Show HN: Run LLaMa2 on the Browser with Ggml.js
3 by anonymousd3vil | 1 comments on Hacker News.
You can now build serverless AI inference web application with ggml.js's LM backends.
Saturday, 12 August 2023
New Show Hacker News story: latest news
Show HN: React Hooks in Python
2 by amitassaraf | 0 comments on Hacker News.
A React inspired way to code in Python. Python Hooks is a very opinionated project and it's not meant to be a replacement for any of the existing state management libraries. It was created as a coding exercise gone sideways.
2 by amitassaraf | 0 comments on Hacker News.
A React inspired way to code in Python. Python Hooks is a very opinionated project and it's not meant to be a replacement for any of the existing state management libraries. It was created as a coding exercise gone sideways.
New Show Hacker News story: latest news
Show HN: Hypersprawl – a metaverse with 4 spatial dimensions
2 by teemur | 0 comments on Hacker News.
Hi all, Years ago I was wondering what kind of problems superhuman AI would be able to solve that would be difficult for humans. I came up with problems requiring higher dimensional spatial awareness. Scratching my head a bit further, the humble me started thinking that actually there is no reason why our brains could not handle higher spatial dimensions, if the brain was just given appropriate feedback from our sensory system. Fast forward a lot of further head scratching, and here is a barely MVP of a VR environment for 4 spatial dimensions, with some navigation aids including acoustic sensory augmentation (sounds fancier than it is...) First I thought to have just the Hypershack where you can locally build 4d stuff on your browser, but then thought it would be cool to have also a 4d metaverse where you could share and see 4d things built by others - and there might be even a way to monetise this for me by renting real estate in there. So, that is Hypersprawl. To manage expectations, the performance is bad for now, I have limited the Hypersprawl to 10 4d objects per address. You can use Hypershack for free, also there are lots of free addresses in Hypersprawl, only if you want to get closer to the Black Sun (sorry Stephenson, I couldn't not copy that), you need to make a donation or subscription. You do not need to have a VR headset to have a look, browser is enough, but of course immersion suffers quite a lot. All feedback is more than welcome. Frankly, I do not expect this to be a major commercial success and actually I think this may be a bit too far in the fringes of geekiness to find anyone else interested, but I guess this community is one of the better places to try to prove me wrong. (So, have I been successful in getting any intuition in 4 spatial dimensions? So far not much luck on that, only that the 4d space is huge. Literally mind-bogglingly huge. I think my next direction is to spend some time figuring out the pedagogy on what objects/functionality I should have there to get the intuition. But do not expect anything to be quick with this one, and I am talking more likely about years than months.) Cheers, Teemu
2 by teemur | 0 comments on Hacker News.
Hi all, Years ago I was wondering what kind of problems superhuman AI would be able to solve that would be difficult for humans. I came up with problems requiring higher dimensional spatial awareness. Scratching my head a bit further, the humble me started thinking that actually there is no reason why our brains could not handle higher spatial dimensions, if the brain was just given appropriate feedback from our sensory system. Fast forward a lot of further head scratching, and here is a barely MVP of a VR environment for 4 spatial dimensions, with some navigation aids including acoustic sensory augmentation (sounds fancier than it is...) First I thought to have just the Hypershack where you can locally build 4d stuff on your browser, but then thought it would be cool to have also a 4d metaverse where you could share and see 4d things built by others - and there might be even a way to monetise this for me by renting real estate in there. So, that is Hypersprawl. To manage expectations, the performance is bad for now, I have limited the Hypersprawl to 10 4d objects per address. You can use Hypershack for free, also there are lots of free addresses in Hypersprawl, only if you want to get closer to the Black Sun (sorry Stephenson, I couldn't not copy that), you need to make a donation or subscription. You do not need to have a VR headset to have a look, browser is enough, but of course immersion suffers quite a lot. All feedback is more than welcome. Frankly, I do not expect this to be a major commercial success and actually I think this may be a bit too far in the fringes of geekiness to find anyone else interested, but I guess this community is one of the better places to try to prove me wrong. (So, have I been successful in getting any intuition in 4 spatial dimensions? So far not much luck on that, only that the 4d space is huge. Literally mind-bogglingly huge. I think my next direction is to spend some time figuring out the pedagogy on what objects/functionality I should have there to get the intuition. But do not expect anything to be quick with this one, and I am talking more likely about years than months.) Cheers, Teemu
Friday, 11 August 2023
New Show Hacker News story: latest news
Show HN: Startup Viability Calculator
3 by brntsllvn | 0 comments on Hacker News.
Last week, my co-founder and I walked away from our revenue-generating startup Jason Cohen's startup viability calculator was the nail in the coffin https://ift.tt/SVY1oTE We spoke to many investors about our startup, and the calculator is the best proxy for vc feedback I've seen Use the calculator as a gut check before speaking with customers, building, or pitching. It should save you time
3 by brntsllvn | 0 comments on Hacker News.
Last week, my co-founder and I walked away from our revenue-generating startup Jason Cohen's startup viability calculator was the nail in the coffin https://ift.tt/SVY1oTE We spoke to many investors about our startup, and the calculator is the best proxy for vc feedback I've seen Use the calculator as a gut check before speaking with customers, building, or pitching. It should save you time
New Show Hacker News story: latest news
Show HN: Pykoi – a Python library for LLM data collection and fine tuning
15 by jaredwilber | 0 comments on Hacker News.
Hi HN, pykoi is an open-source python library for ML scientists. pykoi makes it easier to collect data for LLMs, to use that data for finetuning, and to compare models to each other (e.g. your model pre- and post- finetuning, or your model vs openai vs claude). The library comes from pain points we experienced in LLM development: 1. Collecting feedback data from users isn't as easy as it could be. (The current process usually involves sharing excel files of annotated responses back-and-forth, offering no insight into how users actually engage with your models). 2. RLHF remains complicated to carry out. By complicated , we mean requires a lot of steps, hundreds of configs, lengthy setups, etc. 3. Comparing models to each other as they're used (that is, independent from academic metrics) is full of friction. The current approach: spin up a model, ask questions, write them down. Repeat for other models then compare. At a high-level, we think that the active learning process should be closed-loop: data collection, fine tuning, and inference all feed from the same system. This library is our first step in that direction. The project is still very early but we hope that some if it is useful. Note, we're fully open-source, and actively adding features! Website: https://ift.tt/Uq0pQBl GitHub: https://ift.tt/yaQuDMs We would love your feedback!
15 by jaredwilber | 0 comments on Hacker News.
Hi HN, pykoi is an open-source python library for ML scientists. pykoi makes it easier to collect data for LLMs, to use that data for finetuning, and to compare models to each other (e.g. your model pre- and post- finetuning, or your model vs openai vs claude). The library comes from pain points we experienced in LLM development: 1. Collecting feedback data from users isn't as easy as it could be. (The current process usually involves sharing excel files of annotated responses back-and-forth, offering no insight into how users actually engage with your models). 2. RLHF remains complicated to carry out. By complicated , we mean requires a lot of steps, hundreds of configs, lengthy setups, etc. 3. Comparing models to each other as they're used (that is, independent from academic metrics) is full of friction. The current approach: spin up a model, ask questions, write them down. Repeat for other models then compare. At a high-level, we think that the active learning process should be closed-loop: data collection, fine tuning, and inference all feed from the same system. This library is our first step in that direction. The project is still very early but we hope that some if it is useful. Note, we're fully open-source, and actively adding features! Website: https://ift.tt/Uq0pQBl GitHub: https://ift.tt/yaQuDMs We would love your feedback!
Subscribe to:
Posts (Atom)