topic: quant

10 modules sharing the quant name family (derived from the tree's prefix discipline).

The 'quant' topic family in the Nishi sovereign ecosystem focuses on efficient AI inference through quantization techniques. nx_quant provides foundational primitives, while nx_quant_block implements 4-bit block quantization on an i64 substrate, and nx_quant_block_q8 supports 8-bit block quantization with q8_0 shape, enabling memory-efficient execution. These modules work together to optimize precision and performance in neural network operations.

auto-narrated by the local model from this topic's module headers; links verified against the wiki index.

narrated overview -- maintained by the narration lane, module links verified against this wiki.

module	description	lines	funcs
nx_quant.nx	quantization primitives for AI inference.	661	24
nx_quant_block.nx	4-bit block quantization on i64 substrate.	256	10
nx_quant_block_q8.nx	8-bit block quantization (q8_0 shape).	301	11
nx_quant_block_test.nx	algo-led memory-win correctness.	149	1
nx_quant_calibration_gate.nx	the UNGLAMOROUS production engineering an integer stack actually needs:	84	7
nx_quant_policy.nx	per-layer mixed-precision quantization policy.	364	10
nx_quant_q4k.nx	hierarchical k-quants (ggml q4_K shape).	277	10
nx_quant_q4k_test.nx	algo-led correctness; q4_K BEATS q4_0 on	197	1
nx_quant_table.nx	perceptual quantisation matrices. The two	136	8
quant.nx	quantization primitives for AI inference.	462	18