nx_effbench_research_fetch.nx
buildroot/runtime/nx_effbench_research_fetch.nx
about
nx_effbench_research_fetch.nx -- SOVEREIGN researcher: GROUND the "are we S-class EXCEED vs Unsloth & friends"
benchmark in REAL fetched facts (Rule 4: benchmarks need VERIFIED competitor claims, never assumptions). The
incumbents = efficient LLM training/inference frameworks (Unsloth, llama.cpp, QLoRA, GPTQ, AWQ, bitsandbytes/
LLM.int8, LoRA). We measure our no-float Q16 stack against THEIR real claims (speedup, memory, bit-width) so
the EXCEED verdict is measured + honest (no-wave), not a floor-claim.
Mirrors the proven fetcher: sovereign TLS-1.3 (nx_https_fetch_follow) + Mozilla CA, idempotent skip-if-have,
saves each source to knowledge/fetched/eff_*.raw so the corpus is greppable to CITE. expect_exit: 0
dependencies 4 imports · 0 importers
imports: nx_syscalls.nxnx_x509_trust_store.nxnx_trust_store_load_from_certdata.nxnx_https_fetch_follow.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
| 12 | const K_MAGIC_4194304: i64 = 4194304 |
| 13 | const K_MAGIC_8388608: i64 = 8388608 |
functions
| 15 | func df_puts(s: *u8) -> i64 { var n: i64 = 0; while s[n] != (0 as u8) { n = n + 1 } sys_write(1, s, n); return 0 } |
| 16 | func df_putn(v: i64) -> i64 |
| 26 | func have_file(path: *u8) -> i64 { let fd: i64 = sys_openat_rd(path); if fd < 0 { return 0 } sys_close(fd); return 1 } |
| 27 | func fetch_save(url: *u8, opath: *u8, store: *TrustStore, out: *u8, cap: i64) -> i64 |
| 43 | func main() -> i64 |