nx_nofloat_qwen_kvgen_gate.nx
buildroot/runtime/nx_nofloat_qwen_kvgen_gate.nx
about
nx_nofloat_qwen_kvgen_gate.nx -- KV-CACHED greedy generation through the sovereign no-float Qwen2.5-0.5B.
Prefill the 5-token prompt once (capturing every layer's K/V rows), then each new token is ONE single-token
decode_step_kv (flat cost in sequence length) instead of re-forwarding the whole growing sequence.
LOSSLESSNESS TEETH: the cached path must reproduce the uncached nx_nofloat_qwen_gen_gate tokens EXACTLY
(' Paris' '.' ' It' ' is' = 12095 13 1084 374) -- causal masking makes KV caching bit-exact, so any drift = bug.
No hw writes (Rule 26). expect_exit: 0 license_tier: ORIGINAL
dependencies 10 imports · 0 importers
imports: nx_syscalls.nxnx_tier.nxnx_le.nxnx_tensor.nxnx_gguf.nxnx_gguf_load.nxnx_gguf_meta.nxnx_nofloat_llm.nxnx_nofloat_tok.nxnx_gate_verdict.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
| 18 | func kg_puts(s: *u8) -> i64 { var n: i64=0; while s[n]!=(0 as u8){n=n+1} sys_write(1,s,n); return 0 } |
| 19 | func kg_num(v: i64) -> i64 { let b: *u8=sys_mmap(28); var m: i64=v; if m<0{m=0-m;sys_write(1,"-" as *u8,1)} let t: *u8=sys_mmap(28); var k: i64=0; if m==0{t[0]=48 as u8;k=1} while m>0{t[k]=(48+(m%10)) as u8;m=m/10;k=k+1} var i: i64=0; while i<k{b[i]=t[k-1-i];i=i+1} sys_write(1,b,k); return 0 } |
| 20 | func kg_slen(s: *u8) -> i64 { var n: i64=0; while s[n]!=(0 as u8){n=n+1} return n } called by 1: main |
| 23 | func kg_head_argmax(buf: *u8, oh_base: i64, oh_ty: i64, normed: *i64, row: *i64, tmp: *i64, vocab: i64, ne: i64, lgout: *i64) -> i64 calls 1: dequant_row |
| 38 | func main() -> i64 |