AdamParamsConfig ¶

class AdamParamsConfig(TypedDict, closed=True):
    learning_rate: "float"
    """<p>The learning rate for the optimizer. Must be greater than 0.</p>"""
    beta1: "float"
    """<p>The exponential decay rate for the first moment estimates.</p>"""
    beta2: "float"
    """<p>The exponential decay rate for the second moment estimates.</p>"""
    eps: "float"
    """<p>A small constant added to the denominator for numerical stability.</p>"""
    weight_decay: "float"
    """<p>The weight decay (L2 regularization) coefficient.</p>"""
    grad_clip_norm: "float"
    """<p>The maximum gradient norm for gradient clipping. Gradients are scaled down if their norm exceeds this value.</p>"""
    lora_plus_lr_ratio: NotRequired["float"]
    """<p>The learning rate multiplier for LoRA+ B matrices relative to A matrices. When set, B matrices use <code>learningRate * loraPlusLrRatio</code> while A matrices use the base learning rate.</p>"""