Bilinear — PyTorch 2.7 documentation (original) (raw)

class torch.nn.Bilinear(in1_features, in2_features, out_features, bias=True, device=None, dtype=None)[source][source]¶

Applies a bilinear transformation to the incoming data: y=x1TAx2+by = x_1^T A x_2 + b.

Parameters

• in1_features (int) – size of each first input sample
• in2_features (int) – size of each second input sample
• out_features (int) – size of each output sample
• bias (bool) – If set to False, the layer will not learn an additive bias. Default: True

Shape:

• Input1: (∗,Hin1)(*, H_\text{in1}) where Hin1=in1_featuresH_\text{in1}=\text{in1\_features} and∗* means any number of additional dimensions including none. All but the last dimension of the inputs should be the same.
• Input2: (∗,Hin2)(*, H_\text{in2}) where Hin2=in2_featuresH_\text{in2}=\text{in2\_features}.
• Output: (∗,Hout)(*, H_\text{out}) where Hout=out_featuresH_\text{out}=\text{out\_features}and all but the last dimension are the same shape as the input.

Variables

• weight (torch.Tensor) – the learnable weights of the module of shape(out_features,in1_features,in2_features)(\text{out\_features}, \text{in1\_features}, \text{in2\_features}). The values are initialized from U(−k,k)\mathcal{U}(-\sqrt{k}, \sqrt{k}), wherek=1in1_featuresk = \frac{1}{\text{in1\_features}}
• bias – the learnable bias of the module of shape (out_features)(\text{out\_features}). If bias is True, the values are initialized fromU(−k,k)\mathcal{U}(-\sqrt{k}, \sqrt{k}), wherek=1in1_featuresk = \frac{1}{\text{in1\_features}}

Examples:

m = nn.Bilinear(20, 30, 40) input1 = torch.randn(128, 20) input2 = torch.randn(128, 30) output = m(input1, input2) print(output.size()) torch.Size([128, 40])