# Multilingual Mixed Training

A Japanese-optimized model based on Meta-Llama-3-8B, continuously pretrained on a mixed dataset of Japanese and English with 22 billion tokens

A 7-billion-parameter Bengali large language model optimized based on the LLaMA-2 architecture, supporting instruction-following tasks

DiscoLM German 7B v1 is a 7B-parameter German language model based on the Mistral architecture, supporting both German and English, and released under the Apache-2.0 license.

A Vietnamese-optimized model based on Llama2-chat 7B, significantly improving Vietnamese language processing through incremental pre-training and an efficient tokenizer

KoAlpaca is the Korean version of the Stanford Alpaca model, combining the LLAMA architecture with Polyglot-ko technology, specifically optimized for Korean text generation tasks.