Very happy to announce SlovenianGPT-Instruct! 🥳 The model significantly improves upon the base model across all of the benchmarks and is by far the best LLM in the 7B category for the Slovenian language! Over the next couple of days SlovenianGPT-Instruct will be available for testing through the yugochat.com app. Really looking forward to feedback from native speakers! :) Gemma and some other Slovenian LLMs are not included as they're even weaker than Mistral/LLaMA. The results are likely favorable against much bigger models as well, will do additional evals to confirm. I'm still looking for sponsors to improve the Slovenian LLM evaluation (huggingface.co/datasets/gordi…) if you're willing to help and get a company mention dm me! Big thank you to: * @Hyperstackcloud for providing H100s and @togethercompute for providing A100s that we'd be using to train internal align models * @nljubesic for help providing Slovenian data! I'll be open-sourcing the base model over the next period as well so stay tuned.
@gordic_aleksa What is the base model for SlovenianGPT-Instruct?
@gordic_aleksa Impressive work Aleksa! Looking forward to the open-sourcing of the base model 🙌
@gordic_aleksa @Hyperstackcloud Important work to protect a culture in the age of AI