nx_gpu_export.nx
buildroot/runtime/nx_gpu_export.nx
about
nx_gpu_export.nx -- dump the no-float Qwen2.5-0.5B i8 caches + i32 head + output_norm gamma + freqs +
config + a sequential-decode CPU REFERENCE (input embeddings, final normed, next-token id) to a binary
that the GPU forward server loads. Reuses the PROVEN load/dequant lib (eat-the-debt: DUMP the built
cache, do NOT re-parse GGUF). One-shot offline tool. No hw writes (Rule 26). 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_tok.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
| 14 | const K_MAGIC_65536: i64 = 65536 |
| 15 | const K_MAGIC_5108: i64 = 5108 |
functions
| 17 | func ge_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 ge_num(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 } |
| 19 | func ge_slen(s: *u8) -> i64 { var n: i64=0; while s[n]!=(0 as u8){n=n+1} return n } |
| 21 | func ge_u32(buf: *u8, flen: i64, hdr: *NxGgufHeader, key: *u8) -> i64 |
| 28 | func ge_wall(fd: i64, buf: *u8, count: i64) -> i64 { var off: i64=0; while off<count { let w: i64=sys_write(fd, ((buf as i64)+off) as *u8, count-off); if w<=0 { return 0-1 } off=off+w } return 0 } |
| 29 | func ge_w1(fd: i64, v: i64) -> i64 { let p: *i64=sys_mmap(8) as *i64; p[0]=v; return ge_wall(fd, p as *u8, 8) } |
| 31 | func main() -> i64 |