Update README.md

generate.py: nrng decoding defaults (temp 0.7, rep-pen 1.15) -- best per decode_lab_chat

Format-blended SFT: chat fluency + preserved benchmark formats

Harness: no-repeat-ngram(3) flag, default on (A/B'd per model on real weights, rep-4 -> 0.000 at deployed settings)

Upload chkpt/quazimoto_sft.pt with huggingface_hub

Upload chkpt/quazimoto.pt with huggingface_hub

Upload fractal_phase.pt with huggingface_hub

Upload tokenizer.json with huggingface_hub

Upload visualize.py with huggingface_hub

Upload chat_sft.py with huggingface_hub

Upload generate.py with huggingface_hub

Upload special_tokens.py with huggingface_hub

Upload spike_tokenizer.py with huggingface_hub

Upload fractal.py with huggingface_hub

Upload instrument.py with huggingface_hub

Upload family.py with huggingface_hub

Upload chimera.py with huggingface_hub

Upload model.py with huggingface_hub

Upload README.md with huggingface_hub

initial commit