关于「跨模型路由:复杂度分档、标签误差与级联升级」,《周期配额的路由与倒计时》给出的定量结论是什么?

结论:第四章讲配额变化、复杂度判别出错与级联升级,级联把最优产出从 893.00 点提到 944.65 点。

数据口径:示意参数下,线性规划相对先到先用多出 0.1668 到 0.2217,相对按复杂度分档的固定规则只多出 0.41 % 到 5.05 %,大头来自分档本身。 书中参数为示意,不含任何厂商的价格或配额;全部计算由书中程序生成。

English version

对照数据表

下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 893.00, 944.65

classoptionpoints/taskH unitsF units
routineH0.97004.0000
routineF0.95000.0001
routineC0.98590.5401
standardH1.90006.0000
standardF1.60000.0002
standardC1.89601.4402
complexH4.500010.0000
complexF2.25000.0003
complexC4.20754.8503
criticalH8.500016.0000
criticalF1.50000.0004
criticalC7.257511.1204
columns HFoptimum893.00y_H 0.2250
F0:300.00 F1:150.00 F2:32.00 H2:28.00 H3:20.00
columns HFC optimum944.65y_H 0.2546
C2:60.00 C3:2.25 F0:300.00 F1:150.00 H3:17.75
查看完整验算程序
# Chapter 4. p43: a cascade as a third column of the routing programme.
# Model: besides "H" and "F", a task may take the cascade "C": run F,
# let a checker judge the answer, and escalate to H only when flagged.
# The checker flags a wrong F answer with probability d = 0.8 and a
# right one with probability f = 0.1 (illustrative). Per task of
# class k:  P(flag) = (1 - qF) d + qF f,
#           success = qF (1 - f) + P(flag) qH,
#           F units = cost_F,  H units = P(flag) cost_H.
# The LP chooses H, F or C per class under quotas H 600, F 900, on the
# demand (300, 150, 60, 20) of p21. Units illustrative.
from fractions import Fraction as Fr

def simplex(c, A, b, rule="bland", cap=500):
    m, n = len(A), len(c)
    T = [[Fr(v) for v in A[i]] + [Fr(int(i == j)) for j in range(m)]
         + [Fr(b[i])] for i in range(m)]
    z = [-Fr(v) for v in c] + [Fr(0)] * (m + 1)
    basis = [n + i for i in range(m)]
    steps = 0
    while steps < cap:
        cols = [j for j in range(n + m) if z[j] < 0]
        if not cols:
            break
        if rule == "bland":
            e = cols[0]
        else:
            e = min(cols, key=lambda j: (z[j], j))
        rows = [i for i in range(m) if T[i][e] > 0]
        if not rows:
            return None
        r = min(rows, key=lambda i: (T[i][-1] / T[i][e], basis[i]))
        p = T[r][e]
        T[r] = [v / p for v in T[r]]
        for i in range(m):
            if i != r and T[i][e] != 0:
                f = T[i][e]
                T[i] = [a - f * w for a, w in zip(T[i], T[r])]
        f = z[e]
        z = [a - f * w for a, w in zip(z, T[r])]
        basis[r] = e
        steps += 1
    x = [Fr(0)] * (n + m)
    for i, j in enumerate(basis):
        x[j] = T[i][-1]
    return x[:n], z[-1], z[n:n + m], steps


VALUE = (1, 2, 5, 10)
COST = {"H": (4, 6, 10, 16), "F": (1, 2, 3, 4)}
SUCC = {"H": ("0.97", "0.95", "0.90", "0.85"),
        "F": ("0.95", "0.80", "0.45", "0.15")}
DEM = (300, 150, 60, 20)
NAMES = ("routine", "standard", "complex", "critical")
D_HIT, F_ALARM = Fr("0.8"), Fr("0.1")


def options(k):
    qh, qf = Fr(SUCC["H"][k]), Fr(SUCC["F"][k])
    flag = (1 - qf) * D_HIT + qf * F_ALARM
    succ = qf * (1 - F_ALARM) + flag * qh
    return {"H": (VALUE[k] * qh, COST["H"][k], 0),
            "F": (VALUE[k] * qf, 0, COST["F"][k]),
            "C": (VALUE[k] * succ, flag * COST["H"][k], COST["F"][k])}


def solve(cols, qh=600, qf=900):
    var = [(t, k) for t in cols for k in range(4)]
    c = [options(k)[t][0] for t, k in var]
    A = [[options(k)[t][1] for t, k in var],
         [options(k)[t][2] for t, k in var]]
    b = [qh, qf]
    for j in range(4):
        A.append([int(k == j) for t, k in var])
        b.append(DEM[j])
    x, val, y, _ = simplex(c, A, b)
    return val, dict(zip(var, x)), y[0]


print("class     option  points/task  H units  F units")
for k in range(4):
    for t in "HFC":
        v, h, f = options(k)[t]
        print("%-8s  %6s  %11.4f  %7.3f  %7d" % (NAMES[k], t, float(v),
                                                 float(h), f))
for cols in ("HF", "HFC"):
    val, x, yh = solve(cols)
    alloc = " ".join("%s:%.2f" % (t + str(k), float(x[(t, k)]))
                     for t, k in sorted(x) if x[(t, k)] > 0)
    print("columns %-3s optimum %7.2f  y_H %6.4f" % (cols, float(val),
                                                      float(yh)))
    print("   " + alloc)

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《周期配额削峰填谷与跨模型路由、重置倒计时线性规划与重推理轻量模型算力分配母典》第 4 章「跨模型路由:复杂度分档、标签误差与级联升级」(版本 1.0.0). 5D Library. DOI 待发布

DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-270 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026270,
  author    = {Fan, Yafu},
  title     = {Peak-Shaving Periodic Quotas and Cross-Model Routing: Linear Programming against Reset Countdowns to Split Work between Heavy-Reasoning and Fast Models},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论