|
--- |
|
language: |
|
- en |
|
- ru |
|
license: apache-2.0 |
|
|
|
|
|
tags: |
|
- gpt |
|
- NLG |
|
|
|
--- |
|
# HuYaLM 100B |
|
|
|
**Hu**gging Face **YaLM 100B** (by [BlackSamorez](https://github.com/BlackSamorez)) is a _transformers_ compatible implementation of **YaLM 100B** model, originally trained by Yandex for 65 days on a cluster of 800 A100 graphics cards and 1.7 TB of online texts, books, and countless other sources in both English and Russian. |
|
|
|
This particular implementation was motivated by the fact that the model was originally published with outdated code, incompatible with latest achievemnt in the field. This code, being compatible with _transformers_, should automatically support much needed features such as [quantization](https://huggingface.co/docs/transformers/main_classes/quantization) and [adapter training](https://huggingface.co/docs/peft/index). |
|
|
|
Training details and best practices on acceleration and stabilizations can be found on **[Medium](https://medium.com/p/d1df53d0e9a6)** (English) and **[Habr](https://habr.com/ru/company/yandex/blog/672396/)** (Russian) articles. The original code published by Yandex can be found on [GitHub](https://github.com/yandex/YaLM-100B). |
|
|
|
This code, as well as the model itself, is published under [Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0) license, permitting commercial use. |
