|
// Licensed to the .NET Foundation under one or more agreements.
// The .NET Foundation licenses this file to you under the MIT license.
// See the LICENSE file in the project root for more information.
using Microsoft.ML.TorchSharp.NasBert.Modules.Layers;
using System;
using Microsoft.ML.TorchSharp.Roberta.Modules;
using TorchSharp;
namespace Microsoft.ML.TorchSharp.Roberta.Models
{
internal class Layer : torch.nn.Module<torch.Tensor, torch.Tensor, torch.Tensor>
{
#pragma warning disable MSML_GeneralName // This name should be PascalCased
public readonly Attention attention;
public readonly Intermediate intermediate;
public readonly Output output;
private bool _disposedValue;
public Layer(int numAttentionHeads, long hiddenSize, long ffnHiddenSize, double layerNormEps,
double dropoutRate, double attentionDropoutRate, double outputDropoutRate)
: base(nameof(Layer))
{
attention = new Attention(numAttentionHeads, hiddenSize, layerNormEps, attentionDropoutRate, outputDropoutRate);
intermediate = new Intermediate(hiddenSize, ffnHiddenSize);
output = new Output(ffnHiddenSize, hiddenSize, dropoutRate);
RegisterComponents();
}
public override torch.Tensor forward(torch.Tensor input, torch.Tensor attentionMask)
{
using var disposeScope = torch.NewDisposeScope();
var attentionOutput = attention.forward(input, attentionMask);
var intermediateOutput = intermediate.forward(attentionOutput);
var layerOutput = output.forward(intermediateOutput, attentionOutput);
return layerOutput.MoveToOuterDisposeScope();
}
protected override void Dispose(bool disposing)
{
if (!_disposedValue)
{
if (disposing)
{
attention.Dispose();
intermediate.Dispose();
output.Dispose();
_disposedValue = true;
}
}
base.Dispose(disposing);
}
}
}
|