File size: 1,452 Bytes
c289d87 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 | from __future__ import annotations
from dataclasses import dataclass
from typing import Sequence
@dataclass
class WeightScheduleConfig:
sample_knots: tuple[int, int, int, int] = (20, 50, 100, 200)
weight_knots: tuple[float, float, float, float] = (0.1, 0.3, 0.5, 0.8)
max_weight: float = 0.9
min_weight: float = 0.05
instability_threshold: float = 2.0
instability_decay: float = 0.25
def _interp(x: float, xs: Sequence[float], ys: Sequence[float]) -> float:
if x <= xs[0]:
return float(ys[0])
if x >= xs[-1]:
return float(ys[-1])
for i in range(1, len(xs)):
if x <= xs[i]:
x0, x1 = xs[i - 1], xs[i]
y0, y1 = ys[i - 1], ys[i]
t = (x - x0) / max(1e-9, (x1 - x0))
return float(y0 + t * (y1 - y0))
return float(ys[-1])
def compute_model_weight(
n_samples: int,
instability_ratio: float = 1.0,
config: WeightScheduleConfig | None = None,
) -> float:
cfg = config or WeightScheduleConfig()
base = _interp(float(n_samples), cfg.sample_knots, cfg.weight_knots)
if n_samples > cfg.sample_knots[-1]:
extra = min(cfg.max_weight - base, 0.05 * (n_samples - cfg.sample_knots[-1]) / 100.0)
base += max(0.0, extra)
if instability_ratio > cfg.instability_threshold:
base *= max(0.1, 1.0 - cfg.instability_decay)
base = max(cfg.min_weight, min(cfg.max_weight, base))
return float(base)
|