code wiki / (root) / nx_nofloat_blockfloat_layer_gate.nx

nx_nofloat_blockfloat_layer_gate.nx

buildroot/runtime/nx_nofloat_blockfloat_layer_gate.nx

7526 B147 linesdepth 3pulls 3 transitivereach 0 importersview sourcekind gate/prooftopic nofloat
docsdependenciesstructsconstsfunctions

about

nx_nofloat_blockfloat_layer_gate.nx -- BLOCK-FLOAT weights through a real LINEAR LAYER (W.x + bias -> ReLU), the rung after the bare GEMM: a genuine weight MATRIX (OUT x IN, mixed-magnitude rows) quantized block-float per-row-per-block, run through bias + activation. Proves the dynamic-range win + determinism survive a real layer (the core of FFN + attention projections). 1 block-float layer (computed on mantissas) == dequant-then-layer, BYTE-EXACT 2 EXCEED (MEASURED): block-float layer error vs full-precision < per-tensor INT8 layer error 3 DETERMINISTIC: the block-float layer run twice == bit-identical 4 per-tensor scaling ZEROED a row's small block; block-float kept it (the super-weight fix, in a layer) expect_exit: 0 license_tier: ORIGINAL

dependencies 2 imports · 0 importers

nx_syscalls.nx nx_gate_verdict.nx nx_nofloat_blockfloat_layer_ga

imports: nx_syscalls.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

main bf_puts sys_write sys_mmap bf_scale bf_bitlen fp_layer bf_relu bf_layer bf_relu ↻ bf_absdiff bf_putn sys_mmap ↻ sys_write ↻ bf_chk bf_puts ↻ gv_ctr sys_mmap ↻ gv_verdict gv_puts sys_write ↻ gv_num sys_mmap ↻ sys_write ↻ sys_munmap gv_journal sys_openat_append sys_mmap ↻ gv_catn sys_mmap ↻ sys_munmap ↻ sys_now_realtime_sec sys_mmap ↻ sys_clock_gettime_real gv_cat sys_write ↻ sys_close sys_munmap ↻ sys_exit

structs

none

consts

none

functions

13func bf_puts(s: *u8) -> i64 { var n: i64 = 0; while s[n] != (0 as u8) { n = n + 1 } sys_write(1, s, n); return 0 }
called by 2: bf_chkmain calls 1: sys_write
14func bf_putn(v: i64) -> i64
called by 1: main calls 2: sys_mmapsys_write
24func bf_chk(name: *u8, ok: i64) -> i64
called by 1: main calls 1: bf_puts
28func bf_bitlen(x: i64) -> i64 { var b: i64 = 0; var m: i64 = x; while m > 0 { m = m >> 1; b = b + 1 } return b }
called by 1: bf_scale
29func bf_absdiff(a: i64, b: i64) -> i64 { if a >= b { return a - b } return b - a }
called by 1: main
30func bf_relu(x: i64) -> i64 { if x < 0 { return 0 } return x }
called by 2: fp_layerbf_layer
31func bf_scale(W: *i64, off: i64, B: i64, M: i64) -> i64
called by 1: main calls 1: bf_bitlen
40func fp_layer(W: *i64, x: *i64, bias: *i64, OUT: i64, IN: i64, out: *i64) -> i64
called by 1: main calls 1: bf_relu
51func bf_layer(q_W: *i64, e_bf: *i64, x: *i64, bias: *i64, OUT: i64, IN: i64, B: i64, out: *i64) -> i64
called by 1: main calls 1: bf_relu
69func main() -> i64
calls 11: bf_putssys_mmapbf_scalefp_layerbf_layerbf_absdiff+5