code wiki / _hdl_build / nx_vram_budget_test.nx

nx_vram_budget_test.nx

buildroot/runtime/_hdl_build/nx_vram_budget_test.nx

4076 B60 linesdepth 3pulls 3 transitivereach 0 importersview sourcekind gate/prooftopic vram
docsdependenciesstructsconstsfunctions

about

nx_vram_budget_test.nx -- the team REASONS about VRAM: a real 7B model on the operator's 16GB RTX 5080. Proves the investigation's core finding: fp16 does NOT fit, but accounting for ALL components and reducing each under the quality floor (weights Q4 + KV Q8) fits with huge headroom -- which BUYS much longer context (improving functionality, not just shrinking memory). Exit 0 only if all hold. license_tier: ORIGINAL

dependencies 2 imports · 0 importers

nx_vram_budget.nx nx_syscalls.nx nx_vram_budget_test.nx

imports: nx_vram_budget.nxnx_syscalls.nx

imported by: nobody (leaf or entry point)

call flow from main pre-order; caps 40 nodes / depth 6 declared; ↻ = already shown

main vt_puts sys_write vram_weights_mb vram_kv_mb vram_config_mb vram_weights_mb ↻ vram_kv_mb ↻ vram_total_mb vt_num sys_mmap sys_write ↻ vram_fits sys_mmap ↻ vram_best_config sys_mmap ↻ vram_weight_loss vram_kv_loss vram_config_mb ↻ vram_fits ↻ sys_exit

structs

none

consts

none

functions

10func vt_puts(s: *u8) -> i64 { var n: i64 = 0; while s[n] != (0 as u8) { n = n + 1 } sys_write(1, s, n); return 0 }
called by 1: main calls 1: sys_write
11func vt_num(v: i64) -> i64 { let bb: *u8 = sys_mmap(28); var m: i64=v; if m<0 {m=0-m; sys_write(1,"-" as *u8,1)}; let t: *u8 = sys_mmap(28); var k: i64=0; if m==0 {t[0]=48;k=1}; while m>0 {t[k]=48+(m%10); m=m/10; k=k+1}; var i: i64=0; while i<k {bb[i]=t[k-1-i]; i=i+1}; sys_write(1, bb, k); return 0 }
called by 1: main calls 2: sys_mmapsys_write
13func main() -> i64