nx_nofloat_qwen_tok_gate.nx
buildroot/runtime/nx_nofloat_qwen_tok_gate.nx
about
nx_nofloat_qwen_tok_gate.nx -- TOKENIZER CHECK (fast, no forward): BPE-encode a multi-word prompt and print the
token ids + their decoded pieces. Hypothesis under test: the byte-level space map (0x20 -> U+0120 'Ġ') is missing,
so multi-word prompts tokenize wrong -> the forward sees garbage input -> garbage output (NOT a forward bug).
Correct Qwen2.5 tokenization of "The capital of France is" = [785, 6722, 315, 9625, 374] = The/Ġcapital/Ġof/ĠFrance/Ġis.
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_tok.nxnx_gate_verdict.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
| 23 | const TG_KAT_N: i64 = 5 |
| 24 | const TG_KAT_0: i64 = 785 |
| 25 | const TG_KAT_1: i64 = 6722 |
| 26 | const TG_KAT_2: i64 = 315 |
| 27 | const TG_KAT_3: i64 = 9625 |
| 28 | const TG_KAT_4: i64 = 374 |
| 29 | const TG_MAXT: i64 = 32 |
| 30 | const TG_SLOT: i64 = 8 |
functions
| 32 | func tg_kat_match(ids: *i64, n: i64) -> i64 called by 1: main |
| 42 | func tg_puts(s: *u8) -> i64 { var n: i64=0; while s[n]!=(0 as u8){n=n+1} sys_write(1,s,n); return 0 } |
| 43 | func tg_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 } |
| 44 | func tg_slen(s: *u8) -> i64 { var n: i64=0; while s[n]!=(0 as u8){n=n+1} return n } |
| 46 | func tg_encode_print(buf: *u8, mfirst: i64, nm_c: i64, vfirst: i64, vocab: i64, input: *u8) -> i64 |
| 59 | func main() -> i64 |