nx_tok_probe.nx
buildroot/runtime/nx_tok_probe.nx
about
nx_tok_probe.nx -- tokenizer-parity DIAGNOSTIC: dump the exact token ids
(and each id's decoded string) that nx_bpe_encode_bytelevel produces for an
arithmetic prompt vs a known-good word prompt. Loads GGUF vocab ONLY (no
weights) -- fast. Answers mechanically whether digits are being merged
(Qwen2.5 pre-tokenizer splits every digit into its OWN token; a merged
"47" would be a context the model never saw in arithmetic).
expect_exit: 0
dependencies 9 imports · 0 importers
imports: nx_syscalls.nxnx_tier.nxnx_le.nxnx_strconv.nxnx_bpe.nxnx_gguf.nxnx_gguf_load.nxnx_f32_bpe_load.nxnx_reasoning.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
| none |
functions
| 19 | func tp_w(s: *u8) -> i64 { var n: i64 = 0; while s[n] != (0 as u8) { n = n + 1 } sys_write(1, s, n); return 0 } |
| 20 | func tp_wn(v: i64) -> i64 |
| 28 | func tp_wtext(b: *u8, n: i64) -> i64 |
| 41 | func tp_dump(v: *NxBpeVocab, label: *u8, text: *u8, n: nx_int) -> i64 |
| 63 | func main() -> i64 |