Module furiosa.runtime.furiosa_sdk_runtime.model

Model and its methods to access model metadata

"""Model and its methods to access model metadata"""
from abc import ABC, abstractmethod
from ctypes import c_void_p

from ._api import LIBNUX
from .tensor import TensorDesc, TensorArray
from ._util import list_to_dict


class Model(ABC):
    """NPU model binary compiled by Renegade compiler"""

    @abstractmethod
    def _get_model_ref(self) -> c_void_p:
        """
        Returns a raw model pointer

        :return: a raw pointer of a Model
        """

    def input_num(self) -> int:
        """Number of input tensors of Model"""
        return LIBNUX.nux_input_num(self._get_model_ref())

    def output_num(self) -> int:
        """Number of output tensors of Model"""
        return LIBNUX.nux_output_num(self._get_model_ref())

    def input(self, idx) -> TensorDesc:
        """Return tensor description of i-th input tensor of Model"""
        return TensorDesc(LIBNUX.nux_input_desc(self._get_model_ref(), idx))

    def inputs(self) -> [TensorDesc]:
        """Tensor descriptions of all input tensors of Model"""
        inputs = []
        for idx in range(self.input_num()):
            inputs.append(self.input(idx))

        return inputs

    def output(self, idx) -> TensorDesc:
        """Returns tensor description of i-th output tensor of Model"""
        return TensorDesc(LIBNUX.nux_output_desc(self._get_model_ref(), idx))

    def outputs(self) -> [TensorDesc]:
        """Tensor descriptions of all output tensors of Model"""
        outputs = []
        for idx in range(self.output_num()):
            outputs.append(self.output(idx))

        return outputs

    def allocate_inputs(self) -> TensorArray:
        """Creates an array of input tensors with allocated buffers"""
        return TensorArray(LIBNUX.nux_tensor_array_create_inputs(self._get_model_ref()),
                           self.inputs(), allocated=True)

    def allocate_outputs(self) -> TensorArray:
        """Creates an array of output tensors with allocated buffers"""
        return TensorArray(LIBNUX.nux_tensor_array_allocate_outputs(self._get_model_ref()),
                           self.outputs(), allocated=True)

    def create_outputs(self) -> TensorArray:
        """Creates an array of output tensors without allocated buffers"""
        return TensorArray(LIBNUX.nux_tensor_array_create_outputs(self._get_model_ref()),
                           self.outputs(), allocated=True)

    def summary(self) -> str:
        """Returns the summary of this model"""
        return "Inputs:\n{}\nOutputs:\n{}".format(
            list_to_dict(self.inputs()).__repr__(), list_to_dict(self.outputs()).__repr__())

    def print_summary(self):
        """Prints the summary of this model"""
        print(self.summary())

Classes

class Model

NPU model binary compiled by Renegade compiler

Expand source code

class Model(ABC):
    """NPU model binary compiled by Renegade compiler"""

    @abstractmethod
    def _get_model_ref(self) -> c_void_p:
        """
        Returns a raw model pointer

        :return: a raw pointer of a Model
        """

    def input_num(self) -> int:
        """Number of input tensors of Model"""
        return LIBNUX.nux_input_num(self._get_model_ref())

    def output_num(self) -> int:
        """Number of output tensors of Model"""
        return LIBNUX.nux_output_num(self._get_model_ref())

    def input(self, idx) -> TensorDesc:
        """Return tensor description of i-th input tensor of Model"""
        return TensorDesc(LIBNUX.nux_input_desc(self._get_model_ref(), idx))

    def inputs(self) -> [TensorDesc]:
        """Tensor descriptions of all input tensors of Model"""
        inputs = []
        for idx in range(self.input_num()):
            inputs.append(self.input(idx))

        return inputs

    def output(self, idx) -> TensorDesc:
        """Returns tensor description of i-th output tensor of Model"""
        return TensorDesc(LIBNUX.nux_output_desc(self._get_model_ref(), idx))

    def outputs(self) -> [TensorDesc]:
        """Tensor descriptions of all output tensors of Model"""
        outputs = []
        for idx in range(self.output_num()):
            outputs.append(self.output(idx))

        return outputs

    def allocate_inputs(self) -> TensorArray:
        """Creates an array of input tensors with allocated buffers"""
        return TensorArray(LIBNUX.nux_tensor_array_create_inputs(self._get_model_ref()),
                           self.inputs(), allocated=True)

    def allocate_outputs(self) -> TensorArray:
        """Creates an array of output tensors with allocated buffers"""
        return TensorArray(LIBNUX.nux_tensor_array_allocate_outputs(self._get_model_ref()),
                           self.outputs(), allocated=True)

    def create_outputs(self) -> TensorArray:
        """Creates an array of output tensors without allocated buffers"""
        return TensorArray(LIBNUX.nux_tensor_array_create_outputs(self._get_model_ref()),
                           self.outputs(), allocated=True)

    def summary(self) -> str:
        """Returns the summary of this model"""
        return "Inputs:\n{}\nOutputs:\n{}".format(
            list_to_dict(self.inputs()).__repr__(), list_to_dict(self.outputs()).__repr__())

    def print_summary(self):
        """Prints the summary of this model"""
        print(self.summary())

Ancestors

• abc.ABC

Subclasses

• AsyncSession
• Session

Methods

def allocate_inputs(self) ‑> TensorArray

Creates an array of input tensors with allocated buffers

Expand source code

def allocate_inputs(self) -> TensorArray:
    """Creates an array of input tensors with allocated buffers"""
    return TensorArray(LIBNUX.nux_tensor_array_create_inputs(self._get_model_ref()),
                       self.inputs(), allocated=True)

def allocate_outputs(self) ‑> TensorArray

Creates an array of output tensors with allocated buffers

Expand source code

def allocate_outputs(self) -> TensorArray:
    """Creates an array of output tensors with allocated buffers"""
    return TensorArray(LIBNUX.nux_tensor_array_allocate_outputs(self._get_model_ref()),
                       self.outputs(), allocated=True)

def create_outputs(self) ‑> TensorArray

Creates an array of output tensors without allocated buffers

Expand source code

def create_outputs(self) -> TensorArray:
    """Creates an array of output tensors without allocated buffers"""
    return TensorArray(LIBNUX.nux_tensor_array_create_outputs(self._get_model_ref()),
                       self.outputs(), allocated=True)

def input(self, idx) ‑> TensorDesc

Return tensor description of i-th input tensor of Model

Expand source code

def input(self, idx) -> TensorDesc:
    """Return tensor description of i-th input tensor of Model"""
    return TensorDesc(LIBNUX.nux_input_desc(self._get_model_ref(), idx))

def input_num(self) ‑> int

Number of input tensors of Model

Expand source code

def input_num(self) -> int:
    """Number of input tensors of Model"""
    return LIBNUX.nux_input_num(self._get_model_ref())

def inputs(self) ‑> [TensorDesc'>]

Tensor descriptions of all input tensors of Model

Expand source code

def inputs(self) -> [TensorDesc]:
    """Tensor descriptions of all input tensors of Model"""
    inputs = []
    for idx in range(self.input_num()):
        inputs.append(self.input(idx))

    return inputs

def output(self, idx) ‑> TensorDesc

Returns tensor description of i-th output tensor of Model

Expand source code

def output(self, idx) -> TensorDesc:
    """Returns tensor description of i-th output tensor of Model"""
    return TensorDesc(LIBNUX.nux_output_desc(self._get_model_ref(), idx))

def output_num(self) ‑> int

Number of output tensors of Model

Expand source code

def output_num(self) -> int:
    """Number of output tensors of Model"""
    return LIBNUX.nux_output_num(self._get_model_ref())

def outputs(self) ‑> [TensorDesc'>]

Tensor descriptions of all output tensors of Model

Expand source code

def outputs(self) -> [TensorDesc]:
    """Tensor descriptions of all output tensors of Model"""
    outputs = []
    for idx in range(self.output_num()):
        outputs.append(self.output(idx))

    return outputs

def print_summary(self)

Prints the summary of this model

Expand source code

def print_summary(self):
    """Prints the summary of this model"""
    print(self.summary())

def summary(self) ‑> str

Returns the summary of this model

Expand source code

def summary(self) -> str:
    """Returns the summary of this model"""
    return "Inputs:\n{}\nOutputs:\n{}".format(
        list_to_dict(self.inputs()).__repr__(), list_to_dict(self.outputs()).__repr__())