lairv

11 ms·

151.
by lairv 3y ago
How different are the performance between Bun and Go ? It'd be really convenient to have a decently fast backend runtime, but more expressive than Go and without the hassle of NodeJS
152.
by lairv 3y ago
How is HN different ? To me HN and Twitter hit that sweet spot of productive procrastination, where I feel like I'm keeping up with the tech world and learning new things, while in reality most of the things I read here aren't val
153.
by lairv 3y ago
I've encountered appimages a few time and wasn't really convinced. Maybe I'm missing something but here's what I found annoying: - need to move the x.appimage somewhere in your $PATH - need to create an alias like nvim.a
154.
by lairv 3y ago
The article make it look binary when in reality everyone is guessing to a certain degree. No one has ever come up to me in the street and asked for $1000. Everyone is guessing up to a point, but when there's too much incertitude, some
155.
by lairv 3y ago
Just thought this was funny
156.
by lairv 3y ago
What's the experience of writing web backend with Nim ? Did you use existing libraries / framework ? How good is the concurrency compared to something like Go ?
157.
by lairv 3y ago
Paths, Trees, and Flowers from Jack Edmonds (1965) https://www.cambridge.org/core/journals/canadian-journal-of-... Presents a beautiful algorithm (maximum matching in general graphs) and is very well written
158.
by lairv 3y ago
You seem passionate about it, so you should definitely keep going > Hasbro is litigious > I’m not going to charge without getting sue I don't think this is a thing ? There's this website which is a web Catan game ( https:&#
159.
by lairv 3y ago
- Tried 23.04 and found too many bugs (I guess this doesn't count) - Some day 22.04 randomly started without GUI and I couldn't get it back either with ubuntu-desktop and startx - I installed a Python package without a virtual env
160.
by lairv 3y ago
Had to reinstall ubuntu 3 times already since the beginning of this year and thus switched to debian, hopefully I'll be able to settle in for a while
161.
by lairv 3y ago
Another recent cool work in this field is this paper : https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/ They manage to get the same quality, with <1hr of training, and running at 60fps 1080p, it uses p
162.
by lairv 3y ago
I think vits is just a repo that implements the TTS model
163.
by lairv 3y ago
I wrote a colab to run TTS, it takes some time to install dependencies : https://colab.research.google.com/drive/11tZ1c7kQJ5uKImm9t-C... Overall I got pretty poor results in english and french, I guess it would require
164.
by lairv 3y ago
I don't have much knowledge about TTS models, is it possible/affordable to fine-tune those models on your own voice ?
165.
by lairv 3y ago
From their FAQ https://docs.modular.com/mojo/faq.html "Will Mojo be open-sourced? Yes, we expect that Mojo will be open-sourced. However, Mojo is still young, so we will continue to incubate it within Modular unti
166.
by lairv 3y ago
Read on their discord server that it both support .mojo and the fire emoji
167.
by lairv 3y ago
Exciting stuff indeed, looks like a unified approach to recent developments like (OpenAI) Triton, TVM, OpenXLA, ONNX Runtime, etc. though we'll have to wait for proper benchmarks and compatibility to know how great it is. Also it'
168.
Mojo – a new programming language for AI developers(modular.com)
643 points by lairv 3y ago|261 comments
169.
by lairv 3y ago
I was going to answer the same, I find the approach of machine learning compilers that directly compile models to host and device code better than having to bring a huge runtime. There are exciting projects in this area like TVM Unity [1],
170.
by lairv 4y ago
This doesn't seem that obvious to me, serving LLMs through an API allows to have highly optimized inference with stuff like TensorRT and batched inference while you're stuck with batch size = 1 when processing locally. LLMs doesn&
171.
by lairv 4y ago
LLMs weights are initialized from a normal distribution (or whatever distribution is now SOTA) while some animals can walk the second they are born
172.
by lairv 4y ago
https://en.wikipedia.org/wiki/Byte_pair_encoding tldr: start with unary characters and greedily merge pairs that are the most frequents A consequence is that an encoding is suited for the dataset it was trained on, so
173.
Koala: A fine-tuned Llama chatbot for Academic Research(bair.berkeley.edu)
4 points by lairv 4y ago|1 comments
174.
PyTorch and OpenXLA: The Path Forward(pytorch.org)
3 points by lairv 4y ago|1 comments
175.
by lairv 4y ago
I still don't fully grasp what XLA is, where does XLA sits against CUDA, ROCm, OpenVino ? Against ONNX/ONNX-Runtime ? Against OpenAI Triton ?
176.
by lairv 4y ago
Lot of cynicism in the comments so I'm just wondering, what would you want them to say, if there's anything that could satisfy you ? Now I'm looking like an OpenAI shill, but I'm really just happy that I'm alive in
177.
by lairv 4y ago
Couldn't read the article either but do people actually want to sort their tabs ? My tabs are approximately in a chronological order and I can find a specific tab based on that, sorting them would be a disaster Edit: oh ok it uses gpt-
178.
by lairv 4y ago
Is there any reason to believe that biologically inspired architectures should yield better performance ? Brain are biological systems which have been trained through evolutionary processes. Neural Networks are algorithmic/linear algeb
179.
New ways Google Maps is getting more immersive and sustainable(blog.google)
4 points by lairv 4y ago|0 comments
180.
WebGL2 fluid simulation in less than 5kb of JavaScript(loicmagne.github.io)
2 points by lairv 4y ago|0 comments

More ›