BitNet b1.58-large TQ2_0

Run (UTC)	geist	llama.cpp	pp512 tok/s	vs llama	tg64 tok/s	vs llama	Files
2026-10-12 00:38	4d6e2696a	d81235049	2232.3	+61 %	276.6	+48 %	report · raw
2026-10-09 13:01	53aacaacb	d81235049	2207.1	+61 %	270.4	+47 %	report · raw

Every run of this cell, newest first. pp512: prefill of 512 tokens; tg64: decode of 64 tokens at depth 512; median of 12 samples. The raw file holds every sample and the full host, engine and model provenance. Back to the matrix.