A spore is a portable knowledge artifact: a reference manual run once through a
frozen LLM, with the resulting KV-cache quantized (int4 per-channel) and compressed
(LZMA, ~5.5× vs fp16). At inference time the spore is prepended as a KV prefix, so the
model "remembers" the manual without the corpus occupying context tokens.
Spores are model-specific. This bank binds to… See the full description on the dataset page:
https://huggingface.co/datasets/lucas-mella/daimon-spores.