topic: gguf
20 modules sharing the gguf name family (derived from the tree's prefix discipline).
The 'gguf' topic family in the Nishi sovereign ecosystem focuses on parsing, loading, and testing GGUF v3 model files, which are used to store quantized neural network weights. Modules like nx_gguf provide structural parsing, while nx_gguf_load handles tensor lookup and dequantization. nx_gguf_fixture_tiny supports reusable fixture creation for testing, ensuring consistency across modules. These components work together to enable efficient model loading and validation within the NishiLang framework.
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_gguf.nx | GGUF v3 model-file structural parser. | 426 | 6 |
| nx_gguf_dequant_kat_gate.nx | KATs for the DEBT-EATEN quant family (2026-07-15): every newly added | 180 | 4 |
| nx_gguf_fixture_tiny.nx | reusable tiny-Llama GGUF fixture builder. | 297 | 7 |
| nx_gguf_fixture_tiny_test.nx | smoke for the fixture builder. | 71 | 1 |
| nx_gguf_load.nx | GGUF tensor lookup + dequantize composer. | 747 | 18 |
| nx_gguf_load_block.nx | per-layer Llama-class weight bundle loader. | 325 | 4 |
| nx_gguf_load_block_test.nx | smoke for nx_gguf_load_block.nx. | 172 | 2 |
| nx_gguf_load_f32.nx | load a GGUF tensor and dequantize to f32 bits. | 110 | 2 |
| nx_gguf_load_f32_test.nx | smoke for nx_gguf_load_f32.nx. | 89 | 1 |
| nx_gguf_load_lazy.nx | Phase B of conductor arc. | 119 | 2 |
| nx_gguf_load_lazy_test.nx | smoke for nx_gguf_load_lazy.nx + | 158 | 1 |
| nx_gguf_load_model.nx | top-level model-tensor loader. | 147 | 2 |
| nx_gguf_load_model_test.nx | smoke for nx_gguf_load_model.nx. | 185 | 2 |
| nx_gguf_load_q4k_test.nx | focused smoke for Q4_K dequant. | 146 | 1 |
| nx_gguf_load_test.nx | smoke for nx_gguf_load.nx. | 164 | 1 |
| nx_gguf_meta.nx | GGUF metadata key-value walker + typed readers. | 238 | 18 |
| nx_gguf_meta_test.nx | smoke for nx_gguf_meta.nx. | 159 | 2 |
| nx_gguf_type_census_gate.nx | FACTS FIRST for the dequant-debt rung (2026-07-15): enumerate every | 67 | 4 |
| nx_gguf_typecensus.nx | tally the real model's tensors by ggml_type, | 45 | 1 |
| nx_gguf_write_gate.nx | 131 | 10 |