nx_forge_local_gen.nx
buildroot/runtime/nx_forge_local_gen.nx
about
nx_forge_local_gen.nx -- L4 generation phase: the LOCAL Coder-0.5B (integer no-float tower)
writes fbench candidates. nsv_init(coder) ONCE, then per task: prompt = mini-pack + task
instruction -> nsv_generate (mode 0 i32 greedy) -> fence-strip -> write candidate file
IMMEDIATELY (so a mid-run reap preserves finished tasks). Prints per-task nprompt/ngen: an
nprompt > 384 also PROVES the L2 capacity raise (the pack prompt could not fit the old cap).
Phase 2 = the existing nx_forge driver (fg_run) judges these candidate files byte-exact.
license_tier: ORIGINAL expect_exit: 0
dependencies 2 imports · 0 importers
imports: nx_nofloat_serve_core.nxnx_forge_ctx.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
| 10 | const K_MAGIC_32768: i64 = 32768 |
| 11 | const K_MAGIC_16384: i64 = 16384 |
| 12 | const K_MAGIC_30000: i64 = 30000 |
functions
| 15 | func lg_strip(src: *u8, n: i64, out: *u8) -> i64 called by 1: lg_task |
| 61 | func lg_prompt(taskinstr: *u8, buf: *u8, cap: i64) -> i64 called by 1: lg_task |
| 84 | func lg_read_pack(buf: *u8, cap: i64) -> i64 |
| 103 | func lg_prompt_mp(taskinstr: *u8, buf: *u8, cap: i64) -> i64 |
| 128 | func lg_task(taskinstr: *u8, candpath: *u8, tnum: i64, mp: i64, maxnew: i64) -> i64 |
| 192 | func main(argc: i64, argv: *i64) -> i64 |