nx_embed_gpu_live_gate.nx
buildroot/runtime/nx_embed_gpu_live_gate.nx
about
nx_embed_gpu_live_gate.nx -- e4 LIVE gate: the deployed :8033 daemon's /embed must return the
BYTE-IDENTICAL Q24 vector for mode "i8" (CPU) and mode "gpu" (resident-weight GPU server via the
unix-socket client in serve core). Also asserts the response labels mode:"gpu" and equal ntok.
Requires: coder daemon relaunched on the new serve core + gpu_embed_serve --serve up.
usage: nx_embed_gpu_live_gate [port] (default 8033)
license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0
dependencies 1 imports · 0 importers
imports: nx_code_index.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
| 9 | const EGL_PORT_DEFAULT: i64 = 8033 // the deployed coder daemon's /embed port |
| 10 | const EGL_BODY_CAP: i64 = 4096 // request-body scratch (JSON is well under this) |
| 11 | const EGL_RESP_CAP: i64 = 65536 // /embed response buffer (Q24 vector + labels) |
| 12 | const EGL_ERRDUMP: i64 = 300 // bytes of the failing response to echo on RED |
| 13 | const EGL_I64_BYTES: i64 = 8 // sizeof(i64) -- byte size of one vector element / pointer slot |
| 14 | const EGL_ERR_POST: i64 = 1 // lb_embed sentinel: /embed POST returned nothing |
| 15 | const EGL_ERR_PARSE: i64 = 2 // lb_embed sentinel: response vector length != CI_NE |
functions
| 17 | func lb_body(mode: *u8, out: *u8) -> i64 called by 1: lb_embed |
| 31 | func lb_embed(port: i64, mode: *u8, vec: *i64, resp: *u8, rescap: i64) -> i64 |
| 42 | func main(argc: i64, argv: *i64) -> i64 |