參考：about-pytorch

定義ScaledDotProductAttention結構體：

use candle_core::{Result, Device, Tensor};
use candle_nn::{Linear, Module, linear_no_bias, VarMap, VarBuilder, ops};struct ScaledDotProductAttention {wq: Linear,wk: Linear,wv: Linear,d_model: Tensor,device: Device,
}

為ScaledDotProductAttention結構體實現new方法：

impl ScaledDotProductAttention {fn new(vb: VarBuilder, embedding_dim: usize, out_dim: usize, device: Device) -> Result<Self> {Ok(Self { wq: linear_no_bias(embedding_dim, out_dim, vb.pp("wq"))?, wk: linear_no_bias(embedding_dim, out_dim, vb.pp("wk"))?, wv: linear_no_bias(embedding_dim, out_dim, vb.pp("wv"))?,d_model: Tensor::new(embedding_dim as f32, &device)?,device,})}
}

為結構體實現Module的forward trait:

impl Module for ScaledDotProductAttention {fn forward(&self, xs: &Tensor) -> Result<Tensor> {let q = self.wq.forward(xs)?;let k = self.wk.forward(xs)?;let v = self.wv.forward(xs)?;let attn_score = q.matmul(&k.t()?)?;let attn_score = attn_score.broadcast_div(&self.d_model.sqrt()?)?;let dim = attn_score.rank() - 1;let attn_weights = ops::softmax(&attn_score, dim)?;let attn_output = attn_weights.matmul(&v)?;Ok(attn_output)}
}

融合qkv實現：

定義ScaledDotProductAttentionFusedQKV結構體：

struct ScaledDotProductAttentionFusedQKV {w_qkv: Linear,d_model: Tensor,device: Device,
}

為結構體實現new方法：

impl ScaledDotProductAttentionFusedQKV {fn new(vb: VarBuilder, embedding_dim: usize, out_dim: usize, device: Device) -> Result<Self> {Ok(Self { w_qkv: linear_no_bias(embedding_dim, 3*out_dim, vb.pp("w_qkv"))?,d_model: Tensor::new(embedding_dim as f32, &device)?,device,})}
}

為結構體實現forward trait：

impl Module for ScaledDotProductAttentionFusedQKV {fn forward(&self, xs: &Tensor) -> Result<Tensor> {let qkv = self.w_qkv.forward(xs)?;let (batch_size, seq_len, _) = qkv.dims3()?;let qkv = qkv.reshape((batch_size, seq_len, 3, ()))?;let q = qkv.get_on_dim(2, 0)?;let q = q.reshape((batch_size, seq_len, ()))?;let k = qkv.get_on_dim(2, 1)?;let k = k.reshape((batch_size, seq_len, ()))?;let v = qkv.get_on_dim(2, 2)?;let v = v.reshape((batch_size, seq_len, ()))?;let attn_score = q.matmul(&k.t()?)?;let attn_score = attn_score.broadcast_div(&self.d_model.sqrt()?)?;let dim = attn_score.rank() - 1;let attn_weights = ops::softmax(&attn_score, dim)?;let attn_output = attn_weights.matmul(&v)?;Ok(attn_output)}
}

測試：

fn main() -> Result<()> {let device = Device::cuda_if_available(0)?;let varmap = VarMap::new();let vb = VarBuilder::from_varmap(&varmap, candle_core::DType::F32, &device);let input = Tensor::from_vec(vec![0.43f32, 0.15, 0.89, 0.55, 0.87, 0.66,0.57, 0.85, 0.64,0.22, 0.58, 0.33,0.77, 0.25, 0.10,0.05, 0.80, 0.55, 0.43, 0.15, 0.89, 0.55, 0.87, 0.66,0.57, 0.85, 0.64,0.22, 0.58, 0.33,0.77, 0.25, 0.10,0.05, 0.80, 0.55], (2, 6, 3), &device)?;// let model = ScaledDotProductAttention::new(vb.clone(), 3, 2, device.clone())?;let model = ScaledDotProductAttentionFusedQKV::new(vb.clone(), 3, 2, device.clone())?;let output = model.forward(&input)?;println!("output: {:?}\n", output);println!("output: {:?}\n", output.to_vec3::<f32>()?);Ok(())
}