A new Kazakh language model, SHERKALA, has been launched
Author - Mansur Ismagulov
The Tech AI
Get the key points in seconds - a short summary from The Tech AI.
Companies Inception and G42, together with MBZUAI and Cerebras, have presented SHERKALA — the first high-accuracy Kazakh language model. The model, with 8 billion parameters and trained on 45 billion words, supports Kazakh, English, Russian, and Turkish languages through a 25% expansion of the Llama 3.1 tokenizer. Training was conducted on the Condor Galaxy supercomputer. SHERKALA is available as open-source on Hugging Face for more than 13 million Kazakh speakers.
Inception, G42, together with the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), and with support from Cerebras, has presented SHERKALA — the first high-accuracy Kazakh language model.
SHERKALA is designed for over 13 million Kazakh speakers, providing access to advanced AI technologies. The model, with eight billion parameters and trained on 45 billion words, supports Kazakh, English, Russian, and Turkish. It uses Llama 3.1 with a tokenizer extension of 25%, which improves text processing accuracy. Training was conducted on Condor Galaxy, one of the most powerful supercomputers developed by G42 and Cerebras.
SHERKALA is available as open-source on Hugging Face, offering new opportunities for developing AI solutions in the Kazakh language.
The Tech on Telegram Follow The Tech on Telegram
Keep up with the latest technology and startup news.