Vikhr-7b-0.1 / README.md
AlexWortega's picture
Update README.md
955a0de
|
raw
history blame
400 Bytes
metadata
license: apache-2.0
datasets:
  - IlyaGusev/habr
language:
  - ru
  - en
library_name: transformers

Mistral Based russian language model, pretrained on 400m tokens on 3 epochs over original mistral

dataset AlexWortega/Vikhr-7b-0.1 mistralai/Mistral-7B-v0.1
mmlu_ru 0.59 0.49
xwinograd 0.7443 0.681
xnli 0.3980 0.3691