanding the paper and learning how to implement it?

vpjon PyTorch 1.10

We did a bunch of popular research paper implementations in PyTorch with notes (annotations); might be helpful.

https://nn.labml.ai

vpjon PyTorch 1.10

Have you looked at TorchServe

https://pytorch.org/serve/

vpjon PonderNet: Annotated PyTorch Implementation

Github: https://github.com/labmlai/annotated_deep_learning_paper_imp...

Tweet on the implementation: https://twitter.com/labmlai/status/1425809861028192267

Tweet by Deepmind on PonderNet: https://twitter.com/DeepMind/status/1425814786080608257

vpjon Show HN: Trending Machine Learning Papers on Twitter

Thanks. Just checked my self and it's kind of bad. Will work on it.

vpjon Show HN: Paper Time – Listen to abstracts of CS papers, like a custom podcast

https://papers.labml.ai/

vpjon World to hit temperature tipping point 10 years faster than forecast

Atlantic Ocean currents weaken, signalling big weather changes - study

HN Discussions:

- https://news.ycombinator.com/item?id=28082887

- https://news.ycombinator.com/item?id=28078575

- https://news.ycombinator.com/item?id=28085342

- https://news.ycombinator.com/item?id=28091087

vpjon Scientist says cleaning indoor air could make us healthier and smarter

Nothing

vpjon Scientist says cleaning indoor air could make us healthier and smarter

Bought one about a week ago, because of similar research studies. I don't notice anything different. I use it in the bedroom when sleeping.

May be my air quality was already pretty good. I will check the status of the filter after a few weeks to see how much purifying it had to do.

vpjon Ask HN: Reinforcement learning for single, lower end graphic cards?

Depends on what sort of RL you are doing. If you are trying to train agents to play small games with vision the agent will need a small cnn to process images. This will need a gpu and what you have should be enough.

I was training on atari for a while with 1080ti. The games run on the cpu so you need a decent cpu as well.

vpjon React's UI State Model vs. Vanilla JavaScript

Thanks for finding the typo.

Ton of abstractions is an overstatement. https://github.com/vpj/weya is more like a helper function around `document.createElement`.

vpjon React's UI State Model vs. Vanilla JavaScript

https://github.com/labmlai/labml/tree/master/app

vpjon React's UI State Model vs. Vanilla JavaScript

This is true. We recently re-wrote everything in plain javascript from react, and it didn't increase the lines of code.

The vanilla code was much easier to maintain too. React comes in your way when things get complicated and the work arounds produce really messy code.

Edit:

Project (GitHub): https://github.com/labmlai/labml/tree/master/app

It's a mobile/web app to monitor machine learning experiment.

vpjon Tesla Q2 2021 Vehicle Production and Deliveries

I guess most people were waiting for the Plaid

vpjon Break-It-Fix-It: Unsupervised Learning for Program Repair

PDF: https://arxiv.org/pdf/2106.06600v1.pdf

Twitter thread by author: https://twitter.com/michiyasunaga/status/1407749347714818052

vpjon Lossy Compression for Lossless Prediction

PDF: https://arxiv.org/pdf/2106.10800v1.pdf

vpjon Revisiting Deep Learning Models for Tabular Data

PDF: https://arxiv.org/pdf/2106.11959v1.pdf

vpjon Regret Minimization in Games with Incomplete Information (CFR)

I think it's a derivative of CFR, with a lot of optimizations and abstractions. They reduce the search space by grouping different card combinations together e.g. KK and QQ are treated same. They also limit the action space and only allow half-pot or full pot bets.

This is what I remember. I went through what they published briefly sometime back. So I could be wrong here.

vpjon Regret Minimization in Games with Incomplete Information (CFR)

The twitter thread tries to explain the concepts without the math

https://twitter.com/labmlai/status/1407186002255380484

vpjon Revisiting the Calibration of Modern Neural Networks

PDF: https://arxiv.org/pdf/2106.07998v1.pdf

vpjon Annotated PyTorch implementations of various deep learning normalization layers

Author here. From next month onwards we will pick papers to implement based on votes on Github issues https://github.com/lab-ml/nn/issues

Feel free to open an issue if there's a paper that you like implemented.

vpjon Show HN: Monitor server hardware usage from phone

Hi, I am one of the developers. This makes total sense, we were thinking of doing alerts. Do you have any specific alerts in mind? Do you think they should be customizable or general anomalies be enough?

I don't think you are missing a better workflow. Our current use case is monitoring the hardware while monitoring model training. In this case researchers tend to take a look at how their models are doing from time to time and they could check on hardware at the same time. This wouldn't be the case for general servers.

vpjon Show HN: Collection of deep learning implementations with side-by-side notes

All of the class definition has to be in a single cell. You can’t have notes within it only normal comments

vpjon Show HN: Collection of deep learning implementations with side-by-side notes

We used https://coffeescript.org/#literate for some products at https://forestpin.com and it was so much easier to maintain.

One of the problems with notebooks for literate programming is that it kind of breaks down when you define a class or a long function. The entire code has to be within a single block.

vpjon Show HN: Simple VSCode extension to autocomplete Python with a transformer model

Github repo: https://github.com/lab-ml/python_autocomplete

vpjon Show HN: Autocomplete Python Code with Transformers

Links:

Github repo: https://github.com/lab-ml/python_autocomplete

Training notebook: https://colab.research.google.com/github/lab-ml/python_autoc...

Evaluation notebook: https://colab.research.google.com/github/lab-ml/python_autoc...

vpjon Show HN: Remote – A CLI tool for distributed PyTorch model training

I haven't had time to really use it. Finished off coding a couple of days ago. I only used it with a bunch of old pytorch projects for testing on Lambda Labs cloud. I plan to use it for training some NLP models, that are too slow for my home GPU setup.

For now, starting servers, copying their ips/hostnames to the configuration files and stopping servers when training finishes seem is tedious. I think I will integrate it to the AWS (and other cloud) api so that you can automate starting and terminating servers.

vpjon [dead]

We don’t have an android phone and will test when we get hold of one.

Will add installation instructions.