code wiki / _hdl_build / nx_llm_loopb_gate.nx
nx_llm_loopb_gate.nx
buildroot/runtime/_hdl_build/nx_llm_loopb_gate.nx
about
nx_llm_loopb_gate.nx -- THE LOOP-B CLOSURE: load OUR sovereignly-trained model (nx_f32_qwen2_train_gate's
/tmp/nx_ours_qwen.gguf) via the arch-config NO-FLOAT inference and GENERATE -- proving train (ours, f32 tape)
-> gguf-export -> serve (ours, 100pct-integer) end to end. If our integer inference reproduces the sequence
the f32 trainer learned ('the quick brown fox...'), the whole sovereign loop is closed: we train a model AND
run it, all our own stack, no PyTorch, no CUDA, no float lib on the serve path. Dims come from the gguf
metadata (arch-config), so nothing is hardcoded to this specific model.
T1 arch-config load reads OUR metadata (qwen2 d=16 layers=2 1-head).
T2 all 27 tensors present + loadable by name.
T3 GENERATE on the no-float integer forward reproduces the trained corpus (>=30/44 chars) = LOOP-B CLOSED.
expect_exit: 0 license_tier: ORIGINAL
dependencies 9 imports · 0 importers
imports: nx_syscalls.nxnx_tier.nxnx_le.nxnx_tensor.nxnx_gguf.nxnx_gguf_load.nxnx_gguf_meta.nxnx_nofloat_llm.nxnx_nofloat_arch.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
| 21 | func lw(s: *u8) -> i64 { var n: i64=0; while s[n]!=(0 as u8){n=n+1} sys_write(1,s,n); return 0 } |
| 22 | func ln2(v: i64) -> i64 { let b: *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 as u8;k=1} while m>0{t[k]=(48+(m%10)) as u8;m=m/10;k=k+1} var i: i64=0; while i<k{b[i]=t[k-1-i];i=i+1} sys_write(1,b,k); return 0 } |
| 23 | func lslen(s: *u8) -> i64 { var n: i64=0; while s[n]!=(0 as u8){n=n+1} return n } called by 1: main |
| 25 | func main() -> i64 |