language: - en base_model: - meta-llama/Llama-3.2-3B-Instruct
Dataset : llama tokenizer 4000 distribution : 80% 13% 7% only for attributes
