Telegram Group & Telegram Channel
​​Byte Latent Transformer: Patches Scale Better Than Tokens

Новая статья от META - Byte Latent Transformer. Пробуют новый подход к токенизации - вместо фиксированного словаря используют динамические patches, размер которых определяется по энтропии следующего байта. Модель успешно масштабировали до 8B параметров и 4T байтов, при этом с лучшим качеством. Плюс эффективность и тренировки, и инференса лучше. Каких-то особых недостатков подхода авторы не описали. Ждём Llama 4 на байтах? :)

Paper link
Code link

Мои обзоры:
Personal blog
Medium
Linkedin Pulse

#paperreview



group-telegram.com/datastorieslanguages/342
Create:
Last Update:

​​Byte Latent Transformer: Patches Scale Better Than Tokens

Новая статья от META - Byte Latent Transformer. Пробуют новый подход к токенизации - вместо фиксированного словаря используют динамические patches, размер которых определяется по энтропии следующего байта. Модель успешно масштабировали до 8B параметров и 4T байтов, при этом с лучшим качеством. Плюс эффективность и тренировки, и инференса лучше. Каких-то особых недостатков подхода авторы не описали. Ждём Llama 4 на байтах? :)

Paper link
Code link

Мои обзоры:
Personal blog
Medium
Linkedin Pulse

#paperreview

BY Data, Stories and Languages




Share with your friend now:
group-telegram.com/datastorieslanguages/342

View MORE
Open in Telegram


Telegram | DID YOU KNOW?

Date: |

Following this, Sebi, in an order passed in January 2022, established that the administrators of a Telegram channel having a large subscriber base enticed the subscribers to act upon recommendations that were circulated by those administrators on the channel, leading to significant price and volume impact in various scrips. After fleeing Russia, the brothers founded Telegram as a way to communicate outside the Kremlin's orbit. They now run it from Dubai, and Pavel Durov says it has more than 500 million monthly active users. In addition, Telegram's architecture limits the ability to slow the spread of false information: the lack of a central public feed, and the fact that comments are easily disabled in channels, reduce the space for public pushback. Either way, Durov says that he withdrew his resignation but that he was ousted from his company anyway. Subsequently, control of the company was reportedly handed to oligarchs Alisher Usmanov and Igor Sechin, both allegedly close associates of Russian leader Vladimir Putin. "There is a significant risk of insider threat or hacking of Telegram systems that could expose all of these chats to the Russian government," said Eva Galperin with the Electronic Frontier Foundation, which has called for Telegram to improve its privacy practices.
from us


Telegram Data, Stories and Languages
FROM American