search

bigscience-workshop / petals

🌸 Run LLMs at home, BitTorrent-style. Fine-tuning and inference up to 10x faster than offloading
https://petals.dev
MIT License
8.89k stars 489 forks source link

Add YaLM-100B #512

Closed Aspect004 closed 9 months ago

Aspect004 commented 9 months ago

Could you consider adding the YaLM-100B open-source model by Yandex? I think it would be a nice addition.

borzunov commented 9 months ago

Hi @Aspect004,

Unfortunately, our team does not have capacity to work on this. If necessary, you can follow the "Run a custom model" guide to add YaLM support yourself - you can use this HF Transformers-compatible version (sorry about the name).

Please note that YaLM is an older model and is likely to be inferior to smaller newer models that use Chinchilla scaling laws (such as Llama/Llama 2). For example, see benchmarks on Russian SuperGlue.