Hacker News Reader

Ambixon The Lofi Magic of VHS Audio [video]

And many professional studios still using tape machines and emulations over the final songs to produce that fat, warm sound of analog era.

Like this Ampex recreation I love to use with my Apollo and Luna:

https://www.uaudio.com/uad-plugins/luna-custom/ampex-atr-102...

Ambixon OpenLLaMA: An Open Reproduction of LLaMA

Yeah, it's really so bad on desktops.

With my LLaMA AVX implementation on 32bit floats [0] there no performance gain after 2 threads, so remaining 14 threads available are of no use, there no memory bandwidth to load them with work :)

[0] https://github.com/gotzmann/llama.go

Ambixon OpenLLaMA: An Open Reproduction of LLaMA

It's not about threads number, it about memory bottleneck. Sweet spot for my M1 Pro laptop is around 6 threads and 4bit model - I've managed to get 20 tokens per sec, really impressive

Ambixon OpenLLaMA: An Open Reproduction of LLaMA

Is it Zen1 architecture? It should be much better on Zen2 and newer Epycs

Ambixon [dead]

Happy to announce new version of LLaMA.go - open-source CPU inference framework for GPT models.

That's a big milestone, we've embedded scalable server which allowing access to GPT model with simple REST API.

Now anyone is able to integrate local GPT into micro-service mesh or build fancy ML startup :)

Pre-compiled binary builds for all major platforms released too.

Free access to already converted LLaMA 7B and 13B models as well.

All model computations take place on regular CPU and RAM, so no GPU access needed at all.

Model quantisation is not supported yet, so you'll need at least 32 Gb RAM to grok with 7B model.

I'm ready to help with any questions :)

Ambixon Ask HN: Most interesting tech you built for just yourself?

LLaMA.go - open-source framework for LLM inference on regular CPUs [0]

It took me about a month of full-time, hard, day and night coding (including weekends) to finally build a solid piece which can handle some crazy CPU workloads of tensor math.

[0] https://github.com/gotzmann/llama.go

Ambixon OpenAI’s CEO says the age of giant AI models is already over

Not the TS, but that's actually the same goal I have in mind with [0] project.

Right now I'm building my homelab server which aimed to fit 1 TB RAM and 2 CPUs with ~100 cores total.

It will cost like 0.1% of what I need to pay for GPU cluster with the same memory size :)

[0] https://github.com/gotzmann/llama.go/

Ambixon OpenAI’s CEO says the age of giant AI models is already over

> "a picture is worth a thousand words"

and it might be opposite for the GPT models actually. it's just easier for humans to grasp the bunch of knowledge with one eyes sight, but usually most of useful information might be represented with just of bunch of words and machines are to scan through the millions of words in an instant.

Ambixon MiniGPT-4

I'm developing framework [1] in Golang with this goal in mind :) It successfully runs relatively big LLM right now, and diffusion models will be the next step

[1] https://github.com/gotzmann/llama.go/

Ambixon Choose Your Weapon: Survival Strategies for Depressed AI Academics

Could you point to some actual examples of using FFT for NNs in real projects / GitHub codebases? Would like to dig more into the thing.

Ambixon Show HN: Llama.go – port of llama.cpp to pure Go

Python needed only to converting PyTorch models into more accessible bin format :) All the inference and tensor math code is 100% Golang. I'm going to inject AVX2 assembler soon, but that's too very specific to Go ecosystem.

Ambixon Llama.cpp 30B runs with only 6GB of RAM now

How cool is that! There's also exciting news about early access for GPT5 13T model [1] features from Noosphere AI

[1] https://noosphere.chat

Ambixon Launch HN: Noosphere (VC W23) – First GPT5 chatbot with superhuman abilities

Noosphere AI released GPT5-based chatbot today. Spoiler:

- 10x bigger than GPT4 - 3.14x smarter than average human - Trained with Chinese and Japanese texts - Egyptian hieroglyphs as well - Computing costs doubles the whole Croatia GDP

Join the future here: https://noosphere.chat

Ambixon Why Python keeps growing, explained

> I haven't yet found a replacement for Django

Take a look here https://adonisjs.com

Ambixon Ask HN: Who wants to be hired? (November 2022)

  ---------------------------------------------------------
  Location: Remote
  Remote: Yes
  Willing to relocate: No
  Technologies: Golang, Rust, PostgreSQL, Docker, micro-services, clouds, queues and caches
  Résumé/CV: https://www.linkedin.com/in/gotzmann
  Email: sgotsulyak@gmail.com
  ---------------------------------------------------------

I'm passionate about distributed systems. Looking for a team doing their best to build great product, new tech or maybe Next Big Thing :)

Ambixon Ask HN: Do you still buy physical tech books like “Learn Rust” or “Learn Go”?

Yes, paper books on hard tech just easier for performant and/or thorough reading. And I just love the taste and look of real books.

Ambixon The dangerous playgrounds of 1900s through vintage photographs

I've had the playground very similar to those shown on the first photos in USSR school back 30 years ago. It's hard to imagine what dangerous plays we had there.

Ambixon Ask HN: Where should I live?

~25,000 Russians moved there right after the incident. So maybe many will return or move farther.

Ambixon Russia’s Invasion of Ukraine Is Redrawing the Geopolitics of Space - ISS

There's a big difference between democracy and political chaos. I went through all 90s in Russia - believe me, there were so many freedom and democracy around, that no one in modern America could ever imagine :) Despite all of that, all things around were so bad, that people turned to Putin's stability with great relief.

Ambixon BP quits Russia in up to $25B hit after Ukraine invasion

Exactly. And there still many people shocked by NATO'1999 bombarding of Serbia.

Ambixon Defensive tactics from the modern history of urban warfare

What about the same advice for the citizens of Iran, Iraq, Afghanistan? Should they have fought for their liberty against US troops the same way?

Ambixon Understanding the War in Ukraine

https://en.wikipedia.org/wiki/NATO_bombing_of_Yugoslavia

That was a really eye-opening moment for many people, I felt the same as now about Ukraine.

Ambixon Russian forces invade Ukraine after Putin orders attack

Yep, I remember the real excitement in Russia about US around 1990+ and Kosovo was like a cold shower for many believers there.

Ambixon I was wrong, we need crypto

Thanks for Kotkin recommendation, looks very interesting

Ambixon Show HN: EdgeDB 1.0

I've worked with ArangoDB recently and very much liked their AQL. How Edge compares with Arango?

Ambixon Laravel 9

Laravel is great, but sometimes just overkill for simple tasks. That was the reason for me to start PHP Comet, it's like SlimPHP on async steroids:

https://github.com/gotzmann/comet

Ambixon Ask HN: Who wants to be hired? (December 2021)

  ---------------------------------------------------------
  Location: Remote
  Remote: Yes
  Willing to relocate: No
  Technologies: Golang, Solidity, SQL, PHP, PostgreSQL, ArangoDB, Kafka, Docker
  Résumé/CV: https://www.linkedin.com/in/gotzmann
  Email: sgotsulyak@gmail.com
  ---------------------------------------------------------

I'm looking for a team working within emerging field like crypto, clouds, fintech or biotech doing their best to build great product, new tech or even Next Big Thing :)

So if you'll going to be a game changer for DeFi / NFT / whatever space - please let me know, we could do more together!

Ambixon Ask HN: Finding clients for a test engineering company

Engineers often overlook the fact that most of such deals still closed by old-school sales force I suppose. So if you'd like to have a scaling service business maybe you should hire someone with sales skills?

Ambixon Remote work will break the US monopoly on global talent

Please share with us :)

Ambixon Remote work will break the US monopoly on global talent

I'm living in Moscow, Russia and I know for sure that we have so many crypto companies here but still I was not aware that they are ready to pay Bay Area salaries :)