t27.aiEnglish

GF-T8: порядок в тритах

Вы узнаете

Как GF-T сохраняет деление по phi, но хранит порядок в сбалансированных тритах: 1 + 3 трита + 4 бита.

GF-T — это GoldenFloat с троичным порядком. В GF-T8 1 знак, 3 трита порядка и 4 бита мантиссы. Три трита дают 27 значений порядка, смещение 13 делает их сбалансированными вокруг нуля, а смещение 26 оставлено как особое неконечное значение. Спека проверяет деление как E / M в тысячных: 3000 / 4 = 750.

Попробуйте

Запустите два теста gft8.t27. Затем найдите exp_values и is_finite и скажите, какое смещение не конечно.

Открыть интерактивный урок →

GoldenFloat 24: GF-T8, an exponent in trits
GoldenFloat 24: GF-T8, an exponent in trits ↗

GF-T8 as the spec lays it out, read from gft8.t27. Lesson 24 of the GoldenFloat course.

specs/numeric/gft8.t27

// SPDX-License-Identifier: Apache-2.0
// gft8.t27 -- GF-T8: the golden-ratio ternary ladder, 8-bit class.
//
// TWO AXES, FOUR FAMILIES. GF and GF-T are derived from the golden ratio; BNF and
// TNF are derived from the theorems, as the optimisation result for ternary
// networks. They are not renamings of each other and they answer different
// questions.
//
// GF-T applies GF's rule to POSITIONS, and a trit is a position -- the golden
// section divides the payload the way it divides a segment:
//
//     E_t = round((N-1)/phi^2) = 3,   M = N - 1 - E_t = 4
//
// Every rung lands exactly: 1 + 3 + 4 = 8, no position unspent. The ratio
// E_t/M = 0.7500 against 1/phi = 0.6180, a phi-distance of 0.1320
// which falls toward zero up the ladder, by construction, exactly as in GF.
//
// What this buys and what it costs, measured against TNF8 on the reference
// oracle: GF-T's exponent spans 13 binades either side where TNF sizes its own
// for the range a workload actually visits. GF-T pays for that in mantissa. At 64
// bits GF-T takes 1.3e8 times the range for 1.2e5 times the error -- neither
// dominates, and the corollary on the pair (M_eff, binades) forbids ranking them
// without naming a workload.
//
// Supersedes the ad-hoc parameters this rung carried before 2026-08-09, where the
// exponent was sized at roughly log2(N) trits with no documented rule and left
// positions unspent.
//
//   layout: [ sign(1) | E = 3 balanced-ternary trits | M = 4 binary bits ]
//   value = (-1)^sign * (1 + M/2^4) * 2^e,   e in [-13,+13]

module triformat_gft8 {
    use base::types;

    const SIGN_BITS: u32 = 1;
    const EXP_TRITS: u32 = 3;     // round((N-1)/phi^2)
    const MANT_BITS: u32 = 4;      // the remaining positions, all of them
    const EXP_OFFSET: u32 = 13;
    const OFFSET_MAX: u32 = 26;

    fn is_finite(offset: u32) -> bool { return offset != OFFSET_MAX; }
    fn exp_values() -> u32 { return 27; }

    // ---- Tests / invariants ----

    // The golden section is the rule; this asserts it rather than remembering it.
    test golden_section {
        assert(SIGN_BITS + EXP_TRITS + MANT_BITS == 8, "1 + E_t + M = N, every position spent");
        assert(EXP_TRITS * 1000 / MANT_BITS == 750, "E_t/M holds the golden section");
    }

    test balanced_offsets {
        assert(EXP_OFFSET * 2 == OFFSET_MAX, "balanced: offset_max = 2 * exp_offset");
    }
}

Открыть спеку урока в плеере ↗

Все уроки