code wiki / (root) / nx_gpu_export.nx

nx_gpu_export.nx

buildroot/runtime/nx_gpu_export.nx

9508 B131 linesdepth 8pulls 18 transitivereach 0 importersview sourcekind tooltopic gpu
docsdependenciesstructsconstsfunctions

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

nx_syscalls.nx nx_tier.nx nx_le.nx nx_tensor.nx nx_gguf.nx nx_gguf_load.nx nx_gguf_meta.nx nx_nofloat_llm.nx nx_nofloat_tok.nx nx_gpu_export.nx

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

main ge_puts sys_write sys_mmap sys_read_file sys_openat_rd sys_lseek sys_mmap ↻ sys_read sys_close nx_gguf_parse nx_le_read_u32 nx_le_read_u64 nx_le_read_u32 ↻ _gguf_skip_value nx_le_read_u64 ↻ nx_le_read_u32 ↻ _gguf_skip_value ↻ sys_mmap ↻ ge_u32 sys_mmap ↻ nx_gguf_meta_find nx_le_read_u64 ↻ nx_le_read_u32 ↻ _gmeta_bytes_equal nx_gguf_skip_value _gguf_skip_value ↻ ge_slen nx_gguf_meta_read_u32 nx_le_read_u32 ↻ isqrt ge_num sys_mmap ↻ sys_write ↻ nx_gguf_meta_find ↻ ge_slen ↻ nx_gguf_meta_array_count nx_le_read_u64 ↻ nx_gguf_meta_array_first_e nx_gguf_find_tensor

structs

none

consts

14const K_MAGIC_65536: i64 = 65536
15const K_MAGIC_5108: i64 = 5108

functions

17func 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 }
called by 1: main calls 1: sys_write
18func 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 }
called by 1: main calls 2: sys_mmapsys_write
19func ge_slen(s: *u8) -> i64 { var n: i64=0; while s[n]!=(0 as u8){n=n+1} return n }
called by 2: ge_u32main
21func ge_u32(buf: *u8, flen: i64, hdr: *NxGgufHeader, key: *u8) -> i64
28func 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 }
called by 2: ge_w1main calls 1: sys_write
29func 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) }
called by 1: main calls 2: sys_mmapge_wall
31func main() -> i64