torch

package module
v0.0.0-...-890d392 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Sep 8, 2025 License: AGPL-3.0 Imports: 12 Imported by: 0

README

BioTorch - Neuromorphic Compute(Bio-Neuron Inspired) DeepLearning Engine

build

Inference Compute Runtime Support

  • Built-in Inference Computing Engine (CPU Compute Version Available Under AGPL License / GPU Close Source)
  • ONNX Runtime (Version 7 - Export ONNX Models for Offline Production Computation, Available Under AGPL license / ONNX Runtime Built-in Realtime Compute Integration Close Source)
llama3 example with gpu acceleration built on top of this framework
llama3_1b_cpu_01llama3_1b_cpu-02
CNN From Scratch example AND re-design build-in Graphic Renderer that compatible with Tensor data struct:
CNN From Scratchgraphic_renderer
More Examples: README.md
Notice

Some parts of this project’s code may generate, optimize, or annotate by AI.

Terms of Use
Permission to Use

Citizens, companies, and other organizations from Full democracies and Flawed democracies countries (classified by The Economist Democracy Index) are permitted to use this repository under the GNU Affero General Public License (AGPL).

Prohibited Uses

The following use purposes are strictly prohibited:

  • Any activity to serve the ideology or social system of communism, socialism, or any other ideology in a manner that leads to harm, oppression, or violation of human rights.
  • Human rights violations, including surveillance, oppression, or discrimination.
  • Terrorism or supporting terrorist organizations.
  • Any other malicious or harmful purposes that could cause harm to individuals or society.
  • Military applications, including but not limited to warfare, weapons development, or any defense-related activities.
  • Illegal activities, such as hacking, fraud, or any action that violates applicable laws.

Users must comply with all local, national, and international laws when using this repository. To request permission for a use case not covered above, please contact me.

Documentation

Index

Constants

This section is empty.

Variables

This section is empty.

Functions

func LinearCompute 

func LinearCompute(Bias, Weights, OutputData []float32, inputLength int, inputFloat32 float32, startPos int)

func LoadFlatDataFromCSV 

func LoadFlatDataFromCSV(filePath string) ([]float32, error)

func LoadImageFromCSV 

func LoadImageFromCSV(filename string) *tensor.Tensor

func LoadMatrixFromCSV 

func LoadMatrixFromCSV(filename string) (*tensor.Tensor, error)

func MSE 

func MSE(predictions, targets *tensor.Tensor) float32

func Sigmoid 

func Sigmoid(x float32) float32

func SigmoidDerivative 

func SigmoidDerivative(x float32) float32

Types

type BasicTrainer 

type BasicTrainer struct {
	LossFunc func(predictions, targets *tensor.Tensor) float32
	Verbose  bool
}

func NewBasicTrainer 

func NewBasicTrainer(lossFunc func(predictions, targets *tensor.Tensor) float32) *BasicTrainer

func (*BasicTrainer) Train 

func (t *BasicTrainer) Train(model ModelInterface, inputs, targets *tensor.Tensor, epochs int, learningRate float32)

func (*BasicTrainer) Validate 

func (t *BasicTrainer) Validate(model ModelInterface, inputs, targets *tensor.Tensor) float32

type BatchNormLayer 

type BatchNormLayer struct {
	RunningMean *tensor.Tensor
	// contains filtered or unexported fields
}

func NewBatchNormLayer 

func NewBatchNormLayer(numFeatures int, eps, momentum float32) *BatchNormLayer

func (*BatchNormLayer) Forward 

func (bn *BatchNormLayer) Forward(x *tensor.Tensor) *tensor.Tensor

func (*BatchNormLayer) GetBias 

func (l *BatchNormLayer) GetBias() *tensor.Tensor

func (*BatchNormLayer) GetWeights 

func (l *BatchNormLayer) GetWeights() *tensor.Tensor

func (*BatchNormLayer) SetBias 

func (l *BatchNormLayer) SetBias(data []float32)

func (*BatchNormLayer) SetBiasAndShape 

func (l *BatchNormLayer) SetBiasAndShape(data []float32, shape []int)

func (*BatchNormLayer) SetWeights 

func (l *BatchNormLayer) SetWeights(data []float32)

func (*BatchNormLayer) SetWeightsAndShape 

func (l *BatchNormLayer) SetWeightsAndShape(data []float32, shape []int)

type ConvLayer 

type ConvLayer struct {
	InChannels  int
	OutChannels int
	KernelSize  int
	Stride      int
	Padding     int
	Weights     *tensor.Tensor
	Bias        *tensor.Tensor
	InputCache  *tensor.Tensor
	GradWeights *tensor.Tensor
	GradBias    *tensor.Tensor
}

func NewConvLayer 

func NewConvLayer(inCh, outCh, kSize, stride, pad int) *ConvLayer

func (*ConvLayer) Backward 

func (c *ConvLayer) Backward(gradOutput *tensor.Tensor) *tensor.Tensor

func (*ConvLayer) BackwardWithLR 

func (c *ConvLayer) BackwardWithLR(gradOutput *tensor.Tensor, learningRate float32) *tensor.Tensor

func (*ConvLayer) Forward 

func (c *ConvLayer) Forward(x *tensor.Tensor) *tensor.Tensor

func (*ConvLayer) GetBias 

func (l *ConvLayer) GetBias() *tensor.Tensor

func (*ConvLayer) GetWeights 

func (l *ConvLayer) GetWeights() *tensor.Tensor

func (*ConvLayer) SetBias 

func (l *ConvLayer) SetBias(data []float32)

func (*ConvLayer) SetBiasAndShape 

func (l *ConvLayer) SetBiasAndShape(data []float32, shape []int)

func (*ConvLayer) SetWeights 

func (l *ConvLayer) SetWeights(data []float32)

func (*ConvLayer) SetWeightsAndShape 

func (l *ConvLayer) SetWeightsAndShape(data []float32, shape []int)

func (*ConvLayer) UpdateParameters 

func (c *ConvLayer) UpdateParameters(learningRate float32)

func (*ConvLayer) ZeroGrad 

func (c *ConvLayer) ZeroGrad()

type Embedding 

type Embedding struct {
	Weights     *tensor.Tensor
	GradWeights *tensor.Tensor
	VocabSize   int
	EmbDim      int
	LastIndices []int
}

func NewEmbedding 

func NewEmbedding(vocabSize, embDim int) *Embedding

func (*Embedding) Backward 

func (e *Embedding) Backward(gradOutput *tensor.Tensor, learningRate float32) *tensor.Tensor

func (*Embedding) Forward 

func (e *Embedding) Forward(indices *tensor.Tensor) *tensor.Tensor

func (*Embedding) GetBias 

func (e *Embedding) GetBias() *tensor.Tensor

func (*Embedding) GetWeights 

func (e *Embedding) GetWeights() *tensor.Tensor

func (*Embedding) Parameters 

func (e *Embedding) Parameters() []*tensor.Tensor

func (*Embedding) SetBiasAndShape 

func (e *Embedding) SetBiasAndShape(data []float32, shape []int)

func (*Embedding) SetWeightsAndShape 

func (e *Embedding) SetWeightsAndShape(data []float32, shape []int)

func (*Embedding) ZeroGrad 

func (e *Embedding) ZeroGrad()

type FlattenLayer 

type FlattenLayer struct {
	// contains filtered or unexported fields
}

func NewFlattenLayer 

func NewFlattenLayer() *FlattenLayer

func (*FlattenLayer) Backward 

func (f *FlattenLayer) Backward(dout *tensor.Tensor) *tensor.Tensor

func (*FlattenLayer) Forward 

func (f *FlattenLayer) Forward(x *tensor.Tensor) *tensor.Tensor

func (*FlattenLayer) GetBias 

func (r *FlattenLayer) GetBias() *tensor.Tensor

func (*FlattenLayer) GetWeights 

func (r *FlattenLayer) GetWeights() *tensor.Tensor

type Layer 

type Layer interface {
	Forward(input *tensor.Tensor) *tensor.Tensor

	//DEL
	Backward(gradOutput *tensor.Tensor, learningRate float32) *tensor.Tensor
	ZeroGrad()
	Parameters() []*tensor.Tensor
}

type LayerForTesting 

type LayerForTesting interface {
	GetWeights() *tensor.Tensor
	GetBias() *tensor.Tensor
}

type LayerLoader 

type LayerLoader interface {
	SetWeightsAndShape(data []float32, shape []int)
	SetBiasAndShape(data []float32, shape []int)
}

type LinearLayer 

type LinearLayer struct {
	InputDim          int
	OutputDim         int
	Weights           *tensor.Tensor
	Bias              *tensor.Tensor
	WeightsTransposed bool
}

func NewLinearLayer 

func NewLinearLayer(inputDim, outputDim int) *LinearLayer

func (*LinearLayer) Backward 

func (l *LinearLayer) Backward(x *tensor.Tensor, lr float32) *tensor.Tensor

func (*LinearLayer) Forward 

func (l *LinearLayer) Forward(x *tensor.Tensor) *tensor.Tensor

func (*LinearLayer) ForwardMultiThread 

func (l *LinearLayer) ForwardMultiThread(x *tensor.Tensor) *tensor.Tensor

func (*LinearLayer) ForwardSignalThread 

func (l *LinearLayer) ForwardSignalThread(x *tensor.Tensor) *tensor.Tensor

func (*LinearLayer) ForwardSignalThreadCompute 

func (l *LinearLayer) ForwardSignalThreadCompute(x *tensor.Tensor) *tensor.Tensor

func (*LinearLayer) GetBias 

func (l *LinearLayer) GetBias() *tensor.Tensor

func (*LinearLayer) GetWeights 

func (l *LinearLayer) GetWeights() *tensor.Tensor

func (*LinearLayer) Parameters 

func (l *LinearLayer) Parameters() []*tensor.Tensor

func (*LinearLayer) SetBias 

func (l *LinearLayer) SetBias(data []float32)

func (*LinearLayer) SetBiasAndShape 

func (l *LinearLayer) SetBiasAndShape(data []float32, shape []int)

func (*LinearLayer) SetWeights 

func (l *LinearLayer) SetWeights(data []float32)

func (*LinearLayer) SetWeightsAndShape 

func (l *LinearLayer) SetWeightsAndShape(data []float32, shape []int)

func (*LinearLayer) SetupBackward 

func (l *LinearLayer) SetupBackward(x, out *tensor.Tensor)

func (*LinearLayer) ZeroGrad 

func (l *LinearLayer) ZeroGrad()

type MaxPool2DLayer 

type MaxPool2DLayer struct {
	PoolSize  int
	Stride    int
	Padding   int
	Input     *tensor.Tensor
	ArgMax    [][4]int
	OutputDim []int
}

func NewMaxPool2DLayer 

func NewMaxPool2DLayer(poolSize, stride, padding int) *MaxPool2DLayer

func (*MaxPool2DLayer) Backward 

func (m *MaxPool2DLayer) Backward(gradOutput *tensor.Tensor) *tensor.Tensor

func (*MaxPool2DLayer) Forward 

func (m *MaxPool2DLayer) Forward(x *tensor.Tensor) *tensor.Tensor

func (*MaxPool2DLayer) OutputShape 

func (m *MaxPool2DLayer) OutputShape() []int

type Model 

type Model struct {
	Layers          []Layer
	LayerIndex2Name map[int]string
	LayerName2Index map[string]int
	// contains filtered or unexported fields
}

func NewModel 

func NewModel() *Model

func (*Model) AddLayer 

func (m *Model) AddLayer(layer Layer)

func (*Model) Backward 

func (m *Model) Backward(target *tensor.Tensor, learningRate float32)

func (*Model) Forward 

func (m *Model) Forward(input *tensor.Tensor) *tensor.Tensor

func (*Model) PrintModel 

func (m *Model) PrintModel()

func (*Model) ZeroGrad 

func (m *Model) ZeroGrad()

type ModelInterface 

type ModelInterface interface {
	Forward(input *tensor.Tensor) *tensor.Tensor
	Backward(target *tensor.Tensor, learningRate float32)
	ZeroGrad()
}

type ReLULayer 

type ReLULayer struct {
	// contains filtered or unexported fields
}

func NewLeakyReLULayer 

func NewLeakyReLULayer(negativeSlope float32) *ReLULayer

func NewReLULayer 

func NewReLULayer() *ReLULayer

func (*ReLULayer) ActivationType 

func (r *ReLULayer) ActivationType() string

func (*ReLULayer) Backward 

func (r *ReLULayer) Backward(gradOutput *tensor.Tensor, learningRate float32) *tensor.Tensor

func (*ReLULayer) Forward 

func (r *ReLULayer) Forward(x *tensor.Tensor) *tensor.Tensor

func (*ReLULayer) GetBias 

func (r *ReLULayer) GetBias() *tensor.Tensor

func (*ReLULayer) GetWeights 

func (r *ReLULayer) GetWeights() *tensor.Tensor

func (*ReLULayer) NegativeSlope 

func (r *ReLULayer) NegativeSlope() float32

func (*ReLULayer) Parameters 

func (r *ReLULayer) Parameters() []*tensor.Tensor

func (*ReLULayer) SetInplace 

func (r *ReLULayer) SetInplace(inplace bool)

func (*ReLULayer) ZeroGrad 

func (r *ReLULayer) ZeroGrad()

type SigmoidLayer 

type SigmoidLayer struct {
}

func NewSigmoidLayer 

func NewSigmoidLayer() *SigmoidLayer

func (*SigmoidLayer) Backward 

func (s *SigmoidLayer) Backward(gradOutput *tensor.Tensor, learningRate float32) *tensor.Tensor

func (*SigmoidLayer) Forward 

func (s *SigmoidLayer) Forward(input *tensor.Tensor) *tensor.Tensor

func (*SigmoidLayer) Parameters 

func (s *SigmoidLayer) Parameters() []*tensor.Tensor

func (*SigmoidLayer) ZeroGrad 

func (s *SigmoidLayer) ZeroGrad()

type SoftmaxLayer 

type SoftmaxLayer struct {
	// contains filtered or unexported fields
}

func NewSoftmaxLayer 

func NewSoftmaxLayer(axis int) *SoftmaxLayer

func (*SoftmaxLayer) Backward 

func (s *SoftmaxLayer) Backward(dout *tensor.Tensor) *tensor.Tensor

func (*SoftmaxLayer) Forward 

func (s *SoftmaxLayer) Forward(x *tensor.Tensor) *tensor.Tensor

func (*SoftmaxLayer) GetAxis 

func (s *SoftmaxLayer) GetAxis() int

func (*SoftmaxLayer) SetAxis 

func (s *SoftmaxLayer) SetAxis(axis int)

type TrainerInterface 

type TrainerInterface interface {
	Train(model ModelInterface, inputs, targets *tensor.Tensor, epochs int, learningRate float32)
	Validate(model ModelInterface, inputs, targets *tensor.Tensor) float32
}

Source Files

View all Source files

Directories

Path Synopsis
cmd
00_standalone_example command
01_polynomial_regression command
02_cnn_mnist_handwrite_recognize command
03_autoencoder command
04_variational_autoencoder command
04_variational_autoencoder/py command
06_layer_testing command
07_autoencoder_compute_graph command
data_store
compute_graph
neuromorphic_compute
tensor
datasets_loader
mnist
pkg
fmt
log
risc-v
rv64-core/cmd/vm command
rv64-core/pkg/rv64
rv64-core/pkg/rv64/cmd/make command
rv64-core/pkg/rv64/cmd/rv64 command
thirdparty
float16
fp16_convert
Package float16 defines support for half-precision floating-point numbers.
 Package float16 defines support for half-precision floating-point numbers.
onnx-go
onnx-go/backend
onnx-go/backend/simple
Package simple holds a very simple graph structure suitable to receive an onnx model
 Package simple holds a very simple graph structure suitable to receive an onnx model
onnx-go/backend/testbackend
Package testbackend provides a set of testing helper functions that test backend interface implementations.
 Package testbackend provides a set of testing helper functions that test backend interface implementations.
onnx-go/backend/testbackend/onnx
Package onnxtest contains an export of the onnx test files
 Package onnxtest contains an export of the onnx test files
onnx-go/backend/testbackend/onnx/gen_cmd command
onnx-go/backend/testbackend/testreport
onnx-go/backend/x/gorgonnx
Package gorgonnx creates a temporary graph that is compatible with backend.ComputationBackend.
 Package gorgonnx creates a temporary graph that is compatible with backend.ComputationBackend.
onnx-go/backend/x/gorgonnx/genapi command
onnx-go/doc/introduction command
Present displays slide presentations and articles.
 Present displays slide presentations and articles.
onnx-go/doc/introduction/samples command
onnx-go/doc/introduction/utils command
onnx-go/examples/emotions command
onnx-go/examples/model_zoo_executor command
onnx-go/examples/tiny_yolov2 command
onnx-go/internal/examples/mnist
onnx-go/internal/examples/mnist/gen command
onnx-go/internal/onnx/ir
onnx-go/internal/readme
onnx-go/internal/tools/dump command
onnx-go/internal/x/images
onnx-go/ir
onnx-go/misc/viewer command
tiktoken
tiktoken/cmd/tokenizer command
tiktoken/codec
tiktoken/internal/cmd command
vision
3d/fbx
image/jpg
video/ffmpeg
video/mp4

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.