Table 3.
Hyperparameter | Probability distribution |
---|---|
Learning rate | LogUniform(10−5, 10−2) |
MME tradeoff λ | Uniform(0, 1) |
SLA mix ratio α | Uniform(0, 1) |
SLA temperature τ | Uniform(0, 1) |
SLA warmup W | UniformChoice({100, 500, 1000, 2000, 5000}) |
SLA update interval I | UniformChoice({5, 10, 100, 500, 1000, 2000, 5000}) |