nx_zimage_gguf_arch.nx
buildroot/runtime/nx_zimage_gguf_arch.nx
about
nx_zimage_gguf_arch.nx -- extract the REAL Z-Image text-encoder architecture from its GGUF, sovereignly.
sd-server -> Nishi migration: the real-scale assembly needs the REAL dims to scale our f32 organs to.
This reads the actual `Z-Image_Qwen_3_4b-Q6_K.gguf` and pulls token_embd.weight's dims (hidden, vocab)
+ ggml quant type + tensor_count from the parsed header (metadata at the front -> bounded prefix read,
no multi-GB tensor materialization), and writes them to /tmp/zimg_arch.txt for inspection. Composes the
proven sovereign GGUF reader. Gate: opens, GGUF v3, token_embd found, hidden>0.
license_tier: ORIGINAL
dependencies 10 imports · 0 importers
imports: nx_syscalls.nxnx_tier.nxnx_le.nxnx_strconv.nxnx_tensor.nxnx_gguf.nxnx_gguf_load.nxnx_gguf_meta.nxnx_placement.nxnx_gguf_load_lazy.nx
imported by: nobody (leaf or entry point)
call flow from main pre-order; caps 40 nodes / depth 6 declared; ↻ = already shown
structs
| none |
consts
| none |
functions
| 20 | func za_emit_kv(fd: i64, key: *u8, key_len: i64, value: i64) -> i64 |
| 35 | func za_blk_name(out: *u8, n: i64, suffix: *u8, suffix_len: i64) -> i64 |
| 55 | func za_emit_dims(ofd: i64, hdr: *NxGgufHeader, label: *u8, label_len: i64, suffix: *u8, suffix_len: i64) -> i64 |
| 74 | func main() -> i64 |