Strangedove/CYFRAGOVPL_Llama-PLLuM-70B-instruct-EmbedFix

Strangedove/CYFRAGOVPL_Llama-PLLuM-70B-instruct-EmbedFix is a 70 billion parameter Llama-based instruction-tuned model with a 32768 token context length. This model features a specific modification where added tokens were moved to pre-allocated spaces in the tokenizer, restoring the original embeddings size. This technical adjustment aims to maintain embedding integrity, making it suitable for applications sensitive to tokenizer modifications.