我的反向传播算法实现出了什么问题？

# 前向传播
def forward_propagation(X, weights, biases):
    activations = [X]
    zs = []
    for w, b in zip(weights, biases):
        z = np.dot(activations[-1], w) + b
        zs.append(z)
        activation = sigmoid(z)  # 激活函数
        activations.append(activation)
    return activations, zs

# 反向传播
def backward_propagation(X, y, activations, zs, weights, biases, learning_rate):
    m = X.shape[0]  # 样本数量
    delta = activations[-1] - y
    deltas = [delta]
    
    # 计算隐藏层的delta
    for l in range(2, len(weights)):
        z = zs[-l]
        sp = sigmoid_prime(z)
        delta = np.dot(deltas[-1], weights[-l+1].T) * sp
        deltas.append(delta)
    deltas.reverse()
    
    # 更新权重和偏置
    for l in range(len(weights)):
        weights[l] -= learning_rate * np.dot(activations[l].T, deltas[l]) / m
        biases[l] -= learning_rate * np.sum(deltas[l], axis=0, keepdims=True) / m

# 主训练循环
for epoch in range(num_epochs):
    activations, zs = forward_propagation(X_train, weights, biases)
    backward_propagation(X_train, y_train, activations, zs, weights, biases, learning_rate)