Smerity

15 ms·

31.
by Smerity 6y ago
A random fun note about how Peter Molyneux was able to start Bullfrog, sourced from Wikipedia[0] and mentioned in this blog's previous post[1]: Molyneux tried to sell a game years earlier but failed so instead started a baked bean expo
32.
by Smerity 6y ago
To expand on this from the recent Australian fires[1] causing massive damage to traditionally fire resistant environments: > We’re seeing recurrent fires in tall, wet eucalypt forests, which normally only burn very rarely. A swamp dried
33.
by Smerity 6y ago
From what I understand (speaking casually to someone on the Google Chrome team) Safari is able to integrate into operating system level APIs that are not available for other applications, impacting especially power efficiency.
34.
by Smerity 6y ago
I have been using CapRover and love it. I donate to their OpenCollective[1]. For those noting "why don't you just use Linux / k8s / ...", that feels close to the original complaints re: Dropbox on Hacker News[2]. I&
35.
by Smerity 6y ago
On the linked page it's noted: > The labs used in CS3210 are largely based on the materials originally developed for CS140e: An Experimental Course on Operating Systems by Sergio Benitez.
36.
by Smerity 7y ago
I still strongly disagree. Few of these hand written CUDA kernels outside of the frameworks are about implementing derivative rules, they're about eliminating the CUDA call overheads or avoiding the layered computational / memory
37.
by Smerity 7y ago
The CUDA I've written has never been joyous but it also hasn't been as horrific as I'd expected. There's a period of hair pulling but persistence will get you through it. The majority of CUDA code is closer to C than C++
38.
by Smerity 7y ago
For high performance CUDA kernels people still need to write derivatives by hand. I know this as for my own research, and for many production systems, I'd still need to write it myself. Many of my architectures wouldn't have been
39.
by Smerity 7y ago
PyTorch already has TPU support via XLA and has been used in research at the scale of billions of parameters. https://github.com/pytorch/xla
40.
by Smerity 7y ago
Champion! Thank you =]
41.
by Smerity 7y ago
Does anyone know of an easy way to download the nucleotide sequences for the entire set of pneumonia / coronavirus viruses? I've looked at the FTP mirror but can't connect it with the nucleotide locus I find on the site itsel
42.
by Smerity 7y ago
I ran a computer and network security course back from 2010-2013 and the simplified version of this was part of our wargames. The difficulty scaled with recency of the last commit. Admittedly we had to minimize the number of points awarded
43.
by Smerity 7y ago
Having read it I love the approach - it's very similar to how we handled data at Common Crawl / Web Archive (WARC) files / inspired by MapReduce + object storage interactions. I'll keep it in mind for the next opportunit
44.
by Smerity 7y ago
I made the WikiText-2 and WikiText-103 datasets they compare against and held state of the art results on language modeling over PTB, WT-2, and WT-103 not too long ago. OpenGPT-2's results are near equal to GPT-2's in zero-shot la
45.
by Smerity 7y ago
Given these chips are purely meant for machine learning the issue that "your wafer computer was always going to be sub-optimal at something" is less of an issue now than in traditional scientific programming setups, especially of
46.
by Smerity 7y ago
Thanks for replying with your stance, genuinely. I genuinely wanted to just add the additional context for the Hacker News community to discuss as it seemed relevant. If you believed the person this was a story from their past. If you are u
47.
by Smerity 7y ago
Disclosure: I worked for Freelancer.com seven or eight years ago back in Australia and know the founder. This is a "new" post from an old and now deleted post [1] by the same author. In the initial post [1] they state "I&#x27
48.
by Smerity 8y ago
Chainer and potentially CuPy (which was extracted from Chainer to be independent) were around before PyTorch as it served as inspiration for PyTorch. I feel like that's a good motivation for diversity in packages and ecosystems regardl
49.
by Smerity 8y ago
A third recommendation for CapRover. It's simple, it's beautiful, and it's open source. I've been enjoying it so far :) CapRover also features one click SSL generation via Let's Encrypt, similar to what is offered o
50.
by Smerity 8y ago
This a brilliant move! The most impactful people to me, both in Australia (Sydney) and the US (Boston/SF) are those who fit an altruistic advisor archetype. They're the people I owe the most to and also those who genuinely care th
51.
by Smerity 8y ago
The big news here: Citus Data donated 1% of their equity to non-profit PostgreSQL organizations[1] so this acquisition is a win for the community even in the darkest scenario of Citus Data disappearing into a canyon on the Microsoft campus.
52.
The compute and data moats are dead(smerity.com)
82 points by Smerity 8y ago|23 comments
53.
by Smerity 8y ago
It's polite ribbing amongst work colleagues where an engineer at Cloudflare (@zackbloom) offers praise to a colleague (@kentonv) for a mighty project (libc replacement) and Cloudflare's CTO (@jgrahamc) jokes that the project isn&#
54.
by Smerity 8y ago
I love these stories of engineering and science from the trenches. Prefacing the story with "the early Pixar renderer was stupidly slow for silly archaic (and at the time sensible reason) X" followed by "luckily no-one is tha
55.
by Smerity 8y ago
Actually, yes, the QRNN has all of those features. First figure from our paper: how the LSTM with a twist allows for the equivalent speed of a plain convnet by running efficiently in parallel on GPUs, like image processing convents.[1] Best
56.
by Smerity 8y ago
As a researcher who has work "featured" here, I am confused. I was going across the projects and found my own - "Quasi-Recurrent Neural Network"[1]. I was interested to see what type of model would be trained as the QRNN
57.
by Smerity 8y ago
My original comment was meant for a separate HN article on machine learning and I posted in the wrong tab. My apologies.
58.
by Smerity 8y ago
There are already big names in this list and it was a a tenth (~30) only a few hours ago when I signed. Yoshua Bengio, Jeff Dean, Yann LeCun, Ian Goodfellow, ... Even if it only represents their personal opinion, like herd immunity I'm
59.
by Smerity 9y ago
Agreed that if you have the control (and potentially the time depending on the algorithm), type specific compression is the way to go. Having said that, zstd beats Snappy handily for text ^_^ On enwik8 (100MB of Wikipedia XML encoded articl
60.
by Smerity 9y ago
PyTorch was born from Lua Torch but they're not interchangeable and noting installation issues regarding Lua Torch when the article is discussing PyTorch is likely confusing. PyTorch helpfully provides clear installation instructions f

More ›