# init(modelDimension:keyDimension:valueDimension:headCount:dropout:hasBiases:hasAttentionBiases:addsZeroAttention:) Creates a multi-head attention descriptor with the dimensions, number of attention heads, dropout rate, and bias and padding options you specify. ``` convenience init?(modelDimension: Int, keyDimension: Int, valueDimension: Int, headCount: Int, dropout: Float, hasBiases: Bool, hasAttentionBiases: Bool, addsZeroAttention: Bool) ``` ## Parameters `modelDimension` The total dimension of model space. `keyDimension` The total dimension of key space; the default value is equal to `modelDimension`. `valueDimension` The total dimension of value space; the default value is equal to `modelDimension`. `headCount` The number of parallel attention heads. `dropout` The dropout rate you apply to the output projection weights; the default value is `0.0`. `hasBiases` A Boolean that specifies whether you add a bias to query, key, value, and output projections; the default value is `true`. `hasAttentionBiases` A Boolean that specifies whether you add a row of zeros to projected key and value; the default value is `false`. `addsZeroAttention` A Boolean that specifies whether you add a row of zeros to projected key and value; the default value is `false`. ## Discussion --- Copyright © 2026 Apple Inc. All rights reserved. | [Terms of Use](https://www.apple.com/legal/internet-services/terms/site.html) | [Privacy Policy](https://www.apple.com/privacy/privacy-policy)