Mojo trait
QuantizationEncoding
Describes the encoding for a data type that can be quantized.
Any type that conforms to this trait implicitly knows the relationship
between the buffer and tensor layout for its quantization encoding.
In particular, the quantize() function takes in a tensor with a logical
shape in terms of its elements. Then it returns a uint8 tensor with a
different shape that instead describes the shape of the bytes storage
buffer after applying the quantization encoding.
Implemented traits
AnyType,
UnknownDestructibility
Methods
quantize
static quantize(tensor: Tensor[float32]) -> Tensor[uint8]
Quantizes the full-precision tensor to the quantized type associated with this QuantizationEncoding instance.
Args:
- tensor (
Tensor[float32]): Full-precision tensor to quantize.
Returns:
A Tensor quantized to the quantized storage format of this QuantizationEncoding instance. The tensor datatype is uint8 because this is simply a bytes buffer. The actual data structure in that buffer depends on the encoding.
id
static id() -> String
Returns a unique string identifier for this quantization encoding.
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!