Some shameless shelf promotion. I'm quite the shellfish! nbogoychev.com/protesting-l...
Protesting like a Chinese: Puns
Language is a source of many misunderstandings. Sometimes on purpose...
nbogoychev.com
Nikolay Bogoychev
@xapajiamnu.bsky.social
CTO at https://lasttoken.ai LLMs, neural networks, logographic writing systems. Ex-Meta https://nbogoychev.com
I decided to take some toy online repos that teach people deep learning using the LLaMa 3 architecture and see how far I can go optimising them for training speed. The tl;dr 9X nbogoychev.com/step-by-step...
Step by step pytorch performance optimisation: LLaMa 3 as a case study
The evolution of a powerful transformer.
nbogoychev.com
Hear me out guys, there's a new cool hanzi in town: nbogoychev.com/hanzi-of-the...
Hanzi of the day! 聞
The smells of spring are all over!
nbogoychev.com
Time for another update on my concussion recovery... nbogoychev.com/will-concuss...
Will concussions ever stop sucking? 3 year update.
More than three years still, and my brain is not right.
nbogoychev.com
[1/2] Our benchmark Multiloko is now featured on Kaggle! Submit your model so that you can see how it does on the secret, OOD part of the dataset. www.kaggle.com/benchmarks/m...
MultiLoKo Leaderboard | Kaggle
A multilingual local knowledge benchmark for LLMs spanning 31 languages
kaggle.com
Time to discuss the matter of Queens in the Chinese language nbogoychev.com/hanzi-of-the...
Hanzi of the day! 后
The one stays behind the king...
nbogoychev.com
Back to hanzi, time to talk about dancing! nbogoychev.com/hanzi-of-the...
Hanzi of the day! 舞 and 無
Dancing for the rain...
nbogoychev.com
[1/5] Check out Multiloko, our new multilingual [] knowledge dataset, developed together with Dieuwke Hupkes! We have collected questions in 31 languages, sourced locally, so that the content is relevant to the culture that speaks the language, mirroring closer real world user behaviour:
