nx_batched_gate.nx
buildroot/runtime/nx_batched_gate.nx
about
nx_batched_gate.nx -- MEASURED gate for BATCHED MULTI-SEQUENCE decode
(nx_f32_llama_v4b) on the REAL model.
EQUIV batched decode == sequential fork decode BIT-IDENTICAL:
prefill once, fork N=3 twice (same seeds); decode one set
sequentially (v4p, m=1 per fork) and one set BATCHED (v4b,
one M-row forward per round); texts must match byte-for-byte
SPEED decode phases timed separately from the shared prefill:
sequential N*T m=1 forwards vs T M-row forwards
HYGIENE pool returns to all-free
license_tier: ORIGINAL expect_exit: 0
dependencies 26 imports · 0 importers
diagram shows first 10 each side; +16 more imports, +0 more importers in the complete lists below.
imports: nx_syscalls.nxnx_tier.nxnx_le.nxnx_bpe.nxnx_gguf.nxnx_gguf_load.nxnx_gguf_meta.nxnx_f32.nxnx_f32_kv_cache.nxnx_f32_lazy_weight.nxnx_f32_llama_block.nxnx_f32_llama_block_v4.nxnx_f32_llama_stack_v4.nxnx_f32_llama_layer_lazy_load.nxnx_f32_llm.nxnx_f32_llm_v4.nxnx_f32_llm_read_dims.nxnx_f32_bpe_load.nxnx_f32_llm_special_tokens.nxnx_f32_sampler.nxnx_prng.nxnx_reasoning.nxnx_kvcache.nxnx_f32_attn_paged.nxnx_f32_llama_v4p.nxnx_f32_llama_v4b.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
| 41 | const BG_N: nx_int = 3 |
| 42 | const BG_MAXNEW: nx_int = 8 |
functions
| 44 | func bg_w(s: *u8) -> i64 { var n: i64 = 0; while s[n] != (0 as u8) { n = n + 1 } sys_write(1, s, n); return 0 } |
| 45 | func bg_wn(v: i64) -> i64 |
| 58 | func bg_beq(a: *u8, b: *u8, n: i64) -> i64 called by 1: main |
| 63 | func bg_stop(rc: *NxReasonCfg, t: nx_int) -> nx_int called by 1: main |
| 69 | func bg_emit(rc: *NxReasonCfg, t: nx_int, buf: *u8, len: *i64, one: *i64, db: *u8) -> i64 |
| 82 | func main() -> i64 |