Add model card

Add self-contained model.py (class, loader, generate CLI)

Correct architecture note: it is a fine-tune of Glint-2 (not a frozen-base LoRA); do not claim the shared block is untouched

Add model card

Add config

Add honest model card (corrected training numbers + known limitation) (#1)

Add standalone inference (torch + tokenizers + safetensors)

Add architecture config (max_loops=16 table capacity, run at 8)

Add Glint-2 BPE tokenizer (vocab 4096)

Add fine-tuned weights (1,064,737 params, F32)

Add honest model card for catgirl-1m (Glint-2 tsundere-catgirl fine-tune)

initial commit