torch.func API Reference

Function Transforms

vmap	vmap is the vectorizing map; vmap(func) returns a new function that maps func over some dimension of the inputs.
grad	grad operator helps computing gradients of func with respect to the input(s) specified by argnums.
grad_and_value	Returns a function to compute a tuple of the gradient and primal, or forward, computation.
vjp	Standing for the vector-Jacobian product, returns a tuple containing the results of func applied to primals and a function that, when given cotangents, computes the reverse-mode Jacobian of func with respect to primals times cotangents.
jvp	Standing for the Jacobian-vector product, returns a tuple containing the output of func(*primals) and the "Jacobian of func evaluated at primals" times tangents.
linearize	Returns the value of func at primals and linear approximation at primals.
jacrev	Computes the Jacobian of func with respect to the arg(s) at index argnum using reverse mode autodiff
jacfwd	Computes the Jacobian of func with respect to the arg(s) at index argnum using forward-mode autodiff
hessian	Computes the Hessian of func with respect to the arg(s) at index argnum via a forward-over-reverse strategy.
functionalize	functionalize is a transform that can be used to remove (intermediate) mutations and aliasing from a function, while preserving the function's semantics.

Utilities for working with torch.nn.Modules

In general, you can transform over a function that calls a torch.nn.Module. For example, the following is an example of computing a jacobian of a function that takes three values and returns three values:

model = torch.nn.Linear(3, 3)

def f(x):
    return model(x)

x = torch.randn(3)
jacobian = jacrev(f)(x)
assert jacobian.shape == (3, 3)

However, if you want to do something like compute a jacobian over the parameters of the model, then there needs to be a way to construct a function where the parameters are the inputs to the function. That’s what functional_call() is for: it accepts an nn.Module, the transformed parameters, and the inputs to the Module’s forward pass. It returns the value of running the Module’s forward pass with the replaced parameters.

Here’s how we would compute the Jacobian over the parameters

model = torch.nn.Linear(3, 3)

def f(params, x):
    return torch.func.functional_call(model, params, x)

x = torch.randn(3)
jacobian = jacrev(f)(dict(model.named_parameters()), x)

functional_call	Performs a functional call on the module by replacing the module parameters and buffers with the provided ones.
stack_module_state	Prepares a list of torch.nn.Modules for ensembling with vmap().
replace_all_batch_norm_modules_	In place updates root by setting the running_mean and running_var to be None and setting track_running_stats to be False for any nn.BatchNorm module in root

If you’re looking for information on fixing Batch Norm modules, please follow the guidance here

• Patching Batch Norm