nx_nofloat_speed_bench.nx
buildroot/runtime/nx_nofloat_speed_bench.nx
about
nx_nofloat_speed_bench.nx -- honest tok/s baseline for the no-float Qwen serve (the inference-speed arc's
"measure before you optimize" foundation). nsv_generate already self-times (meta[1]=gen_tokens,
meta[3]=ms_per_token); this drives it with a fixed prompt + max_new and prints prompt-tokens, gen-tokens,
ms/tok, and tok/s for BOTH i32-lossless (mode 0) and i8-fast (mode 1) so the real number + the i8 speed
delta are on the record (memory had conflicting 3.2 vs 14.9 tok/s). Low-risk: reads the model, no serve edits.
usage: nx_nofloat_speed_bench [model.gguf] [max_new] Sovereign: nx_nofloat_serve_core. license_tier: ORIGINAL
dependencies 3 imports · 0 importers
imports: nx_syscalls.nxnx_itoa_lib.nxnx_nofloat_serve_core.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
| 11 | const SB_MODEL: *u8 = "/home/elderwesto/nx_stage/nx_real_model.gguf" |
functions
| 13 | func sb_puts(s: *u8) -> i64 { var n: i64=0; while s[n] != (0 as u8) { n=n+1 } sys_write(1, s, n); return 0 } |
| 18 | func sb_putn(v: i64) -> i64 { nxi_out(v); return 0 } |
| 19 | func sb_slen(s: *u8) -> i64 { var n: i64=0; while s[n]!=(0 as u8){ n=n+1 } return n } called by 1: sb_run |
| 20 | func sb_int(s: *u8) -> i64 { var v: i64=0; var i: i64=0; while s[i]!=(0 as u8){ let c: i64=s[i] as i64; if c>=48 { if c<=57 { v=v*10+(c-48) } } i=i+1 } return v } called by 1: main |
| 22 | func sb_run(mode: i64, label: *u8, maxnew: i64, pr: *u8, out: *u8, meta: *i64, gp: *i64) -> i64 |
| 41 | func main(argc: i64, argv: *i64) -> i64 |