Add model card with exact and within-1 confusion matrices and per-class metrics for non-fine-tuned LLaMA evaluation

Upload non-fine-tuned LLaMA-3-8B CEFR evaluation with exact and within-1 confusion matrices

initial commit