This project is based on Llama2, and we open-source Tibetan-Llama2 (foundation model) and Tibetan-Alpaca (instruction-following model). These models have been expanded and optimized with Tibetan vocabulary, surpassing the original Llama-2. We utilized a considerable amount of Tibetan data for incremental pre-training, which further enhanced the fundamental semantic understanding of the Tibetan language. The relevant models support a 4K context and can be expanded up to 18K+ using the NTK method.
The main contents of this project include:
🚀 New extended Tibetan vocabulary beyond Llama2, open-sourcing the Tibetan-Llama2 and Tibetan-Alpaca LLMs.
🚀 Quickly deploy and experience the quantized LLMs on CPU/GPU of personal PC.
🚀 Support for Llama ecosystems like 🤗transformers, Llama.cpp, text-generation-webui, LangChain, vLLM etc.
