tensor_parser

Classes

Functions

parse_blockwise_quantization

parse_blockwise_quantization(details: BlockwiseQuantizationT) -> AirBlockwiseQuantization

Create AirBlockwiseQuantization from LiteRT BlockwiseQuantizationT.

parse_dimension_meta_data

parse_dimension_meta_data(meta: DimensionMetadataT) -> AirDimensionMetaData

Create AirDimensionMetaData from LiteRT DimensionMetadataT.

parse_sparsity

parse_sparsity(sparsity: SparsityParametersT | None) -> AirSparsityParameters | None

Create sparsity parameters from LiteRT SparsityParametersT.

parse_quantization_parameters

parse_quantization_parameters(quantization: QuantizationParametersT | None) -> AirQuantizationParameters | None

Create quantization parameters from LiteRT QuantizationParametersT.

Returns None for unquantized tensors (fp16/fp32), which carry an empty QuantizationParameters stub in the flatbuffer with no scale or zero_point.

parse_tensor

parse_tensor(id: int, tensor: TensorT, model: ModelT, subgraph: SubGraphT) -> AirTensor

Create an AirTensor from a LiteRT Tensor.