Replace VarB: v3k_iq2xs_q5k (combined 3.051 BPW, fp16 cos 0.894 vs B 0.870). Q3_K vision + IQ2_XS/Q5_K HSIC LLM + 20-ep Fisher.

Upload pi05.gguf with huggingface_hub

Upload tokenizer.model with huggingface_hub

Upload norm_stats.json with huggingface_hub

initial commit