quark.onnx.finetuning.create_torch.quant_base_ops

Module Contents

Classes

class quark.onnx.finetuning.create_torch.quant_base_ops.RoundHalfToEven(*args, **kwargs)

Base class to create custom autograd.Function.

To create a custom autograd.Function, subclass this class and implement the :meth:`forward` and :meth:`backward` static methods. Then, to use your custom op in the forward pass, call the class method apply. Do not call :meth:`forward` directly.

To ensure correctness and best performance, make sure you are calling the correct methods on ctx and validating your backward function using torch.autograd.gradcheck().

See extending-autograd for more details on how to use this class.

Examples:

>>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_AUTOGRAD)
>>> class Exp(Function):
>>>     @staticmethod
>>>     def forward(ctx, i):
>>>         result = i.exp()
>>>         ctx.save_for_backward(result)
>>>         return result
>>>
>>>     @staticmethod
>>>     def backward(ctx, grad_output):
>>>         result, = ctx.saved_tensors
>>>         return grad_output * result
>>>
>>> # Use it by calling the apply method:
>>> # xdoctest: +SKIP
>>> output = Exp.apply(input)

static forward(ctx: torch.autograd.Function, t: torch.Tensor) → torch.Tensor

Define the forward of the custom autograd Function.

This function is to be overridden by all subclasses. There are two ways to define forward:

Usage 1 (Combined forward and ctx):

@staticmethod
def forward(ctx: Any, *args: Any, **kwargs: Any) -> Any:
    pass

• It must accept a context ctx as the first argument, followed by any number of arguments (tensors or other types).

• See combining-forward-context for more details

Usage 2 (Separate forward and ctx):

@staticmethod
def forward(*args: Any, **kwargs: Any) -> Any:
    pass

@staticmethod
def setup_context(ctx: Any, inputs: Tuple[Any, ...], output: Any) -> None:
    pass

• The forward no longer accepts a ctx argument.

• Instead, you must also override the :meth:`torch.autograd.Function.setup_context` staticmethod to handle setting up the ctx object. output is the output of the forward, inputs are a Tuple of inputs to the forward.

• See extending-autograd for more details

The context can be used to store arbitrary data that can be then retrieved during the backward pass. Tensors should not be stored directly on ctx (though this is not currently enforced for backward compatibility). Instead, tensors should be saved either with ctx.save_for_backward() if they are intended to be used in backward (equivalently, vjp) or ctx.save_for_forward() if they are intended to be used for in jvp.

static backward(ctx: torch.autograd.Function, grad_output: torch.Tensor) → torch.Tensor

Define a formula for differentiating the operation with backward mode automatic differentiation.

This function is to be overridden by all subclasses. (Defining this function is equivalent to defining the vjp function.)

It must accept a context :attr:`ctx` as the first argument, followed by as many outputs as the forward() returned (None will be passed in for non tensor outputs of the forward function), and it should return as many tensors, as there were inputs to forward(). Each argument is the gradient w.r.t the given output, and each returned value should be the gradient w.r.t. the corresponding input. If an input is not a Tensor or is a Tensor not requiring grads, you can just pass None as a gradient for that input.

The context can be used to retrieve tensors saved during the forward pass. It also has an attribute :attr:`ctx.needs_input_grad` as a tuple of booleans representing whether each input needs gradient. E.g., backward() will have ctx.needs_input_grad[0] = True if the first input to forward() needs gradient computed w.r.t. the output.

class quark.onnx.finetuning.create_torch.quant_base_ops.BaseQuantFunction(*args, **kwargs)

Base class to create custom autograd.Function.

To create a custom autograd.Function, subclass this class and implement the :meth:`forward` and :meth:`backward` static methods. Then, to use your custom op in the forward pass, call the class method apply. Do not call :meth:`forward` directly.

To ensure correctness and best performance, make sure you are calling the correct methods on ctx and validating your backward function using torch.autograd.gradcheck().

See extending-autograd for more details on how to use this class.

Examples:

>>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_AUTOGRAD)
>>> class Exp(Function):
>>>     @staticmethod
>>>     def forward(ctx, i):
>>>         result = i.exp()
>>>         ctx.save_for_backward(result)
>>>         return result
>>>
>>>     @staticmethod
>>>     def backward(ctx, grad_output):
>>>         result, = ctx.saved_tensors
>>>         return grad_output * result
>>>
>>> # Use it by calling the apply method:
>>> # xdoctest: +SKIP
>>> output = Exp.apply(input)

static forward(ctx: Any, tensor: torch.Tensor, scale: torch.Tensor, zero_point: torch.Tensor, min_q: torch.Tensor, max_q: torch.Tensor, round_func: Any) → torch.Tensor

Define the forward of the custom autograd Function.

This function is to be overridden by all subclasses. There are two ways to define forward:

Usage 1 (Combined forward and ctx):

@staticmethod
def forward(ctx: Any, *args: Any, **kwargs: Any) -> Any:
    pass

• It must accept a context ctx as the first argument, followed by any number of arguments (tensors or other types).

• See combining-forward-context for more details

Usage 2 (Separate forward and ctx):

@staticmethod
def forward(*args: Any, **kwargs: Any) -> Any:
    pass

@staticmethod
def setup_context(ctx: Any, inputs: Tuple[Any, ...], output: Any) -> None:
    pass

• The forward no longer accepts a ctx argument.

• Instead, you must also override the :meth:`torch.autograd.Function.setup_context` staticmethod to handle setting up the ctx object. output is the output of the forward, inputs are a Tuple of inputs to the forward.

• See extending-autograd for more details

The context can be used to store arbitrary data that can be then retrieved during the backward pass. Tensors should not be stored directly on ctx (though this is not currently enforced for backward compatibility). Instead, tensors should be saved either with ctx.save_for_backward() if they are intended to be used in backward (equivalently, vjp) or ctx.save_for_forward() if they are intended to be used for in jvp.

static backward(ctx: Any, grad_output: torch.Tensor) → Tuple[Optional[torch.Tensor], None, None, None, None, None]

Define a formula for differentiating the operation with backward mode automatic differentiation.

This function is to be overridden by all subclasses. (Defining this function is equivalent to defining the vjp function.)

It must accept a context :attr:`ctx` as the first argument, followed by as many outputs as the forward() returned (None will be passed in for non tensor outputs of the forward function), and it should return as many tensors, as there were inputs to forward(). Each argument is the gradient w.r.t the given output, and each returned value should be the gradient w.r.t. the corresponding input. If an input is not a Tensor or is a Tensor not requiring grads, you can just pass None as a gradient for that input.

The context can be used to retrieve tensors saved during the forward pass. It also has an attribute :attr:`ctx.needs_input_grad` as a tuple of booleans representing whether each input needs gradient. E.g., backward() will have ctx.needs_input_grad[0] = True if the first input to forward() needs gradient computed w.r.t. the output.

class quark.onnx.finetuning.create_torch.quant_base_ops.BaseDeQuantFunction(*args, **kwargs)

Base class to create custom autograd.Function.

To create a custom autograd.Function, subclass this class and implement the :meth:`forward` and :meth:`backward` static methods. Then, to use your custom op in the forward pass, call the class method apply. Do not call :meth:`forward` directly.

To ensure correctness and best performance, make sure you are calling the correct methods on ctx and validating your backward function using torch.autograd.gradcheck().

See extending-autograd for more details on how to use this class.

Examples:

>>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_AUTOGRAD)
>>> class Exp(Function):
>>>     @staticmethod
>>>     def forward(ctx, i):
>>>         result = i.exp()
>>>         ctx.save_for_backward(result)
>>>         return result
>>>
>>>     @staticmethod
>>>     def backward(ctx, grad_output):
>>>         result, = ctx.saved_tensors
>>>         return grad_output * result
>>>
>>> # Use it by calling the apply method:
>>> # xdoctest: +SKIP
>>> output = Exp.apply(input)

static forward(ctx: torch.autograd.Function, tensor: torch.Tensor, scale: torch.Tensor, zero_point: torch.Tensor) → torch.Tensor

Define the forward of the custom autograd Function.

This function is to be overridden by all subclasses. There are two ways to define forward:

Usage 1 (Combined forward and ctx):

@staticmethod
def forward(ctx: Any, *args: Any, **kwargs: Any) -> Any:
    pass

• It must accept a context ctx as the first argument, followed by any number of arguments (tensors or other types).

• See combining-forward-context for more details

Usage 2 (Separate forward and ctx):

@staticmethod
def forward(*args: Any, **kwargs: Any) -> Any:
    pass

@staticmethod
def setup_context(ctx: Any, inputs: Tuple[Any, ...], output: Any) -> None:
    pass

• The forward no longer accepts a ctx argument.

• Instead, you must also override the :meth:`torch.autograd.Function.setup_context` staticmethod to handle setting up the ctx object. output is the output of the forward, inputs are a Tuple of inputs to the forward.

• See extending-autograd for more details

The context can be used to store arbitrary data that can be then retrieved during the backward pass. Tensors should not be stored directly on ctx (though this is not currently enforced for backward compatibility). Instead, tensors should be saved either with ctx.save_for_backward() if they are intended to be used in backward (equivalently, vjp) or ctx.save_for_forward() if they are intended to be used for in jvp.

static backward(ctx: torch.autograd.Function, grad_output: torch.Tensor) → Tuple[torch.Tensor, None, None]

Define a formula for differentiating the operation with backward mode automatic differentiation.

This function is to be overridden by all subclasses. (Defining this function is equivalent to defining the vjp function.)

It must accept a context :attr:`ctx` as the first argument, followed by as many outputs as the forward() returned (None will be passed in for non tensor outputs of the forward function), and it should return as many tensors, as there were inputs to forward(). Each argument is the gradient w.r.t the given output, and each returned value should be the gradient w.r.t. the corresponding input. If an input is not a Tensor or is a Tensor not requiring grads, you can just pass None as a gradient for that input.

The context can be used to retrieve tensors saved during the forward pass. It also has an attribute :attr:`ctx.needs_input_grad` as a tuple of booleans representing whether each input needs gradient. E.g., backward() will have ctx.needs_input_grad[0] = True if the first input to forward() needs gradient computed w.r.t. the output.

class quark.onnx.finetuning.create_torch.quant_base_ops.BaseQuantDequantFunction(*args, **kwargs)

Base class to create custom autograd.Function.

To create a custom autograd.Function, subclass this class and implement the :meth:`forward` and :meth:`backward` static methods. Then, to use your custom op in the forward pass, call the class method apply. Do not call :meth:`forward` directly.

To ensure correctness and best performance, make sure you are calling the correct methods on ctx and validating your backward function using torch.autograd.gradcheck().

See extending-autograd for more details on how to use this class.

Examples:

>>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_AUTOGRAD)
>>> class Exp(Function):
>>>     @staticmethod
>>>     def forward(ctx, i):
>>>         result = i.exp()
>>>         ctx.save_for_backward(result)
>>>         return result
>>>
>>>     @staticmethod
>>>     def backward(ctx, grad_output):
>>>         result, = ctx.saved_tensors
>>>         return grad_output * result
>>>
>>> # Use it by calling the apply method:
>>> # xdoctest: +SKIP
>>> output = Exp.apply(input)

static forward(ctx: Any, tensor: torch.Tensor, scale: torch.Tensor, zero_point: torch.Tensor, min_q: torch.Tensor, max_q: torch.Tensor, round_func: Any) → torch.Tensor

Define the forward of the custom autograd Function.

This function is to be overridden by all subclasses. There are two ways to define forward:

Usage 1 (Combined forward and ctx):

@staticmethod
def forward(ctx: Any, *args: Any, **kwargs: Any) -> Any:
    pass

• It must accept a context ctx as the first argument, followed by any number of arguments (tensors or other types).

• See combining-forward-context for more details

Usage 2 (Separate forward and ctx):

@staticmethod
def forward(*args: Any, **kwargs: Any) -> Any:
    pass

@staticmethod
def setup_context(ctx: Any, inputs: Tuple[Any, ...], output: Any) -> None:
    pass

• The forward no longer accepts a ctx argument.

• Instead, you must also override the :meth:`torch.autograd.Function.setup_context` staticmethod to handle setting up the ctx object. output is the output of the forward, inputs are a Tuple of inputs to the forward.

• See extending-autograd for more details

The context can be used to store arbitrary data that can be then retrieved during the backward pass. Tensors should not be stored directly on ctx (though this is not currently enforced for backward compatibility). Instead, tensors should be saved either with ctx.save_for_backward() if they are intended to be used in backward (equivalently, vjp) or ctx.save_for_forward() if they are intended to be used for in jvp.

static backward(ctx: Any, grad_output: torch.Tensor) → Tuple[torch.Tensor, None, None, None, None, None]

Define a formula for differentiating the operation with backward mode automatic differentiation.

This function is to be overridden by all subclasses. (Defining this function is equivalent to defining the vjp function.)

It must accept a context :attr:`ctx` as the first argument, followed by as many outputs as the forward() returned (None will be passed in for non tensor outputs of the forward function), and it should return as many tensors, as there were inputs to forward(). Each argument is the gradient w.r.t the given output, and each returned value should be the gradient w.r.t. the corresponding input. If an input is not a Tensor or is a Tensor not requiring grads, you can just pass None as a gradient for that input.

The context can be used to retrieve tensors saved during the forward pass. It also has an attribute :attr:`ctx.needs_input_grad` as a tuple of booleans representing whether each input needs gradient. E.g., backward() will have ctx.needs_input_grad[0] = True if the first input to forward() needs gradient computed w.r.t. the output.

class quark.onnx.finetuning.create_torch.quant_base_ops.INTQuantFunction(*args, **kwargs)

Base class to create custom autograd.Function.

To create a custom autograd.Function, subclass this class and implement the :meth:`forward` and :meth:`backward` static methods. Then, to use your custom op in the forward pass, call the class method apply. Do not call :meth:`forward` directly.

To ensure correctness and best performance, make sure you are calling the correct methods on ctx and validating your backward function using torch.autograd.gradcheck().

See extending-autograd for more details on how to use this class.

Examples:

>>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_AUTOGRAD)
>>> class Exp(Function):
>>>     @staticmethod
>>>     def forward(ctx, i):
>>>         result = i.exp()
>>>         ctx.save_for_backward(result)
>>>         return result
>>>
>>>     @staticmethod
>>>     def backward(ctx, grad_output):
>>>         result, = ctx.saved_tensors
>>>         return grad_output * result
>>>
>>> # Use it by calling the apply method:
>>> # xdoctest: +SKIP
>>> output = Exp.apply(input)

static forward(ctx: torch.autograd.Function, tensor: torch.Tensor, scale: torch.Tensor, zero_point: torch.Tensor, min_q: torch.Tensor, max_q: torch.Tensor, round_func: torch.autograd.Function) → torch.Tensor

Define the forward of the custom autograd Function.

This function is to be overridden by all subclasses. There are two ways to define forward:

Usage 1 (Combined forward and ctx):

@staticmethod
def forward(ctx: Any, *args: Any, **kwargs: Any) -> Any:
    pass

• It must accept a context ctx as the first argument, followed by any number of arguments (tensors or other types).

• See combining-forward-context for more details

Usage 2 (Separate forward and ctx):

@staticmethod
def forward(*args: Any, **kwargs: Any) -> Any:
    pass

@staticmethod
def setup_context(ctx: Any, inputs: Tuple[Any, ...], output: Any) -> None:
    pass

• The forward no longer accepts a ctx argument.

• Instead, you must also override the :meth:`torch.autograd.Function.setup_context` staticmethod to handle setting up the ctx object. output is the output of the forward, inputs are a Tuple of inputs to the forward.

• See extending-autograd for more details

The context can be used to store arbitrary data that can be then retrieved during the backward pass. Tensors should not be stored directly on ctx (though this is not currently enforced for backward compatibility). Instead, tensors should be saved either with ctx.save_for_backward() if they are intended to be used in backward (equivalently, vjp) or ctx.save_for_forward() if they are intended to be used for in jvp.

static backward(ctx: torch.autograd.Function, grad_output: torch.Tensor) → Tuple[torch.Tensor, None, None, None, None, None]

Define a formula for differentiating the operation with backward mode automatic differentiation.

This function is to be overridden by all subclasses. (Defining this function is equivalent to defining the vjp function.)

It must accept a context :attr:`ctx` as the first argument, followed by as many outputs as the forward() returned (None will be passed in for non tensor outputs of the forward function), and it should return as many tensors, as there were inputs to forward(). Each argument is the gradient w.r.t the given output, and each returned value should be the gradient w.r.t. the corresponding input. If an input is not a Tensor or is a Tensor not requiring grads, you can just pass None as a gradient for that input.

The context can be used to retrieve tensors saved during the forward pass. It also has an attribute :attr:`ctx.needs_input_grad` as a tuple of booleans representing whether each input needs gradient. E.g., backward() will have ctx.needs_input_grad[0] = True if the first input to forward() needs gradient computed w.r.t. the output.

class quark.onnx.finetuning.create_torch.quant_base_ops.INTDeQuantFunction(*args, **kwargs)

Base class to create custom autograd.Function.

To create a custom autograd.Function, subclass this class and implement the :meth:`forward` and :meth:`backward` static methods. Then, to use your custom op in the forward pass, call the class method apply. Do not call :meth:`forward` directly.

To ensure correctness and best performance, make sure you are calling the correct methods on ctx and validating your backward function using torch.autograd.gradcheck().

See extending-autograd for more details on how to use this class.

Examples:

>>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_AUTOGRAD)
>>> class Exp(Function):
>>>     @staticmethod
>>>     def forward(ctx, i):
>>>         result = i.exp()
>>>         ctx.save_for_backward(result)
>>>         return result
>>>
>>>     @staticmethod
>>>     def backward(ctx, grad_output):
>>>         result, = ctx.saved_tensors
>>>         return grad_output * result
>>>
>>> # Use it by calling the apply method:
>>> # xdoctest: +SKIP
>>> output = Exp.apply(input)

static forward(ctx: Any, tensor: torch.Tensor, scale: torch.Tensor, zero_point: torch.Tensor) → torch.Tensor

Define the forward of the custom autograd Function.

This function is to be overridden by all subclasses. There are two ways to define forward:

Usage 1 (Combined forward and ctx):

@staticmethod
def forward(ctx: Any, *args: Any, **kwargs: Any) -> Any:
    pass

• It must accept a context ctx as the first argument, followed by any number of arguments (tensors or other types).

• See combining-forward-context for more details

Usage 2 (Separate forward and ctx):

@staticmethod
def forward(*args: Any, **kwargs: Any) -> Any:
    pass

@staticmethod
def setup_context(ctx: Any, inputs: Tuple[Any, ...], output: Any) -> None:
    pass

• The forward no longer accepts a ctx argument.

• Instead, you must also override the :meth:`torch.autograd.Function.setup_context` staticmethod to handle setting up the ctx object. output is the output of the forward, inputs are a Tuple of inputs to the forward.

• See extending-autograd for more details

The context can be used to store arbitrary data that can be then retrieved during the backward pass. Tensors should not be stored directly on ctx (though this is not currently enforced for backward compatibility). Instead, tensors should be saved either with ctx.save_for_backward() if they are intended to be used in backward (equivalently, vjp) or ctx.save_for_forward() if they are intended to be used for in jvp.

static backward(ctx: Any, grad_output: torch.Tensor) → Tuple[torch.Tensor, None, None]

Define a formula for differentiating the operation with backward mode automatic differentiation.

This function is to be overridden by all subclasses. (Defining this function is equivalent to defining the vjp function.)

It must accept a context :attr:`ctx` as the first argument, followed by as many outputs as the forward() returned (None will be passed in for non tensor outputs of the forward function), and it should return as many tensors, as there were inputs to forward(). Each argument is the gradient w.r.t the given output, and each returned value should be the gradient w.r.t. the corresponding input. If an input is not a Tensor or is a Tensor not requiring grads, you can just pass None as a gradient for that input.

The context can be used to retrieve tensors saved during the forward pass. It also has an attribute :attr:`ctx.needs_input_grad` as a tuple of booleans representing whether each input needs gradient. E.g., backward() will have ctx.needs_input_grad[0] = True if the first input to forward() needs gradient computed w.r.t. the output.

class quark.onnx.finetuning.create_torch.quant_base_ops.INTQuantDequantFunction(*args, **kwargs)

Base class to create custom autograd.Function.

To create a custom autograd.Function, subclass this class and implement the :meth:`forward` and :meth:`backward` static methods. Then, to use your custom op in the forward pass, call the class method apply. Do not call :meth:`forward` directly.

To ensure correctness and best performance, make sure you are calling the correct methods on ctx and validating your backward function using torch.autograd.gradcheck().

See extending-autograd for more details on how to use this class.

Examples:

>>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_AUTOGRAD)
>>> class Exp(Function):
>>>     @staticmethod
>>>     def forward(ctx, i):
>>>         result = i.exp()
>>>         ctx.save_for_backward(result)
>>>         return result
>>>
>>>     @staticmethod
>>>     def backward(ctx, grad_output):
>>>         result, = ctx.saved_tensors
>>>         return grad_output * result
>>>
>>> # Use it by calling the apply method:
>>> # xdoctest: +SKIP
>>> output = Exp.apply(input)

static forward(ctx: torch.autograd.Function, tensor: torch.Tensor, scale: torch.Tensor, zero_point: torch.Tensor, min_q: torch.Tensor, max_q: torch.Tensor, round_func: torch.autograd.Function) → torch.Tensor

Define the forward of the custom autograd Function.

This function is to be overridden by all subclasses. There are two ways to define forward:

Usage 1 (Combined forward and ctx):

@staticmethod
def forward(ctx: Any, *args: Any, **kwargs: Any) -> Any:
    pass

• It must accept a context ctx as the first argument, followed by any number of arguments (tensors or other types).

• See combining-forward-context for more details

Usage 2 (Separate forward and ctx):

@staticmethod
def forward(*args: Any, **kwargs: Any) -> Any:
    pass

@staticmethod
def setup_context(ctx: Any, inputs: Tuple[Any, ...], output: Any) -> None:
    pass

• The forward no longer accepts a ctx argument.

• Instead, you must also override the :meth:`torch.autograd.Function.setup_context` staticmethod to handle setting up the ctx object. output is the output of the forward, inputs are a Tuple of inputs to the forward.

• See extending-autograd for more details

The context can be used to store arbitrary data that can be then retrieved during the backward pass. Tensors should not be stored directly on ctx (though this is not currently enforced for backward compatibility). Instead, tensors should be saved either with ctx.save_for_backward() if they are intended to be used in backward (equivalently, vjp) or ctx.save_for_forward() if they are intended to be used for in jvp.

static backward(ctx: torch.autograd.Function, grad_output: torch.Tensor) → Tuple[torch.Tensor, None, None, None, None, None]

Define a formula for differentiating the operation with backward mode automatic differentiation.

This function is to be overridden by all subclasses. (Defining this function is equivalent to defining the vjp function.)

It must accept a context :attr:`ctx` as the first argument, followed by as many outputs as the forward() returned (None will be passed in for non tensor outputs of the forward function), and it should return as many tensors, as there were inputs to forward(). Each argument is the gradient w.r.t the given output, and each returned value should be the gradient w.r.t. the corresponding input. If an input is not a Tensor or is a Tensor not requiring grads, you can just pass None as a gradient for that input.

The context can be used to retrieve tensors saved during the forward pass. It also has an attribute :attr:`ctx.needs_input_grad` as a tuple of booleans representing whether each input needs gradient. E.g., backward() will have ctx.needs_input_grad[0] = True if the first input to forward() needs gradient computed w.r.t. the output.

class quark.onnx.finetuning.create_torch.quant_base_ops.Quantizer(scale: torch.Tensor, zero_point: torch.Tensor, min_q: torch.Tensor, max_q: torch.Tensor, ch_axis: int = 0, q_folded: bool = False)

Standard Quantizer has three functions including quantize, dequantize and quantize_dequantize, which is corresponding to ONNX QuantizeLinear, DequantizeLinear and Q/DQ pair separately. By default in forward, it works in quantize_dequantize mode.

round_impl(tensor: torch.Tensor) → None

Implement the round function, designed for adaround quantizer

tensor_sync(tensor: torch.Tensor) → None

The Pre-processing of the parameter according to the input tensor

class quark.onnx.finetuning.create_torch.quant_base_ops.AdaroundConstants

Constants used for Adarounding

class quark.onnx.finetuning.create_torch.quant_base_ops.AdaroundQuantizer(scale: torch.Tensor, zero_point: torch.Tensor, min_q: torch.Tensor, max_q: torch.Tensor, ch_axis: int = 0, q_folded: bool = False)

Adaround Quantizer has a alpha paramter for optimizing weight rounding

round_impl(tensor: torch.Tensor) → None

Implement the rounding function for adaround :param weight: The tensor to be ada-rounded

initialize_alpha(tensor: torch.Tensor) → None

Initializes alpha parameter, same shape as the tensor :param tensor: The tensor to be ada-rounded

class quark.onnx.finetuning.create_torch.quant_base_ops.QDQModule(scale: torch.Tensor, zero_point: torch.Tensor, min_q: torch.Tensor, max_q: torch.Tensor, ch_axis: int = 0, q_folded: bool = False)

A pytorch module that behaves as ONNX’s QDQ pair

class quark.onnx.finetuning.create_torch.quant_base_ops.QuantizeWrapper(w_alpha: float = 1.0, b_beta: float = 1.0, **kwargs: Dict[str, Any])

A wrapper for torch layer’s input/weight/bias quantization