Intelligence has weight.
Each block on the floor holds exactly 100 million model weights. The last block holds the exact remainder. Change precision and the same weights physically repack.
Weight precision16-bit
The adaptation view keeps the base model’s measured scale as a reference; it does not measure descendant files. Storage is raw weight data: parameters × bits ÷ 8, decimal GB. It excludes inference runtime, KV cache, temporary buffers and operating-system use. Precision changes may affect quality, speed and hardware support.