Mila::Dnn::Quant::Weight::PerGroupInt4< kGroupSize > Struct Template Reference

Static Public Attributes
static constexpr bool	kIsFp4E2M1 = false
static constexpr bool	kIsQuantized = true
static constexpr bool	kPerChannel = false
static constexpr int	kQuantizationGroupSize = kGroupSize
static constexpr TensorDataType	kScaleDtype = TensorDataType::FP32
static constexpr TensorDataType	kStorageDtype = TensorDataType::UINT8

Member Data Documentation

kIsFp4E2M1

template<int kGroupSize = 128>

bool Mila::Dnn::Quant::Weight::PerGroupInt4< kGroupSize >::kIsFp4E2M1 = false

staticconstexpr

kIsQuantized

template<int kGroupSize = 128>

bool Mila::Dnn::Quant::Weight::PerGroupInt4< kGroupSize >::kIsQuantized = true

staticconstexpr

kPerChannel

template<int kGroupSize = 128>

bool Mila::Dnn::Quant::Weight::PerGroupInt4< kGroupSize >::kPerChannel = false

staticconstexpr

kQuantizationGroupSize

template<int kGroupSize = 128>

int Mila::Dnn::Quant::Weight::PerGroupInt4< kGroupSize >::kQuantizationGroupSize = kGroupSize

staticconstexpr

kScaleDtype

template<int kGroupSize = 128>

TensorDataType Mila::Dnn::Quant::Weight::PerGroupInt4< kGroupSize >::kScaleDtype = TensorDataType::FP32

staticconstexpr

kStorageDtype

template<int kGroupSize = 128>

TensorDataType Mila::Dnn::Quant::Weight::PerGroupInt4< kGroupSize >::kStorageDtype = TensorDataType::UINT8

staticconstexpr

---

The documentation for this struct was generated from the following file:

• /__w/Mila/Mila/Mila/Src/Dnn/Quantization/Weight/Policies.ixx