| import torch |
| from torch.types import _TensorOrTensors |
| import torch.testing |
| from torch.overrides import is_tensor_like |
| import collections |
| from itertools import product |
| import warnings |
| from typing import Callable, Union, Optional, Iterable, List, Tuple, Dict |
| from torch._vmap_internals import vmap |
| import functools |
| |
| |
| class GradcheckError(RuntimeError): |
| # Custom error so that user errors are not caught in the main try catch (see gradcheck()) |
| pass |
| |
| |
| def is_float_or_complex_tensor(obj): |
| return is_tensor_like(obj) and (obj.is_floating_point() or obj.is_complex()) |
| |
| |
| def allocate_jacobians_with_inputs(input_tensors: Tuple, numel_output) -> Tuple[torch.Tensor, ...]: |
| # Makes zero-filled tensors from inputs. If `numel_output` is not None, for each tensor in |
| # `input_tensors`, returns a new zero-filled tensor with height of `t.numel` and width |
| # of `numel_output`. Otherwise, for each tensor, returns a 1-d tensor with size `(t.numel,)`. |
| # Each new tensor will be strided and have the same dtype and device as those of the |
| # corresponding input |
| out: List[torch.Tensor] = [] |
| for t in input_tensors: |
| if is_float_or_complex_tensor(t) and t.requires_grad: |
| out.append(t.new_zeros((t.numel(), numel_output), layout=torch.strided)) |
| return tuple(out) |
| |
| |
| def allocate_jacobians_with_outputs(output_tensors: Tuple, numel_input, dtype=None, |
| device=None) -> Tuple[torch.Tensor, ...]: |
| # Makes zero-filled tensors from outputs. If `dim` is not None, for each tensor in |
| # `output_tensors`, returns a new zero-filled tensor with height of `dim` and width of |
| # `t.numel`. Otherwise, for each tensor, returns a 1-d tensor with size (t.numel,). |
| out: List[torch.Tensor] = [] |
| options = {"dtype": dtype, "device": device, "layout": torch.strided} |
| for t in output_tensors: |
| if is_float_or_complex_tensor(t): |
| out.append(t.new_zeros((numel_input, t.numel()), **options)) |
| return tuple(out) |
| |
| |
| def iter_tensors(x: Union[torch.Tensor, Iterable[torch.Tensor]], only_requiring_grad: bool = False) -> Iterable[torch.Tensor]: |
| if is_tensor_like(x): |
| # mypy doesn't narrow type of `x` to torch.Tensor |
| if x.requires_grad or not only_requiring_grad: # type: ignore[union-attr] |
| yield x # type: ignore[misc] |
| elif isinstance(x, collections.abc.Iterable) and not isinstance(x, str): |
| for elem in x: |
| for result in iter_tensors(elem, only_requiring_grad): |
| yield result |
| |
| |
| def iter_tensor(x_tensor): |
| # Enumerates over a tensor and provides a corresponding flat index that translates |
| # to a given rol/col in the jacobian matrix. The order is the same as as if we flatten |
| # a contiguous tensor. iter_tensor also returns a strided version of the original |
| # tensor that is able to be modified inplace. If the input tensor is strided or sparse, |
| # the returned tensor will share storage with the original. Otherwise, for opaque tensor |
| # types like mkldnn, a copy is returned. |
| # |
| # Example: |
| # for a tensor t with size (2, 2), it will yield: |
| # `x, (0, 0), 0`, `x, (0, 1), 1`, `x, (1, 0), 2`, `x, (1, 1), 3` |
| # |
| # where x is the t.data of the original tensor. Since input t has numel 4, the |
| # Jacobian should have 4 columns. So having a d_idx of 3 and idx of (1, 1) |
| # indicates that perturbing t[(1, 1)] allows us to updating the third (last) |
| # column of any jacobian corresponding to this particular input. |
| # |
| if x_tensor.is_sparse: |
| def get_stride(size): |
| dim = len(size) |
| tmp = 1 |
| stride = [0] * dim |
| for i in reversed(range(dim)): |
| stride[i] = tmp |
| tmp *= size[i] |
| return stride |
| |
| x_nnz = x_tensor._nnz() |
| x_size = list(x_tensor.size()) |
| x_indices = x_tensor._indices().t() |
| x_values = x_tensor._values() |
| x_stride = get_stride(x_size) |
| |
| # Use .data here to get around the version check |
| x_values = x_values.data |
| |
| for i in range(x_nnz): |
| x_value = x_values[i] |
| for x_idx in product(*[range(m) for m in x_values.size()[1:]]): |
| indices = x_indices[i].tolist() + list(x_idx) |
| d_idx = sum(indices[k] * x_stride[k] for k in range(len(x_size))) |
| yield x_value, x_idx, d_idx |
| elif x_tensor.layout == torch._mkldnn: # type: ignore[attr-defined] |
| for d_idx, x_idx in enumerate(product(*[range(m) for m in x_tensor.size()])): |
| # this is really inefficient, but without indexing implemented, there's |
| # not really a better way than converting back and forth |
| x_tensor_dense = x_tensor.to_dense() |
| yield x_tensor_dense, x_idx, d_idx |
| else: |
| # Use .data here to get around the version check |
| x_tensor = x_tensor.data |
| for d_idx, x_idx in enumerate(product(*[range(m) for m in x_tensor.size()])): |
| yield x_tensor, x_idx, d_idx |
| |
| |
| def _get_numerical_jacobian(fn, inputs, outputs=None, target=None, eps=1e-3, |
| grad_out=1.0) -> List[Tuple[torch.Tensor, ...]]: |
| """Computes the numerical jacobian for a given fn and inputs. Returns M * N jacobians |
| where M is the number of input tensors that require grad, and N is the number of output |
| float/complex tensors. |
| |
| Args: |
| fn: the function to compute the jacobian for |
| inputs: inputs to `fn` |
| outputs: provide precomputed outputs to avoid one extra invocation of fn |
| target: the Tensors wrt whom Jacobians are calculated (default=`inputs`) |
| eps: the magnitude of the perturbation during finite differencing (default=`1e-3`) |
| grad_out: grad output value used to calculate gradients. |
| |
| Returns: |
| A list of M N-tuples of tensors |
| |
| Note that `target` may not even be part of `input` to `fn`, so please be |
| **very careful** in this to not clone `target`. |
| """ |
| jacobians: List[Tuple[torch.Tensor, ...]] = [] |
| if outputs is None: |
| outputs = _as_tuple(fn(*_as_tuple(inputs))) |
| if target is None: |
| target = inputs |
| inp_indices = [i for i, a in enumerate(target) if is_tensor_like(a) and a.requires_grad] |
| for i, (inp, inp_idx) in enumerate(zip(iter_tensors(target, True), inp_indices)): |
| jacobians += [get_numerical_jacobian_wrt_specific_input(fn, inp_idx, inputs, outputs, eps, grad_out, input=inp)] |
| return jacobians |
| |
| |
| def get_numerical_jacobian(fn, inputs, target=None, eps=1e-3, grad_out=1.0): |
| """Deprecated api to compute numerical jacobian for a given fn and inputs. |
| Args: |
| fn: the function to compute the jacobian for (must take inputs as a tuple) |
| input: input to `fn` |
| target: the Tensors wrt whom Jacobians are calculated (default=`input`) |
| eps: the magnitude of the perturbation during finite differencing (default=`1e-3`) |
| grad_out: grad output value used to calculate gradients. |
| |
| Returns: |
| A list of jacobians wrt each input (or target) and the first output |
| |
| Note that `target` may not even be part of `input` to `fn`, so please be |
| **very careful** in this to not clone `target`. |
| """ |
| warnings.warn("get_numerical_jacobian was part of PyTorch's private API and not " |
| "meant to be exposed. We are deprecating it and it will be removed " |
| "in a future version of PyTorch. If you have a specific use for " |
| "this or feature request for this to be a stable API, please file " |
| "us an issue at https://github.com/pytorch/pytorch/issues/new") |
| |
| def fn_pack_inps(*inps): |
| return fn(inps) |
| jacobians = _get_numerical_jacobian(fn_pack_inps, inputs, None, target, eps, grad_out) |
| |
| return tuple(jacobian_for_each_output[0] for jacobian_for_each_output in jacobians) |
| |
| |
| def compute_numerical_gradient(fn, entry, v, norm_v, nbhd_checks_fn): |
| # Performs finite differencing by perturbing `entry` in-place by `v` and |
| # returns the gradient of each of the outputs wrt to x at idx. |
| orig = entry.clone() |
| entry.copy_(orig - v) |
| outa = fn() |
| entry.copy_(orig + v) |
| outb = fn() |
| entry.copy_(orig) |
| |
| def compute(a, b): |
| nbhd_checks_fn(a, b) |
| ret = (b - a) / (2 * norm_v) |
| return ret.detach().reshape(-1) |
| |
| return tuple(compute(a, b) for (a, b) in zip(outa, outb)) |
| |
| |
| def compute_numerical_jacobian_cols(jvp_fn, delta, input_is_complex, grad_out) -> List[torch.Tensor]: |
| # Computing the jacobian only works for pure real or pure imaginary delta |
| # for details on the algorithm used here, refer: |
| # Section 3.5.3 https://arxiv.org/pdf/1701.00392.pdf |
| # s = fn(z) where z = x for real valued input |
| # and z = x + yj for complex valued input |
| jacobians_cols: List[torch.Tensor] = [] |
| ds_dx_tup = jvp_fn(delta) |
| |
| if input_is_complex: # C -> C, C -> R |
| ds_dy_tup = jvp_fn(delta * 1j) |
| for ds_dx, ds_dy in zip(ds_dx_tup, ds_dy_tup): |
| if not ds_dx.is_complex() and isinstance(grad_out, complex): |
| # placeholder if grad_out is complex but output is not |
| jacobians_cols.append(torch.zeros_like(ds_dx)) |
| continue |
| # conjugate wirtinger derivative |
| conj_w_d = 0.5 * (ds_dx + ds_dy * 1j) |
| # wirtinger derivative |
| w_d = 0.5 * (ds_dx - ds_dy * 1j) |
| jacobians_cols.append(grad_out.conjugate() * conj_w_d + grad_out * w_d.conj()) |
| else: |
| for ds_dx in ds_dx_tup: |
| if ds_dx.is_complex(): # R -> C |
| # w_d = conj_w_d = 0.5 * ds_dx |
| # dL_dz_conj = 0.5 * [grad_out.conj() * ds_dx + grad_out * ds_dx.conj()] |
| # = 0.5 * [grad_out.conj() * ds_dx + (grad_out.conj() * ds_dx).conj()] |
| # = 0.5 * 2 * real(grad_out.conj() * ds_dx) |
| # = real(grad_out.conj() * ds_dx) |
| jacobians_cols.append(torch.real(grad_out.conjugate() * ds_dx)) |
| else: # R -> R |
| if isinstance(grad_out, complex): |
| # placeholder if grad_out is complex but output is not |
| jacobians_cols.append(torch.zeros_like(ds_dx)) |
| continue |
| jacobians_cols.append(ds_dx * grad_out) |
| return jacobians_cols |
| |
| |
| def combine_jacobian_cols(jacobians_cols: Dict[int, List[torch.Tensor]], outputs, input, |
| numel) -> Tuple[torch.Tensor, ...]: |
| # jacobian_cols is a data structure that maps column_idx -> output_idx -> column of jacobian Tensor |
| # we return a list that maps output_idx -> full jacobian Tensor |
| jacobians = allocate_jacobians_with_outputs(outputs, numel, input.dtype, input.device) |
| for i, jacobian in enumerate(jacobians): |
| for k, v in jacobians_cols.items(): |
| jacobian[k] = v[i] |
| return jacobians |
| |
| |
| def prepped_input(input: torch.Tensor, maybe_perturbed_input: Optional[torch.Tensor], |
| fast_mode=False) -> torch.Tensor: |
| # Prepares the inputs to be passed into the function while including the new modified input. |
| if input.layout == torch._mkldnn: # type: ignore[attr-defined] # no attr _mkldnn |
| # Convert back to mkldnn |
| if maybe_perturbed_input is not None: |
| return maybe_perturbed_input.to_mkldnn() |
| else: |
| return input |
| elif input.layout == torch.sparse_coo: |
| if fast_mode and maybe_perturbed_input is not None: |
| # entry is already a "cloned" version of the original tensor |
| # thus changes to entry are not reflected in the input |
| return maybe_perturbed_input |
| else: |
| return input |
| else: |
| # We cannot use entry (input.data) if we want gradgrad to work because |
| # fn (in the gradgrad case) needs to compute grad wrt input |
| return input |
| |
| |
| def check_outputs_same_dtype_and_shape(output1, output2, eps, idx=None) -> None: |
| # Check that the returned outputs don't have different dtype or shape when you |
| # perturb the input |
| on_index = "on index {idx} " if idx is not None else "" |
| assert output1.shape == output2.shape, \ |
| (f"Expected `func` to return outputs with the same shape" |
| f" when inputs are perturbed {on_index}by {eps}, but got:" |
| f" shapes {output1.shape} and {output2.shape}.") |
| assert output1.dtype == output2.dtype, \ |
| (f"Expected `func` to return outputs with the same dtype" |
| f" when inputs are perturbed {on_index}by {eps}, but got:" |
| f" dtypes {output1.dtype} and {output2.dtype}.") |
| |
| |
| def get_numerical_jacobian_wrt_specific_input(fn, input_idx, inputs, outputs, eps, |
| grad_out, input=None) -> Tuple[torch.Tensor, ...]: |
| # Computes the numerical jacobians wrt to a single input. Returns N jacobian |
| # tensors, where N is the number of outputs |
| # We use a dictionary because for sparse inputs, d_idx aren't necessarily consecutive |
| jacobian_cols: Dict[int, List[torch.Tensor]] = {} |
| input = inputs[input_idx] if input is None else input |
| assert input.requires_grad |
| for x, idx, d_idx in iter_tensor(input): |
| wrapped_fn = with_prepped_inputs(fn, inputs, input_idx, x) |
| input_to_perturb = x[idx] |
| nbhd_checks_fn = functools.partial(check_outputs_same_dtype_and_shape, idx=idx, eps=eps) |
| jvp_fn = get_jvp_fn(wrapped_fn, input_to_perturb, eps, nbhd_checks_fn) |
| jacobian_cols[d_idx] = compute_numerical_jacobian_cols(jvp_fn, eps, x.is_complex(), grad_out) |
| return combine_jacobian_cols(jacobian_cols, outputs, input, input.numel()) |
| |
| |
| def get_input_to_perturb(input): |
| if input.layout == torch._mkldnn: # type: ignore[attr-defined] # no attr _mkldnn |
| # Convert to dense so we can perform operations that require strided tensors |
| input_to_perturb = input.to_dense() |
| elif input.layout == torch.sparse_coo: |
| # Clone because input may require grad, and copy_ calls resize_, |
| # which is not allowed for .data |
| input_to_perturb = input.clone() |
| else: |
| input_to_perturb = input.data |
| return input_to_perturb |
| |
| |
| def with_prepped_inputs(fn, inputs, input_idx, input_to_perturb, fast_mode=False): |
| def wrapped_fn(): |
| inp = tuple(prepped_input(a, input_to_perturb if i == input_idx else None, fast_mode) if is_tensor_like(a) else a |
| for i, a in enumerate(_as_tuple(inputs))) |
| return tuple(a.clone() for a in _as_tuple(fn(*inp))) |
| return wrapped_fn |
| |
| |
| def get_jvp_fn(wrapped_fn, input_to_perturb, eps, nbhd_checks_fn): |
| def jvp_fn(delta): |
| return compute_numerical_gradient(wrapped_fn, input_to_perturb, delta, eps, nbhd_checks_fn) |
| return jvp_fn |
| |
| |
| def get_jvp_wrt_specific_input(fn, input_idx, inputs, outputs, u, eps, grad_out) -> List[torch.Tensor]: |
| # If fast_mode=False, iter_tensor handles the below cases: |
| # basically we want to prepare the input so that it can be modified in-place and do certain |
| # operations that require the tensor to have strides |
| input = inputs[input_idx] |
| input_to_perturb = get_input_to_perturb(input) |
| wrapped_fn = with_prepped_inputs(fn, inputs, input_idx, input_to_perturb, True) |
| nbhd_checks_fn = functools.partial(check_outputs_same_dtype_and_shape, eps=eps) |
| jvp_fn = get_jvp_fn(wrapped_fn, input_to_perturb, eps, nbhd_checks_fn) |
| if u.layout != torch.sparse_coo: |
| u = u.reshape(input_to_perturb.shape) |
| return compute_numerical_jacobian_cols(jvp_fn, u * eps, input.is_complex(), grad_out) |
| |
| |
| def check_jacobians_equal(j1, j2, atol): |
| # Check whether the max diff between two jacobians are within some tolerance `atol` |
| for j1_x, j2_x in zip(j1, j2): |
| if j1_x.numel() != 0 and (j1_x - j2_x).abs().max() > atol: |
| return False |
| return True |
| |
| |
| def stack_and_check_tensors(jacobians_rows, inputs, numel_outputs) -> Tuple[Tuple[torch.Tensor, ...], bool, bool]: |
| out_jacobians = allocate_jacobians_with_inputs(inputs, numel_outputs) |
| diff_input_list = list(iter_tensors(inputs, True)) |
| correct_grad_sizes = True |
| correct_grad_types = True |
| for i, rows in enumerate(jacobians_rows): |
| inp = diff_input_list[i] |
| out_jacobian = out_jacobians[i] |
| for j, row in enumerate(rows): |
| if row is not None and row.size() != inp.size(): |
| correct_grad_sizes = False |
| elif row is not None and row.dtype != inp.dtype: |
| correct_grad_types = False |
| if row is None: |
| out_jacobian[:, j].zero_() |
| else: |
| row_dense = row.to_dense() if not row.layout == torch.strided else row |
| assert out_jacobian[:, j].numel() == row_dense.numel() |
| out_jacobian[:, j] = row_dense.reshape(-1) |
| return out_jacobians, correct_grad_sizes, correct_grad_types |
| |
| |
| FAILED_NONDET_MSG = """ |
| |
| NOTE: If your op relies on non-deterministic operations i.e., it is listed here: |
| https://pytorch.org/docs/stable/generated/torch.use_deterministic_algorithms.html |
| this failure might be expected. |
| |
| If you are adding a new operator, please file an issue and then use one of the |
| workarounds. The workaround depends on how your test invokes gradcheck/gradgradcheck. |
| If the test |
| - manually invokes gradcheck/gradgradcheck, then call gradcheck/gradgradcheck |
| with `nondet_tol=<tol>` as a keyword argument. |
| - is OpInfo-based (e.g., in test_ops.py), then modify the OpInfo for the test |
| to have `gradcheck_nondet_tol=<tol>`. |
| - is a Module test (e.g., in common_nn.py), then modify the corresponding |
| module_test entry to have `gradcheck_nondet_tol=<tol>` |
| """ |
| |
| |
| def check_analytical_jacobian_attributes(inputs, output, nondet_tol, grad_out_scale, check_grad_dtypes, |
| fast_mode=False, v=None) -> Tuple[torch.Tensor, ...]: |
| diff_input_list = list(iter_tensors(inputs, True)) |
| |
| def vjp_fn(grad_output): |
| return torch.autograd.grad(output, diff_input_list, grad_output, |
| retain_graph=True, allow_unused=True) |
| if fast_mode: |
| # vjp can be seen as a linear combination of the jacobians rows, we still call into stack_and_check |
| # because we'd like to reuse the checks for dtype and shape |
| jacobians_rows = get_vjp_wrt_specific_output(vjp_fn, output.clone(), v, grad_out_scale) |
| jacobians_rows_reentrant = get_vjp_wrt_specific_output(vjp_fn, output.clone(), v, grad_out_scale) |
| else: |
| jacobians_rows = compute_analytical_jacobian_rows(vjp_fn, output.clone(), grad_out_scale) |
| jacobians_rows_reentrant = compute_analytical_jacobian_rows(vjp_fn, output.clone(), grad_out_scale) |
| output_numel = output.numel() if not fast_mode else 1 |
| |
| jacobians, correct_grad_types, correct_grad_sizes = stack_and_check_tensors(jacobians_rows, inputs, output_numel) |
| jacobians_reentrant, _, _ = stack_and_check_tensors(jacobians_rows_reentrant, inputs, output_numel) |
| |
| reentrant = check_jacobians_equal(jacobians, jacobians_reentrant, nondet_tol) |
| |
| complex_str = '(calculated using complex valued grad output) ' \ |
| if isinstance(grad_out_scale, complex) else '' |
| |
| if not correct_grad_types and check_grad_dtypes: |
| raise GradcheckError(f'Gradient{complex_str} has dtype mismatch') |
| if not correct_grad_sizes: |
| raise GradcheckError(f'Analytical gradient{complex_str} has incorrect size') |
| if not reentrant: |
| raise GradcheckError(f'Backward{complex_str} is not reentrant, i.e., running backward with ' |
| 'same input and grad_output multiple times gives different values, ' |
| 'although analytical gradient matches numerical gradient.' |
| f'The tolerance for nondeterminism was {nondet_tol}.' + FAILED_NONDET_MSG) |
| return jacobians |
| |
| |
| def get_analytical_jacobian(inputs, output, nondet_tol=0.0, grad_out=1.0): |
| # Replicates the behavior of the old get_analytical_jacobian before the refactor |
| warnings.warn("get_analytical_jacobian was part of PyTorch's private API and not " |
| "meant to be exposed. We are deprecating it and it will be removed " |
| "in a future version of PyTorch. If you have a specific use for " |
| "this or feature request for this to be a stable API, please file " |
| "us an issue at https://github.com/pytorch/pytorch/issues/new") |
| |
| diff_input_list = list(iter_tensors(inputs, True)) |
| |
| def backward_fn(grad_output): |
| return torch.autograd.grad(output, diff_input_list, grad_output, |
| retain_graph=True, allow_unused=True) |
| |
| jacobians_rows = compute_analytical_jacobian_rows(backward_fn, output.clone(), grad_out) |
| jacobians_rows_reentrant = compute_analytical_jacobian_rows(backward_fn, output.clone(), grad_out) |
| |
| output_numel = output.numel() |
| jacobians, correct_grad_types, correct_grad_sizes = stack_and_check_tensors(jacobians_rows, inputs, output_numel) |
| jacobians_reentrant, _, _ = stack_and_check_tensors(jacobians_rows_reentrant, inputs, output_numel) |
| reentrant = check_jacobians_equal(jacobians, jacobians_reentrant, nondet_tol) |
| |
| return jacobians, reentrant, correct_grad_sizes, correct_grad_types |
| |
| |
| def _get_analytical_jacobian(inputs, outputs, input_idx, output_idx): |
| # Computes the analytical jacobian in slow mode for a single input_idx, output_idx pair |
| # without performing checks for dtype, shape, and reentrancy |
| jacobians = check_analytical_jacobian_attributes(inputs, outputs[output_idx], |
| nondet_tol=float('inf'), grad_out_scale=1.0, check_grad_dtypes=False) |
| return jacobians[input_idx] |
| |
| |
| def compute_analytical_jacobian_rows(vjp_fn, sample_output, grad_out_scale) -> List[List[Optional[torch.Tensor]]]: |
| # Computes Jacobian row-by-row using backward function `vjp_fn` = v^T J |
| # NB: this function does not assume vjp_fn(v) to return tensors with |
| # the same number of elements for different v. This is checked when we |
| # later combine the rows into a single tensor. |
| grad_out_base = torch.zeros_like(sample_output, memory_format=torch.legacy_contiguous_format) |
| flat_grad_out = grad_out_base.view(-1) |
| # jacobians_rows[i][j] represents the jth row of the ith input |
| jacobians_rows: List[List[Optional[torch.Tensor]]] = [] |
| |
| for j in range(flat_grad_out.numel()): |
| flat_grad_out.zero_() |
| flat_grad_out[j] = grad_out_scale |
| grad_inputs = vjp_fn(grad_out_base) |
| for i, d_x in enumerate(grad_inputs): |
| if j == 0: |
| jacobians_rows.append([]) |
| jacobians_rows[i] += [d_x.clone() if isinstance(d_x, torch.Tensor) else None] |
| return jacobians_rows |
| |
| |
| def get_vjp_wrt_specific_output(vjp_fn, sample_output, v, grad_out_scale) -> List[List[Optional[torch.Tensor]]]: |
| jacobians_rows: List[List[Optional[torch.Tensor]]] = [] |
| grad_inputs = vjp_fn(v.reshape(sample_output.shape) * grad_out_scale) |
| for i, d_x in enumerate(grad_inputs): |
| jacobians_rows.append([d_x.clone() if isinstance(d_x, torch.Tensor) else None]) |
| return jacobians_rows |
| |
| |
| def check_inputs(tupled_inputs, check_sparse_nnz) -> bool: |
| if not check_sparse_nnz and any(t.is_sparse for t in tupled_inputs if isinstance(t, torch.Tensor)): |
| raise GradcheckError('gradcheck expects all tensor inputs are dense when check_sparse_nnz is set to False.') |
| # Make sure that gradients are saved for at least one input |
| any_input_requiring_grad = False |
| for idx, inp in enumerate(tupled_inputs): |
| if is_tensor_like(inp) and inp.requires_grad: |
| if not (inp.dtype == torch.float64 or inp.dtype == torch.complex128): |
| warnings.warn( |
| f'Input #{idx} requires gradient and ' |
| 'is not a double precision floating point or complex. ' |
| 'This check will likely fail if all the inputs are ' |
| 'not of double precision floating point or complex. ') |
| content = inp._values() if inp.is_sparse else inp |
| # TODO: To cover more problematic cases, replace stride = 0 check with |
| # "any overlap in memory" once we have a proper function to check it. |
| if content.layout is not torch._mkldnn: # type: ignore[attr-defined] |
| if not all(st > 0 or sz <= 1 for st, sz in zip(content.stride(), content.size())): |
| raise RuntimeError( |
| f'The {idx}th input has a dimension with stride 0. gradcheck only ' |
| 'supports inputs that are non-overlapping to be able to ' |
| 'compute the numerical gradients correctly. You should call ' |
| '.contiguous on the input before passing it to gradcheck.') |
| any_input_requiring_grad = True |
| inp.retain_grad() |
| if not any_input_requiring_grad: |
| raise ValueError( |
| 'gradcheck expects at least one input tensor to require gradient, ' |
| 'but none of the them have requires_grad=True.') |
| return True |
| |
| |
| def check_outputs(outputs) -> None: |
| if any(t.layout == torch.sparse_coo for t in outputs if isinstance(t, torch.Tensor)): |
| # it is easier to call to_dense() on the sparse output than |
| # to modify analytical jacobian |
| raise ValueError('Sparse output is not supported at gradcheck yet. ' |
| 'Please call to_dense() on the output of fn for gradcheck.') |
| if any(t.layout == torch._mkldnn for t in outputs if isinstance(t, torch.Tensor)): # type: ignore[attr-defined] |
| raise ValueError('MKLDNN output is not supported at gradcheck yet. ' |
| 'Please call to_dense() on the output of fn for gradcheck.') |
| |
| |
| def check_no_differentiable_outputs(func, inputs, func_out, eps) -> bool: |
| # When there are no differentiable outputs, numerical gradient for a function is |
| # expected to be zero. |
| jacobians_all_inputs_outputs = _get_numerical_jacobian(func, inputs, func_out, eps=eps) |
| for jacobians_all_outputs_and_fixed_input in jacobians_all_inputs_outputs: |
| for jacobian in jacobians_all_outputs_and_fixed_input: |
| if torch.ne(jacobian, 0).sum() > 0: |
| raise GradcheckError('Numerical gradient for function expected to be zero') |
| return True |
| |
| |
| def check_no_differentiable_outputs_fast(func, func_out, all_inputs, inputs_indices, all_u, eps, nondet_tol): |
| for inp_idx, u in zip(inputs_indices, all_u): |
| numerical_jacobians = get_jvp_wrt_specific_input(func, inp_idx, all_inputs, _as_tuple(func_out), u, eps, 1.0) |
| for jacobian in numerical_jacobians: |
| if jacobian.numel() == 0: |
| continue |
| if (jacobian - torch.zeros_like(jacobian)).abs().max() > nondet_tol: |
| raise GradcheckError('Numerical gradient for function expected to be zero') |
| return True |
| |
| |
| FAILED_BATCHED_GRAD_MSG = """ |
| gradcheck or gradgradcheck failed while testing batched gradient computation. |
| This could have been invoked in a number of ways (via a test that calls |
| gradcheck/gradgradcheck directly or via an autogenerated test). |
| |
| If you are adding a new operator, please file an issue and then use one of the |
| workarounds. The workaround depends on how your test invokes gradcheck/gradgradcheck. |
| If the test |
| - manually invokes gradcheck/gradgradcheck, then call gradcheck/gradgradcheck |
| with `check_batched_grad=False` as a keyword argument. |
| - is OpInfo-based (e.g., in test_ops.py), then modify the OpInfo for the test |
| to have `check_batched_grad=False` and/or `check_batched_gradgrad=False`. |
| - is common_method_invocations-based, then add your test to the denylist |
| EXCLUDE_BATCHED_GRAD_TESTS in test_autograd.py |
| |
| If you're modifying an existing operator that supports batched grad computation, |
| or wish to make a new operator work with batched grad computation, please read |
| the following. |
| |
| To compute batched grads (e.g., jacobians, hessians), we vmap over the backward |
| computation. The most common failure case is if there is a 'vmap-incompatible |
| operation' in the backward pass. Please see |
| NOTE: [How to write vmap-compatible backward formulas] |
| in the codebase for an explanation of how to fix this. |
| """.strip() |
| |
| def get_failed_batched_grad_test_msg(output_idx, input_idx, res, exp): |
| return f""" |
| For output {output_idx} and input {input_idx}: |
| |
| {FAILED_BATCHED_GRAD_MSG} |
| |
| Got: |
| {res} |
| |
| Expected: |
| {exp} |
| """.strip() |
| |
| |
| def test_batched_grad(input, output, output_idx) -> bool: |
| # NB: test_batched_grad compares two autograd.grad invocations with a single |
| # vmap(autograd.grad) invocation. It's not exactly a "gradcheck" in the |
| # sense that we're not comparing an analytical jacobian with a numeric one, |
| # but it is morally similar (we could have computed a full analytic jac |
| # via vmap, but that is potentially slow) |
| diff_input_list = list(iter_tensors(input, True)) |
| grad = functools.partial(torch.autograd.grad, output, diff_input_list, retain_graph=True, allow_unused=True) |
| |
| def vjp(v): |
| results = grad(v) |
| results = tuple(grad if grad is not None else |
| torch.zeros([], dtype=inp.dtype, device=inp.device).expand(inp.shape) |
| for grad, inp in zip(results, diff_input_list)) |
| return results |
| |
| grad_outputs = [torch.randn_like(output) for _ in range(2)] |
| |
| expected = [vjp(gO) for gO in grad_outputs] |
| expected = [torch.stack(shards) for shards in zip(*expected)] |
| |
| # Squash warnings since these are expected to happen in most cases |
| # NB: this doesn't work for CUDA tests: https://github.com/pytorch/pytorch/issues/50209 |
| with warnings.catch_warnings(): |
| warnings.filterwarnings("ignore", message="Batching rule not implemented") |
| warnings.filterwarnings("ignore", message="torch.vmap is an experimental prototype") |
| try: |
| result = vmap(vjp)(torch.stack(grad_outputs)) |
| except RuntimeError as ex: |
| # It's OK that we're not raising the error at the correct callsite. |
| # That's because the callsite is always going to inside the Python |
| # autograd.grad instead of the C++ traceback of what line in the |
| # backward formula |
| raise GradcheckError( |
| f'While computing batched gradients, got: {ex}\n\n{FAILED_BATCHED_GRAD_MSG}') |
| |
| for input_idx, (res, exp) in enumerate(zip(result, expected)): |
| if torch.allclose(res, exp): |
| continue |
| raise GradcheckError(get_failed_batched_grad_test_msg(output_idx, input_idx, res, exp)) |
| return True |
| |
| |
| def test_backward_mul_by_grad_output(outputs, inputs, check_sparse_nnz) -> bool: |
| # Tests that backward is multiplied by grad_output |
| diff_input_list: List[torch.Tensor] = list(iter_tensors(inputs, True)) |
| if not diff_input_list: |
| raise GradcheckError("no Tensors requiring grad found in input") |
| grads_input = torch.autograd.grad(outputs, diff_input_list, |
| [torch.zeros_like(o, memory_format=torch.legacy_contiguous_format) for o in outputs], |
| allow_unused=True) |
| for gi, di in zip(grads_input, diff_input_list): |
| if gi is None: |
| continue |
| if isinstance(gi, torch.Tensor) and gi.layout != torch.strided: |
| if gi.layout != di.layout: |
| raise GradcheckError('grad is incorrect layout (' + str(gi.layout) + ' is not ' + str(di.layout) + ')') |
| if gi.layout == torch.sparse_coo: |
| if gi.sparse_dim() != di.sparse_dim(): |
| raise GradcheckError('grad is sparse tensor, but has incorrect sparse_dim') |
| if gi.dense_dim() != di.dense_dim(): |
| raise GradcheckError('grad is sparse tensor, but has incorrect dense_dim') |
| gi = gi.to_dense() |
| di = di.to_dense() |
| |
| if check_sparse_nnz: |
| if not torch.allclose(gi, torch.zeros_like(gi)): |
| raise GradcheckError('backward not multiplied by grad_output') |
| elif not gi.eq(0).all(): |
| raise GradcheckError('backward not multiplied by grad_output') |
| if gi.dtype != di.dtype or gi.device != di.device or gi.is_sparse != di.is_sparse: |
| raise GradcheckError("grad is incorrect type") |
| if gi.size() != di.size(): |
| raise GradcheckError('grad is incorrect size') |
| return True |
| |
| |
| def test_undefined_grad(func, outputs, inputs) -> bool: |
| diff_input_list: List[torch.Tensor] = list(iter_tensors(inputs, True)) |
| if not diff_input_list: |
| raise GradcheckError("no Tensors requiring grad found in input") |
| |
| def warn_bc_breaking(): |
| warnings.warn(( |
| 'Backwards compatibility: New undefined gradient support checking ' |
| 'feature is enabled by default, but it may break existing callers ' |
| 'of this function. If this is true for you, you can call this ' |
| 'function with "check_undefined_grad=False" to disable the feature')) |
| |
| def check_undefined_grad_support(output_to_check): |
| grads_output = [torch.zeros_like(o, memory_format=torch.legacy_contiguous_format) for o in output_to_check] |
| try: |
| grads_input = torch.autograd.grad(output_to_check, diff_input_list, |
| grads_output, allow_unused=True) |
| except RuntimeError: |
| warn_bc_breaking() |
| raise GradcheckError(( |
| 'Expected backward function to handle undefined output grads. ' |
| 'Please look at "Notes about undefined output gradients" in ' |
| '"tools/autograd/derivatives.yaml"')) |
| |
| for gi, i in zip(grads_input, diff_input_list): |
| if (gi is not None) and (not gi.eq(0).all()): |
| warn_bc_breaking() |
| raise GradcheckError(( |
| 'Expected all input grads to be undefined or zero when all output grads are undefined ' |
| 'or zero. Please look at "Notes about undefined output gradients" in ' |
| '"tools/autograd/derivatives.yaml"')) |
| return True |
| |
| # All backward functions must work properly if all output grads are undefined |
| outputs_to_check = [[ |
| torch._C._functions.UndefinedGrad()(o) for o in _differentiable_outputs(func(*inputs)) |
| # This check filters out Tensor-likes that aren't instances of Tensor. |
| if isinstance(o, torch.Tensor) |
| ]] |
| |
| # If there are multiple output grads, we should be able to undef one at a time without error |
| if len(outputs_to_check[0]) > 1: |
| for undef_grad_idx in range(len(outputs)): |
| output_to_check = _differentiable_outputs(func(*inputs)) |
| outputs_to_check.append([ |
| torch._C._functions.UndefinedGrad()(o) if idx == undef_grad_idx else o |
| for idx, o in enumerate(output_to_check)]) |
| |
| return all(check_undefined_grad_support(output) for output in outputs_to_check) |
| |
| |
| def _as_tuple(x): |
| if isinstance(x, tuple): |
| return x |
| elif isinstance(x, list): |
| return tuple(x) |
| else: |
| return x, |
| |
| |
| def _differentiable_outputs(x): |
| return tuple(o for o in _as_tuple(x) if o.requires_grad) |
| |
| |
| def get_notallclose_msg(analytical, numerical, output_idx, input_idx, error_str='') -> str: |
| return error_str + 'Jacobian mismatch for output %d with respect to input %d,\n' \ |
| 'numerical:%s\nanalytical:%s\n' % (output_idx, input_idx, numerical, analytical) |
| |
| def transpose(matrix_of_tensors): |
| # returns list of tuples |
| return list(zip(*matrix_of_tensors)) |
| |
| |
| def slow_gradcheck(func, func_out, tupled_inputs, outputs, eps, rtol, |
| atol, check_grad_dtypes, nondet_tol): |
| if not outputs: |
| return check_no_differentiable_outputs(func, tupled_inputs, _as_tuple(func_out), eps) |
| |
| numerical = transpose(_get_numerical_jacobian(func, tupled_inputs, outputs, eps=eps)) |
| if any(isinstance(o, torch.Tensor) and o.is_complex() for o in _as_tuple(func_out)): |
| numerical_from_imag_grad_out = transpose(_get_numerical_jacobian(func, tupled_inputs, outputs, eps=eps, grad_out=1j)) |
| |
| for i, o in enumerate(outputs): |
| analytical = check_analytical_jacobian_attributes(tupled_inputs, o, nondet_tol, 1.0, check_grad_dtypes) |
| |
| if o.is_complex(): |
| analytical_from_imag_grad_out = check_analytical_jacobian_attributes( |
| tupled_inputs, o, nondet_tol, 1j, check_grad_dtypes) |
| |
| inp_tensors = iter_tensors(tupled_inputs, True) |
| |
| for j, (a, n, inp) in enumerate(zip(analytical, numerical[i], inp_tensors)): |
| if a.numel() != 0 or n.numel() != 0: |
| if o.is_complex(): # C -> C, R -> C |
| if not torch.allclose(analytical_from_imag_grad_out[j], numerical_from_imag_grad_out[i][j], rtol, atol): |
| raise GradcheckError(get_notallclose_msg(analytical_from_imag_grad_out[j], |
| numerical_from_imag_grad_out[i][j], i, j, |
| "Gradients failed to compare equal for grad output = 1j. ")) |
| if inp.is_complex(): # C -> R, C -> C |
| if not torch.allclose(a, n, rtol, atol): |
| raise GradcheckError(get_notallclose_msg(a, n, i, j, |
| "Gradients failed to compare equal for grad output = 1. ")) |
| else: # R -> R, R -> C |
| if not torch.allclose(a, n, rtol, atol): |
| raise GradcheckError(get_notallclose_msg(a, n, i, j)) |
| return True |
| |
| |
| def dot_with_type_promotion(u, v): |
| assert u.dim() == 1 and v.dim() == 1 |
| return (u * v).sum() |
| |
| |
| def allclose_with_type_promotion(a, b, rtol, atol): |
| promoted_type = torch.promote_types(a.dtype, b.dtype) |
| a = a.to(dtype=promoted_type) |
| b = b.to(dtype=promoted_type) |
| return torch.allclose(a, b, rtol, atol) |
| |
| |
| def vec_from_tensor(x, generator): |
| # Create a random vector with the same number of elements as x and the same dtype/device |
| # If x is complex, we create a complex tensor with only real component |
| if x.layout == torch.sparse_coo: |
| # For sparse, create a random sparse vec with random values in the same |
| # indices. Make sure size is set so that it isn't inferred to be smaller. |
| x_values = x._values() |
| values = torch.rand(x_values.numel(), generator=generator) \ |
| .to(dtype=x.dtype, device=x.device) \ |
| .reshape(x_values.shape) |
| values /= values.norm() |
| vec = torch.sparse_coo_tensor(x._indices(), values, x.size()) |
| else: |
| vec = torch.rand(x.numel(), generator=generator).to(dtype=x.dtype, device=x.device) |
| vec /= vec.norm() |
| return vec |
| |
| |
| def adjusted_atol(atol, u, v): |
| # In slow gradcheck, we compare A and B element-wise, i.e., for some a, b we allow |a - b| < atol + rtol * b |
| # but since we now compare q1 = v^T A u and q2 = v^T B u, we must allow |q1 - q2| < v^T E u + rtol * v^T B u |
| # where E is the correctly sized matrix where each entry is atol |
| # |
| # We see that atol needs to be scaled by v^T M u (where M is an all-ones M x N matrix): |
| # v^T M u = \sum_{i} \sum_{j} u_i * v_j = (\sum_{i} u_i)(\sum_{i} v_i) |
| sum_u = torch.sparse.sum(u) if u.layout == torch.sparse_coo else u.sum() |
| sum_v = torch.sparse.sum(v) if v.layout == torch.sparse_coo else v.sum() |
| return atol * sum_u.item() * sum_v.item() |
| |
| |
| FAST_FAIL_SLOW_OK_MSG = """ |
| Fast gradcheck failed but element-wise differences are small. This means that the |
| test might've passed in slow_mode! |
| |
| If you are adding a new operator, please file an issue and then use one of the |
| workarounds. The workaround depends on how your test invokes gradcheck/gradgradcheck: |
| |
| If the test |
| - manually invokes gradcheck/gradgradcheck, then call gradcheck/gradgradcheck |
| with `fast_mode=False` as a keyword argument. |
| - is OpInfo-based (e.g., in test_ops.py), then modify the OpInfo for the test |
| to have `gradcheck_fast_mode=False` |
| - is a Module test (e.g., in common_nn.py), then modify the corresponding |
| module_test entry to have `gradcheck_fast_mode=False` |
| """.strip() |
| |
| |
| def run_slow_mode_and_get_error(func, tupled_inputs, outputs, input_idx, output_idx, rtol, atol): |
| # Compute jacobians in slow mode for better error message |
| slow_numerical = _get_numerical_jacobian(func, tupled_inputs, outputs)[input_idx][output_idx] |
| slow_analytical = _get_analytical_jacobian(tupled_inputs, outputs, input_idx, output_idx) |
| |
| # Assume jacobians are non-empty and have the same shape |
| slow_max_diff = (slow_numerical - slow_analytical).abs().max() |
| |
| slow_allclose = torch.allclose(slow_analytical, slow_numerical, rtol, atol) |
| msg = ("\nThe above quantities relating the numerical and analytical jacobians are computed \n" |
| "in fast mode. See: https://github.com/pytorch/pytorch/issues/53876 for more background \n" |
| "about fast mode. Below, we recompute numerical and analytical jacobians in slow mode:\n\n" |
| f"Numerical:\n {slow_numerical}\n" |
| f"Analytical:\n{slow_analytical}\n\n" |
| f"The max per-element difference (slow mode) is: {slow_max_diff}.\n") |
| if slow_allclose: |
| # Slow gradcheck would've passed! |
| msg += FAST_FAIL_SLOW_OK_MSG |
| return msg |
| |
| |
| def fast_gradcheck(func, func_out, tupled_inputs, outputs, eps, rtol, |
| atol, check_grad_dtypes, nondet_tol): |
| # Perform the fast version of gradcheck |
| # See https://github.com/pytorch/pytorch/issues/53876 for details |
| inp_tensors = [t for t in tupled_inputs if is_tensor_like(t) and t.requires_grad] |
| inp_tensor_indices = [i for i, t in enumerate(tupled_inputs) if is_tensor_like(t) and t.requires_grad] |
| |
| # Use our own generator to avoid messing with the user's RNG state |
| g_cpu = torch.Generator() |
| all_u = [vec_from_tensor(inp, g_cpu) for inp in inp_tensors] |
| all_v = [vec_from_tensor(out, g_cpu) for out in outputs] |
| |
| if not outputs: |
| check_no_differentiable_outputs_fast(func, func_out, tupled_inputs, inp_tensor_indices, all_u, eps, nondet_tol) |
| |
| any_complex = any(o.is_complex() for o in outputs) |
| complex_output_indices = [i for i, o in enumerate(outputs) if o.is_complex()] |
| |
| # Initialize list of lists to store jacobians for each input, output pair |
| all_analytical: List[List[torch.Tensor]] = [[] for _ in outputs] |
| all_numerical: List[List[torch.Tensor]] = [[] for _ in inp_tensors] |
| all_analytical_from_imag_grad_out: List[List[torch.Tensor]] = [[] for _ in complex_output_indices] |
| all_numerical_from_imag_grad_out: List[List[torch.Tensor]] = [[] for _ in inp_tensors] |
| |
| # Numerically approximate v^T (J u) |
| for i, (input_idx, u) in enumerate(zip(inp_tensor_indices, all_u)): |
| numerical = get_jvp_wrt_specific_input(func, input_idx, tupled_inputs, outputs, u, eps, 1.0) |
| for j, (a, v) in enumerate(zip(numerical, all_v)): |
| all_numerical[i].append(dot_with_type_promotion(a, v)) |
| |
| if any_complex: |
| numerical_from_imag_grad_out = get_jvp_wrt_specific_input( |
| func, input_idx, tupled_inputs, outputs, u, eps, 1j) |
| for j in complex_output_indices: |
| a, v = numerical_from_imag_grad_out[j], all_v[j] |
| all_numerical_from_imag_grad_out[i].append(dot_with_type_promotion(a, v)) |
| |
| # Analytically calculate (v^T J) u |
| all_u_dense = [u.to_dense().reshape(-1) if u.layout == torch.sparse_coo else u for u in all_u] |
| for i, (out, v) in enumerate(zip(outputs, all_v)): |
| analytical = check_analytical_jacobian_attributes(tupled_inputs, out, nondet_tol, 1.0, check_grad_dtypes, |
| fast_mode=True, v=v) |
| for a, u in zip(analytical, all_u_dense): |
| all_analytical[i].append(a.T.squeeze(0).dot(u)) |
| |
| if out.is_complex(): |
| analytical_from_imag_grad_out = check_analytical_jacobian_attributes( |
| tupled_inputs, out, nondet_tol, 1j, check_grad_dtypes, fast_mode=True, v=v) |
| |
| for j, (a, u) in enumerate(zip(analytical_from_imag_grad_out, all_u_dense)): |
| all_analytical_from_imag_grad_out[i].append(a.T.squeeze(0).dot(u)) |
| |
| prefix = "Gradients failed to compare equal for grad output = 1j (in fast mode). " |
| # Make sure analytical and numerical is same when calcaluted using grad_out = 1j |
| for i, all_numerical_for_input_i in enumerate(all_numerical_from_imag_grad_out): |
| for j, n in enumerate(all_numerical_for_input_i): |
| a = all_analytical_from_imag_grad_out[j][i] |
| n = n.to(device=a.device) |
| if not allclose_with_type_promotion(a, n, rtol, adjusted_atol(atol, all_u[i], all_v[j])): |
| jacobians_str = run_slow_mode_and_get_error(func, tupled_inputs, outputs, i, j, rtol, atol) |
| raise GradcheckError(get_notallclose_msg(a, n, j, i, prefix) + jacobians_str) |
| |
| # Make sure analytical and numerical is the same |
| for i, (all_numerical_for_input_i, inp) in enumerate(zip(all_numerical, inp_tensors)): |
| prefix = "" if not inp.is_complex() else \ |
| "Gradients failed to compare equal for grad output = 1 (in fast mode). " |
| for j, n in enumerate(all_numerical_for_input_i): |
| a = all_analytical[j][i] |
| n = n.to(device=a.device) |
| if not allclose_with_type_promotion(a, n, rtol, adjusted_atol(atol, all_u[i], all_v[j])): |
| jacobians_str = run_slow_mode_and_get_error(func, tupled_inputs, outputs, i, j, rtol, atol) |
| raise GradcheckError(get_notallclose_msg(a, n, j, i, prefix) + jacobians_str) |
| |
| return True |
| |
| |
| def has_complex_inputs_or_outputs(tupled_inputs, func_out): |
| return any(is_tensor_like(o) and o.is_complex() for o in _as_tuple(func_out)) or \ |
| any(is_tensor_like(i) and i.is_complex() for i in tupled_inputs) |
| |
| |
| # Note [VarArg of Tensors] |
| # ~~~~~~~~~~~~~~~~~~~~~~~~ |
| # 'func' accepts a vararg of tensors, which isn't expressable in the type system at the moment. |
| # If https://mypy.readthedocs.io/en/latest/additional_features.html?highlight=callable#extended-callable-types is accepted, |
| # the '...' first argument of Callable can be replaced with VarArg(Tensor). |
| # For now, we permit any input. |
| # the '...' first argument of Callable can be replaced with VarArg(Tensor). |
| # For now, we permit any input. |
| def gradcheck( |
| func: Callable[..., Union[_TensorOrTensors]], # See Note [VarArg of Tensors] |
| inputs: _TensorOrTensors, |
| eps: float = 1e-6, |
| atol: float = 1e-5, |
| rtol: float = 1e-3, |
| raise_exception: bool = True, |
| check_sparse_nnz: bool = False, |
| nondet_tol: float = 0.0, |
| check_undefined_grad: bool = True, |
| check_grad_dtypes: bool = False, |
| check_batched_grad: bool = False, |
| fast_mode: bool = False, |
| ) -> bool: |
| r"""Check gradients computed via small finite differences against analytical |
| gradients w.r.t. tensors in :attr:`inputs` that are of floating point or complex type |
| and with ``requires_grad=True``. |
| |
| The check between numerical and analytical gradients uses :func:`~torch.allclose`. |
| |
| For complex functions, no notion of Jacobian exists. Gradcheck verifies if the numerical and |
| analytical values of Wirtinger and Conjugate Wirtinger derivative are consistent. The gradient |
| computation is done under the assumption that the overall function has a real valued output. |
| For functions with complex output, gradcheck compares the numerical and analytical gradients |
| for two values of :attr:`grad_output`: 1 and 1j. For more details, check out |
| :ref:`complex_autograd-doc`. |
| |
| .. note:: |
| The default values are designed for :attr:`input` of double precision. |
| This check will likely fail if :attr:`input` is of less precision, e.g., |
| ``FloatTensor``. |
| |
| .. warning:: |
| If any checked tensor in :attr:`input` has overlapping memory, i.e., |
| different indices pointing to the same memory address (e.g., from |
| :func:`torch.expand`), this check will likely fail because the numerical |
| gradients computed by point perturbation at such indices will change |
| values at all other indices that share the same memory address. |
| |
| Args: |
| func (function): a Python function that takes Tensor inputs and returns |
| a Tensor or a tuple of Tensors |
| inputs (tuple of Tensor or Tensor): inputs to the function |
| eps (float, optional): perturbation for finite differences |
| atol (float, optional): absolute tolerance |
| rtol (float, optional): relative tolerance |
| raise_exception (bool, optional): indicating whether to raise an exception if |
| the check fails. The exception gives more information about the |
| exact nature of the failure. This is helpful when debugging gradchecks. |
| check_sparse_nnz (bool, optional): if True, gradcheck allows for SparseTensor input, |
| and for any SparseTensor at input, gradcheck will perform check at nnz positions only. |
| nondet_tol (float, optional): tolerance for non-determinism. When running |
| identical inputs through the differentiation, the results must either match |
| exactly (default, 0.0) or be within this tolerance. |
| check_undefined_grad (bool, optional): if True, check if undefined output grads |
| are supported and treated as zeros, for ``Tensor`` outputs. |
| check_batched_grad (bool, optional): if True, check if we can compute |
| batched gradients using prototype vmap support. Defaults to False. |
| fast_mode (bool, optional): Fast mode for gradcheck and gradgradcheck is currently only |
| implemented for R to R functions. If none of the inputs and outputs are complex |
| a faster implementation of gradcheck that no longer computes the entire jacobian |
| is run; otherwise, we fall back to the slow implementation. |
| |
| Returns: |
| True if all differences satisfy allclose condition |
| """ |
| # This is just a wrapper that handles the raise_exception logic |
| args = locals().copy() |
| args.pop("raise_exception") |
| if not raise_exception: |
| try: |
| return gradcheck_helper(**args) |
| except GradcheckError: |
| return False |
| else: |
| return gradcheck_helper(**args) |
| |
| |
| def gradcheck_helper(func, inputs, eps, atol, rtol, check_sparse_nnz, nondet_tol, check_undefined_grad, |
| check_grad_dtypes, check_batched_grad, fast_mode): |
| tupled_inputs = _as_tuple(inputs) |
| check_inputs(tupled_inputs, check_sparse_nnz) |
| func_out = func(*tupled_inputs) |
| |
| if has_complex_inputs_or_outputs(tupled_inputs, func_out): |
| fast_mode = False |
| |
| outputs = _differentiable_outputs(func_out) |
| check_outputs(outputs) |
| |
| if fast_mode: |
| fast_gradcheck(func, func_out, tupled_inputs, outputs, eps, rtol, |
| atol, check_grad_dtypes, nondet_tol) |
| else: |
| slow_gradcheck(func, func_out, tupled_inputs, outputs, eps, rtol, |
| atol, check_grad_dtypes, nondet_tol) |
| |
| for i, o in enumerate(outputs): |
| if check_batched_grad: |
| test_batched_grad(tupled_inputs, o, i) |
| |
| test_backward_mul_by_grad_output(outputs, tupled_inputs, check_sparse_nnz) |
| |
| if check_undefined_grad: |
| test_undefined_grad(func, outputs, tupled_inputs) |
| return True |
| |
| |
| def gradgradcheck( |
| func: Callable[..., _TensorOrTensors], # See Note [VarArg of Tensors] |
| inputs: _TensorOrTensors, |
| grad_outputs: Optional[_TensorOrTensors] = None, |
| eps: float = 1e-6, |
| atol: float = 1e-5, |
| rtol: float = 1e-3, |
| gen_non_contig_grad_outputs: bool = False, |
| raise_exception: bool = True, |
| nondet_tol: float = 0.0, |
| check_undefined_grad: bool = True, |
| check_grad_dtypes: bool = False, |
| check_batched_grad: bool = False, |
| fast_mode: bool = False, |
| ) -> bool: |
| r"""Check gradients of gradients computed via small finite differences |
| against analytical gradients w.r.t. tensors in :attr:`inputs` and |
| :attr:`grad_outputs` that are of floating point or complex type and with |
| ``requires_grad=True``. |
| |
| This function checks that backpropagating through the gradients computed |
| to the given :attr:`grad_outputs` are correct. |
| |
| The check between numerical and analytical gradients uses :func:`~torch.allclose`. |
| |
| .. note:: |
| The default values are designed for :attr:`input` and |
| :attr:`grad_outputs` of double precision. This check will likely fail if |
| they are of less precision, e.g., ``FloatTensor``. |
| |
| .. warning:: |
| If any checked tensor in :attr:`input` and :attr:`grad_outputs` has |
| overlapping memory, i.e., different indices pointing to the same memory |
| address (e.g., from :func:`torch.expand`), this check will likely fail |
| because the numerical gradients computed by point perturbation at such |
| indices will change values at all other indices that share the same |
| memory address. |
| |
| Args: |
| func (function): a Python function that takes Tensor inputs and returns |
| a Tensor or a tuple of Tensors |
| inputs (tuple of Tensor or Tensor): inputs to the function |
| grad_outputs (tuple of Tensor or Tensor, optional): The gradients with |
| respect to the function's outputs. |
| eps (float, optional): perturbation for finite differences |
| atol (float, optional): absolute tolerance |
| rtol (float, optional): relative tolerance |
| gen_non_contig_grad_outputs (bool, optional): if :attr:`grad_outputs` is |
| ``None`` and :attr:`gen_non_contig_grad_outputs` is ``True``, the |
| randomly generated gradient outputs are made to be noncontiguous |
| raise_exception (bool, optional): indicating whether to raise an exception if |
| the check fails. The exception gives more information about the |
| exact nature of the failure. This is helpful when debugging gradchecks. |
| nondet_tol (float, optional): tolerance for non-determinism. When running |
| identical inputs through the differentiation, the results must either match |
| exactly (default, 0.0) or be within this tolerance. Note that a small amount |
| of nondeterminism in the gradient will lead to larger inaccuracies in |
| the second derivative. |
| check_undefined_grad (bool, optional): if True, check if undefined output grads |
| are supported and treated as zeros |
| check_batched_grad (bool, optional): if True, check if we can compute |
| batched gradients using prototype vmap support. Defaults to False. |
| fast_mode (bool, optional): if True, run a faster implementation of gradgradcheck that |
| no longer computes the entire jacobian. |
| |
| Returns: |
| True if all differences satisfy allclose condition |
| """ |
| tupled_inputs = _as_tuple(inputs) |
| |
| if grad_outputs is None: |
| # If grad_outputs is not specified, create random Tensors of the same |
| # shape, type, and device as the outputs |
| def randn_like(x): |
| y = torch.testing.randn_like( |
| x if (x.is_floating_point() or x.is_complex()) else x.double(), memory_format=torch.legacy_contiguous_format) |
| if gen_non_contig_grad_outputs: |
| y = torch.testing.make_non_contiguous(y) |
| return y.requires_grad_() |
| outputs = _as_tuple(func(*tupled_inputs)) |
| tupled_grad_outputs = tuple(randn_like(x) for x in outputs) |
| else: |
| tupled_grad_outputs = _as_tuple(grad_outputs) |
| |
| num_outputs = len(tupled_grad_outputs) |
| |
| def new_func(*args): |
| input_args = args[:-num_outputs] |
| grad_outputs = args[-num_outputs:] |
| outputs = _differentiable_outputs(func(*input_args)) |
| input_args = tuple(x for x in input_args if isinstance(x, torch.Tensor) and x.requires_grad) |
| grad_inputs = torch.autograd.grad(outputs, input_args, grad_outputs, create_graph=True) |
| return grad_inputs |
| |
| return gradcheck( |
| new_func, tupled_inputs + tupled_grad_outputs, eps, atol, rtol, raise_exception, |
| nondet_tol=nondet_tol, check_undefined_grad=check_undefined_grad, |
| check_grad_dtypes=check_grad_dtypes, check_batched_grad=check_batched_grad, fast_mode=fast_mode) |
