Skip to content

heuristic_reward

heuristic_reward

Heuristic reward function — weighted score from latency, cost, efficiency.

Classes

HeuristicRewardFunction 

HeuristicRewardFunction(
    *,
    weight_latency: float = 0.4,
    weight_cost: float = 0.3,
    weight_efficiency: float = 0.3,
    max_latency: float = 30.0,
    max_cost: float = 0.01,
)

Bases: RewardFunction

Computes a scalar reward based on latency, cost, and token efficiency.

Each component is normalised to [0, 1] and combined via a weighted sum.

Source code in src/diapason/learning/routing/heuristic_reward.py 
16
17
18
19
20
21
22
23
24
25
26
27
28
29
def __init__(
    self,
    *,
    weight_latency: float = 0.4,
    weight_cost: float = 0.3,
    weight_efficiency: float = 0.3,
    max_latency: float = 30.0,
    max_cost: float = 0.01,
) -> None:
    self.weight_latency = weight_latency
    self.weight_cost = weight_cost
    self.weight_efficiency = weight_efficiency
    self.max_latency = max_latency
    self.max_cost = max_cost