code wiki / (root) / nx_llm_diff_probe.nx

nx_llm_diff_probe.nx

buildroot/runtime/nx_llm_diff_probe.nx

5344 B112 linesdepth 13pulls 54 transitivereach 0 importersview sourcekind probetopic llm
docsdependenciesstructsconstsfunctions

about

nx_llm_diff_probe.nx -- NUMERICAL cross-check vs the transformers oracle (same GGUF). Dumps our forward's last-token ("is") hidden-state L2-norm after embed + each of the 24 layers, as milli-units (x1000) for direct compare to the reference norms [390, 4468, 7320, ... 279079]. The FIRST layer whose norm diverges localizes the residual bug. Also prints embed[0..6] + after-L0[0..6] as f32 hex (ref embed ~[-0.005,0.015,-0.0067,...], ref afterL0 ~[0.2343,-0.093,0.1314,...]). Fully sovereign forward; oracle is a throwaway debug reference. exp_exit:0

dependencies 16 imports · 0 importers

nx_syscalls.nx nx_tier.nx nx_gguf.nx nx_gguf_load.nx nx_gguf_meta.nx nx_f32.nx nx_f32_rmsnorm.nx nx_f32_kv_cache.nx nx_f32_lazy_weight.nx nx_f32_llama_block.nx nx_llm_diff_probe.nx

diagram shows first 10 each side; +6 more imports, +0 more importers in the complete lists below.

imports: nx_syscalls.nxnx_tier.nxnx_gguf.nxnx_gguf_load.nxnx_gguf_meta.nxnx_f32.nxnx_f32_rmsnorm.nxnx_f32_kv_cache.nxnx_f32_lazy_weight.nxnx_f32_llama_block.nxnx_f32_llama_block_v4.nxnx_f32_llama_stack_v4.nxnx_f32_llama_layer_lazy_load.nxnx_f32_llm.nxnx_f32_llm_v4.nxnx_f32_llm_read_dims.nx

imported by: nobody (leaf or entry point)

call flow from main pre-order; caps 40 nodes / depth 6 declared; ↻ = already shown

main 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 ↻ nx_f32_llama_model_alloc sys_mmap ↻ nx_f32_llm_read_dims_from_ sys_mmap ↻ nx_gguf_meta_find nx_le_read_u64 ↻ nx_le_read_u32 ↻ _gmeta_bytes_equal nx_gguf_skip_value _gguf_skip_value ↻ nx_gguf_meta_read_string_l nx_le_read_u64 ↻ nx_gguf_meta_read_string_p _fld_read_arch_u32 sys_mmap ↻ _fld_concat_key nx_gguf_meta_find ↻ nx_gguf_meta_read_u32 nx_le_read_u32 ↻ _fld_read_vocab_size_from_ sys_mmap ↻ nx_gguf_find_tensor nx_gguf_tensor_at

structs

none

consts

none

functions

23func pw(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: dump8main calls 1: sys_write
24func pn(v: i64) -> i64 { let bb: *u8=sys_mmap(28); var m: i64=v; if m<0{sys_write(1,"-" as *u8,1);m=0-m} 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{bb[i]=t[k-1-i];i=i+1} sys_write(1,bb,k); return 0 }
called by 1: main calls 2: sys_mmapsys_write
25func ph(v: i64) -> i64 { let b: *u8=sys_mmap(8); var i: i64=0; while i<8 { let nib: i64=(v>>((7-i)*4))&0xF; if nib<10 {b[i]=(48+nib) as u8} else {b[i]=(87+nib) as u8} i=i+1 } sys_write(1,b,8); return 0 }
called by 1: dump8 calls 2: sys_mmapsys_write
28func milli(v: i64) -> i64
called by 1: main calls 1: nx_f32_mul
40func l2norm(h: *i64, n: nx_int) -> i64
called by 1: main calls 3: nx_f32_addnx_f32_mulnx_f32_sqrt
46func dump8(label: *u8, h: *i64) -> i64
called by 1: main calls 2: pwph
53func main() -> i64
calls 14: sys_mmapsys_read_filenx_gguf_parsenx_f32_llama_model_allocnx_f32_llm_read_dims_from_ggufnx_f32_llm_load_weights_v4_from_gguf+8