8000 update torchsharp to 0.105.0 · dotnet/machinelearning@6145645 · GitHub
[go: up one dir, main page]
Skip to content

Commit 6145645

Browse files
LittleLittleCloudLittleLittleCloud
committed
update torchsharp to 0.105.0
1 parent e19aa2a commit 6145645

File tree

4 files changed

+12
-12
lines changed

4 files changed

+12
-12
lines changed

‎eng/Versions.props

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -73,8 +73,8 @@
7373
<TorchSharpPyBridgeVersion>1.4.1</TorchSharpPyBridgeVersion>
7474
<AutoGenVersion>0.1.0</AutoGenVersion>
7575
<SemanticKernelVersion>1.15.0</SemanticKernelVersion>
76-
<TorchSharpVersion>0.102.7</TorchSharpVersion>
77-
<LibTorchVersion>2.2.1.1</LibTorchVersion>
76+
<TorchSharpVersion>0.105.0</TorchSharpVersion>
77+
<LibTorchVersion>2.5.1</LibTorchVersion>
7878
<!-- Build/infrastructure Dependencies -->
7979
<CodecovVersion>1.12.4</CodecovVersion>
8080
<CoverletCollectorVersion>6.0.2</CoverletCollectorVersion>

‎src/Microsoft.ML.GenAI.Core/Module/QuantizedLinear.cs

Lines changed: 8 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -87,9 +87,9 @@ public override Tensor forward(Tensor input)
8787
{
8888
// 8bit quantization
8989
using var dispose = torch.NewDisposeScope();
90-
var weight = this.get_buffer("8bit_weight").to(ScalarType.Float32);
91-
var zeroPoint = this.get_buffer("zeroPoint").to(ScalarType.Float32);
92-
var scale = this.get_buffer("scale").to(ScalarType.Float32);
90+
var weight = this.get_buffer("8bit_weight")!.to(ScalarType.Float32);
91+
var zeroPoint = this.get_buffer("zeroPoint")!.to(ScalarType.Float32);
92+
var scale = this.get_buffer("scale")!.to(ScalarType.Float32);
9393
var restoreWeight = (weight - zeroPoint.view(-1, 1)) / scale.view(-1, 1);
9494
// use float32
9595
var result = torch.matmul(input.to(ScalarType.Float32), restoreWeight.T);
@@ -106,17 +106,17 @@ public override Tensor forward(Tensor input)
106106
{
107107
using var dispose = torch.NewDisposeScope();
108108
var weight = this.get_buffer("4bit_weight");
109-
var weightLower = weight % 16;
110-
var weightUpper = weight / 16;
109+
var weightLower = weight! % 16;
110+
var weightUpper = weight! / 16;
111111
weight = torch.cat([weightUpper, weightLower], 0).to(ScalarType.Float32);
112112
weight = weight.view(this._outFeatures, this._inFeatures);
113113
weight -= 8;
114114
var zeroPoint = this.get_buffer("zeroPoint");
115-
var zeroPointLower = zeroPoint % 16;
116-
var zeroPointUpper = zeroPoint / 16;
115+
var zeroPointLower = zeroPoint! % 16;
116+
var zeroPointUpper = zeroPoint! / 16;
117117
zeroPoint = torch.cat([zeroPointUpper, zeroPointLower], 0).to(ScalarType.Float32);
118118
zeroPoint -= 8;
119-
var scale = this.get_buffer("scale").to(ScalarType.Float32);
119+
var scale = this.get_buffer("scale")!.to(ScalarType.Float32);
120120
var restoreWeight = (weight - zeroPoint.view(-1, 1)) / scale.view(-1, 1);
121121
// use float32
122122
var result = torch.matmul(input.to(ScalarType.Float32), restoreWeight.T);

‎src/Microsoft.ML.GenAI.Core/Module/RotaryEmbedding.cs

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -109,7 +109,7 @@ public override RotaryEmbeddingOutput forward(RotaryEmbeddingInput input)
109109
var seqLen = input.SeqLen;
110110
// TODO
111111
// can be calculated once and cached
112-
var invFreq = this.get_buffer("inv_freq").to(x.device);
112+
var invFreq = this.get_buffer("inv_freq")!.to(x.device);
113113
var invFreqExpanded = invFreq.unsqueeze(0).unsqueeze(-1);
114114
invFreqExpanded = invFreqExpanded.expand(new long[] { positionIds.shape[0], -1, 1 });
115115
var positionIdsExpanded = positionIds.unsqueeze(1).to(torch.float32);

‎src/Microsoft.ML.TorchSharp/AutoFormerV2/ConvModule.cs

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -35,7 +35,7 @@ public class ConvModule : Module<Tensor, Tensor>
3535
public ConvModule(int inChannel, int outChannel, int kernelSize, int stride = 1, int padding = 0, int dilation = 1, bool bias = true, bool useRelu = true)
3636
: base(nameof(ConvModule))
3737
{
38-
this.conv = nn.Conv2d(in_channels: inChannel, out_channels: outChannel, kernelSize: kernelSize, stride: stride, padding: padding, dilation: dilation, bias: bias);
38+
this.conv = nn.Conv2d(in_channels: inChannel, out_channels: outChannel, kernel_size: kernelSize, stride: stride, padding: padding, dilation: dilation, bias: bias);
3939
this.useRelu = useRelu;
4040
if (this.useRelu)
4141
{

0 commit comments

Comments
 (0)
0