metadata
license: llama2
datasets:
- evilfreelancer/toxicator-ru
language:
- ru
tags:
- toxify
- detoxify
- seq2seq
pipeline_tag: translation
LLaMA 2 7B - Toxicator RU
This fine-tuned model based on meta-llama/Llama-2-7b-chat-hf, it utilizes evilfreelancer/toxicator-ru dataset created from samples in s-nlp/russe_detox_2022 project.
Model was tuned just for lulz for experimenting with TorchTune tool.
100 examples on GitHub Gist.
Links
- https://github.com/EvilFreelancer/toxicator-ru - GitHub repository with train scripts and scripts for generating dataset
- https://huggingface.co/datasets/evilfreelancer/toxicator-ru - dataset
- https://api.wandb.ai/links/evilfreelancer/33t8pqze - wandb report about training