AutocastScope
Models & Types · AiDotNet.Tensors.Engines.Gpu
No summary documentation available yet.
Implements: IDisposable
Constructors
| Constructor | Summary |
|---|---|
AutocastScope(PrecisionMode precision) |
|
AutocastScope(PrecisionMode precision, LayerPrecisionPolicy policy) |
Properties
| Property | Summary |
|---|---|
AutocastScope Current { get; } |
|
LayerPrecisionPolicy Policy { get; } |
|
PrecisionMode ActivePrecision { get; } |
|
PrecisionMode Precision { get; } |
|
bool IsEnabled { get; } |
Methods
| Method | Summary |
|---|---|
AutocastScope EnableFromEnvironment() |
|
IGpuBuffer MaybeConvertInput(IDirectGpuBackend backend, IGpuBuffer fp32Buffer, int size, GpuPrecisionOperation operation) |
|
IGpuBuffer MaybeConvertOutput(IDirectGpuBackend backend, IGpuBuffer resultBuffer, int size, GpuPrecisionOperation operation) |
|
PrecisionMode EnvironmentRequestedPrecision() |
|
Tensor<Half> CastToFP16(Tensor<float> fp32) |
|
Tensor<Half> GetFP16Tensor(string name) |
|
Tensor<Half> RegisterAndCastToFP16(string name, Tensor<float> fp32) |
|
Tensor<float> CastToFP32(Tensor<Half> fp16) |
|
Tensor<float> GetFP32Tensor(string name) |
|
bool HasTensor(string name) |
|
bool ShouldAutocast(string operationName) |
|
bool ShouldUseFP32(string layerName) |
|
bool ShouldUseHigherPrecision(string layerName) |
|
void ClearTensors() |
|
void Dispose() |