Ahmad Ahmadov
@ahmadatum
Senior Data Scientist at Porsche AG. Passionate about data, AI, blockchain, cars and e-mobility. Travelotographer 🇦🇿🇩🇪🇹🇷🇰🇷🇺🇸 | Opinions are my own.
ID: 1427566046790262784
https://www.linkedin.com/in/ahmedahmedov/ 17-08-2021 09:41:12
1,1K Tweet
148 Followers
162 Following
@prerationalist According to Chinchilla scaling laws for LLMs: For every parameter in the model, you need 20 training tokens to be "optimally" trained. And you should always strive to train for 200B+ tokens to have a meaningful LLM. So if you want to train a 25B model well, you need to train it
New Years fireworks in Baku, Azerbaijan. I was practicing some night time photographs from one of the windows on the International Space Station at the end of the work day on New Years Eve. I had just finished passing over my targets when I noticed something funny – the city below me was