code wiki / _hdl_build / nx_intfp_lora_gradcheck_gate.nx

nx_intfp_lora_gradcheck_gate.nx

buildroot/runtime/_hdl_build/nx_intfp_lora_gradcheck_gate.nx

5952 B81 linesdepth 2pulls 2 transitivereach 0 importersview sourcekind gate/prooftopic intfp

docs dependencies structs consts functions

about

nx_intfp_lora_gradcheck_gate.nx -- LoRA / parameter-efficient fine-tuning in Q20 INTEGER, gradchecked, no float. modelwright census marks LoRA ABSENT; this closes it. Base weight W is FROZEN; a low-rank adapter dW = A@B (A[DM,r], B[r,DM], rank r << DM) is the only trainable part -> tiny # of trained params. y = x@W + (x@A)@B. Train ONLY A,B. Gradcheck A and B (the adapter). Composes proven integer matmuls. license_tier: ORIGINAL

dependencies 1 imports · 0 importers

imports: nx_syscalls.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 S: i64 = 1048576 // Q20
12	const T: i64 = 3
13	const DM: i64 = 6
14	const R: i64 = 2 // LoRA rank (<< DM)

functions

7	func w(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 2: gcheck main calls 1: sys_write
8	func wn(v: i64) -> i64 { if v==0 { sys_write(1,"0" as u8,1); return 0 } var m: i64=v; if m<0{sys_write(1,"-" as u8,1);m=0-m} let t: u8=sys_mmap(24); var k: i64=0; while m>0{t[k]=(48+(m%10)) as u8;m=m/10;k=k+1} let o: u8=sys_mmap(24); var q: i64=k-1; var i: i64=0; while q>=0{o[i]=t[q];i=i+1;q=q-1} sys_write(1,o,i); return 0 } called by 2: gcheck main calls 2: sys_write sys_mmap
9	func iabs(v: i64) -> i64 { if v<0 { return 0-v } return v } called by 1: gcheck
17	func lora_fwd(P: *i64) -> i64 called by 2: gcheck main
29	func lora_bwd(P: *i64) -> i64 called by 1: main calls 1: sys_mmap
42	func gcheck(name: u8, P: i64, Wt: i64, dW: i64, ncell: i64) -> i64 called by 1: main calls 4: iabs lora_fwd w wn
59	func main() -> i64 calls 6: w wn sys_mmap lora_fwd lora_bwd gcheck