I'm excited to see what cuTile-rs unlocks. Like the direction of HuggingFace's grout https://github.com/huggingface/grout project for local LLM inference:
- state of the art performance
- codebase that fits in a context window (including kernel definitions!)
- single binary deployment
Similar to antirez's ds4.c, but in Rust and with cuTile making kernels both easier to author and higher performance.
Intel | Compiler/Software Engineers | San Diego, Santa Clara, San Francisco, Folsom OR | REMOTE/VISA for strong candidates. Full-time.
Currently in deep learning, if one wants to implement a low level kernel (LSTM, batch-norm etc), you're either going to be writing CUDA or device specific low level code. Join us as we grow the Intel Nervana Graph project (https://www.nervanasys.com/intel-nervana-graph-preview-relea...) the deep learning equivalent of LLVM: an open source deep learning compiler and IR. We plan on pushing the boundaries of tensor optimizations and enable the same benefits that the programming language and compiler community have reaped from LLVM.
Required: C++, modern software engineering, at_least_3(humility, curiosity, drive, on_the_right_side_of_Dunning_Kruger_effect).
Desired: Python, low level optimization, numerical linear algebra, deep learning/ML
So parakey is on the application side of the fence when compared to those like YouOS... an interesting idea.
"When people need to call their "computer friend" in 2007 to install a program, scan a document, burn a CD or show a picture to a friend, there is a problem." So is the main aim of parakey to give applications the ability to interface seamlessly with the web? Or is it more about the simplicity of web 2.0 design?
This past week I was able to talk a partner of an oil and gas VC that specializes in small scale (less than 10M) projects. One of the things I asked her was if the 'small-fry' oil and gas industry was being smothered by the giants of Exxon, BP, and Shell much as some people claim is being done by Microsoft, Google, and Yahoo. She replied that the big three are only able to pursue and develop the projects which show a high margin due to their high operating costs thereby allowing small ventures to flourish in projects that aren't 'profitable enough'.
Certainly this argument has also been made for the internet market too, but the oil and gas market is heavily capital dependent with large 'retooling' costs whereas the internet industry is driven largely by a smart labor force (ie: a more agile industry). So is it a possibility that small time internet ventures are more vulnerable to attack than the equivalent oil and gas company (and other industries for that matter)?
Interesting, its always a wakeup call to see the amount of real world conclusions that can be gleaned from a (seemingly useless) dataset.
An interesting parallel to human psychology as well: often the people you are quite certain have nothing to teach you are the ones with the most valuable lessons for you to learn.
Without seeing the work going into YC firsthand, I can only speculate on the limitations of that end.
However, looking further into the lack of quality applicants, I believe strongly that the failures of the public education system are a direct cause of this lack of applicants. In fact, I would be willing to wager that most applicants are some of the few that 'slipped through the cracks' of the public schooling experience with their minds and ambitions intact. To avoid infringement: this is all covered in Paul's oldie-but-goodie article, "Why Nerds are Unpopular."
And then you have the even higher end (less direct) influences. ie: the modern world diluting the American dream through the 'satisfaction' derived from a big screen TV and a dozen maxed out credit cards. (anyone feel the urge to reread Great Gatsby?). But thats more the realm of philosophers and sociologists.
- state of the art performance
- codebase that fits in a context window (including kernel definitions!)
- single binary deployment
Similar to antirez's ds4.c, but in Rust and with cuTile making kernels both easier to author and higher performance.