U dD @sJdZddlZddlZddlmZddlmZmZmZm Z m Z m Z m Z m Z ddlmZddlmZmZddlmZmZdd lmZmZmZmZdd lmZmZd d lmZmZm Z dd l!m"Z"ddl!m#Z#ddl!m$Z$d dl!m%Z%ddddgZ&e ej'Z(e ej'e e(e)e e(dfdddZ*e ee+e e(dfdddZ,d-ee ee e)e)e ee eddddZ-d.eee ee e)e)e)e)e)e ej'dfd d!d"Z.d#d$Z/d%d&Zee_ej01se2d'dd(l3m4Z4m5Z5m6Z6m7Z7m8Z8m9Z9m:Z:m;Z;mZ>m?Z?m@Z@mAZAmBZBmCZCmDZDmEZEdd)l3mFZFmGZGmHZHmIZImJZJmKZKdd*l!mLZLd+d,ZMdS)/a ``torch.autograd`` provides classes and functions implementing automatic differentiation of arbitrary scalar valued functions. It requires minimal changes to the existing code - you only need to declare :class:`Tensor` s for which gradients should be computed with the ``requires_grad=True`` keyword. As of now, we only support autograd for floating point :class:`Tensor` types ( half, float, double and bfloat16) and complex :class:`Tensor` types (cfloat, cdouble). N)_TensorOrTensors)AnyCallableListOptionalSequenceTupleUnioncast)Variable)FunctionNestedIOFunction) gradcheck gradgradcheck)no_grad enable_gradset_grad_enabledinference_mode)detect_anomalyset_detect_anomaly)has_torch_functionhandle_torch_functionis_tensor_like) functional) forward_ad)graph)_vmap_internalsr r backward grad_mode.)outputsgradsis_grads_batchedreturncCsg}t||D]\}}t|tjrD|s0|jn |jdd}|j|ks|rtdt||dt|jdt||dt|jdnHtdt||dt|jdt||dt|jd|jj |jj kr8tdt||d t|jdt||d t|jd| |q|dkr|j r| dkrltd | tj |tjd n | dqtd t|jqt|S) Nr zIf `is_grads_batched=True`, we interpret the first dimension of each grad_output as the batch dimension. The sizes of the remaining dimensions are expected to match the shape of corresponding output, but a mismatch was detected: grad_output[z] has a shape of z and output[z_. If you only want some tensors in `grad_output` to be considered batched, consider using vmap.zMismatch in shape: grad_output[.zuFor complex Tensors, both grad_output and output are required to have the same dtype. Mismatch in dtype: grad_output[z] has a dtype of z6grad can be implicitly created only for scalar outputs)Z memory_formatz1gradients can be either Tensors or None, but got )zip isinstancetorchTensorshape RuntimeErrorstrindexZdtypeZ is_complexappendZ requires_gradZnumelZ ones_likeZpreserve_format TypeErrortype__name__tuple)r!r"r#Z new_gradsoutgradZ grad_shaper5;/tmp/pip-unpacked-wheel-ua33x9lu/torch/autograd/__init__.py _make_gradss           r7)tensorslengthr$cCs*|dkrd|St|tjr"|fSt|S)NN)r'r(r)r2)r8r9r5r5r6_tensor_or_tensors_to_tupleMs  r;F)r8 grad_tensors retain_graph create_graphgrad_variablesinputsr$c Cs|dk r(td|dkr |}ntd|dk rDt|dkrDtdt|tjrV|fnt|}t|tjrp|fn|dk rt|nt}t|t|}t ||dd}|dkr|}t j j |||||ddd dS) a Computes the sum of gradients of given tensors with respect to graph leaves. The graph is differentiated using the chain rule. If any of ``tensors`` are non-scalar (i.e. their data has more than one element) and require gradient, then the Jacobian-vector product would be computed, in this case the function additionally requires specifying ``grad_tensors``. It should be a sequence of matching length, that contains the "vector" in the Jacobian-vector product, usually the gradient of the differentiated function w.r.t. corresponding tensors (``None`` is an acceptable value for all tensors that don't need gradient tensors). This function accumulates gradients in the leaves - you might need to zero ``.grad`` attributes or set them to ``None`` before calling it. See :ref:`Default gradient layouts` for details on the memory layout of accumulated gradients. .. note:: Using this method with ``create_graph=True`` will create a reference cycle between the parameter and its gradient which can cause a memory leak. We recommend using ``autograd.grad`` when creating the graph to avoid this. If you have to use this function, make sure to reset the ``.grad`` fields of your parameters to ``None`` after use to break the cycle and avoid the leak. .. note:: If you run any forward ops, create ``grad_tensors``, and/or call ``backward`` in a user-specified CUDA stream context, see :ref:`Stream semantics of backward passes`. .. note:: When ``inputs`` are provided and a given input is not a leaf, the current implementation will call its grad_fn (even though it is not strictly needed to get this gradients). It is an implementation detail on which the user should not rely. See https://github.com/pytorch/pytorch/pull/60521#issuecomment-867061780 for more details. Args: tensors (Sequence[Tensor] or Tensor): Tensors of which the derivative will be computed. grad_tensors (Sequence[Tensor or None] or Tensor, optional): The "vector" in the Jacobian-vector product, usually gradients w.r.t. each element of corresponding tensors. None values can be specified for scalar Tensors or ones that don't require grad. If a None value would be acceptable for all grad_tensors, then this argument is optional. retain_graph (bool, optional): If ``False``, the graph used to compute the grad will be freed. Note that in nearly all cases setting this option to ``True`` is not needed and often can be worked around in a much more efficient way. Defaults to the value of ``create_graph``. create_graph (bool, optional): If ``True``, graph of the derivative will be constructed, allowing to compute higher order derivative products. Defaults to ``False``. inputs (Sequence[Tensor] or Tensor, optional): Inputs w.r.t. which the gradient be will accumulated into ``.grad``. All other Tensors will be ignored. If not provided, the gradient is accumulated into all the leaf Tensors that were used to compute the attr::tensors. Nz;'grad_variables' is deprecated. Use 'grad_tensors' instead.zu'grad_tensors' and 'grad_variables' (deprecated) arguments both passed to backward(). Please only use 'grad_tensors'.rz0'inputs' argument to backward() cannot be empty.Fr#T)Zallow_unreachableaccumulate_grad) warningswarnr+lenr'r(r)r2r;r7r _execution_engine run_backward)r8r<r=r>r?r@Z grad_tensors_r5r5r6rUs0A T) r!r@ grad_outputsr=r> only_inputs allow_unusedr#r$c stttjdft|r|fnt|tttjdft|rB|fnt|}t|rxtt||||d S|st dt |t } t | |d} dkr|r܇fdd} tj| ddd d | Stjj| d d SdS) a Computes and returns the sum of gradients of outputs with respect to the inputs. ``grad_outputs`` should be a sequence of length matching ``output`` containing the "vector" in vector-Jacobian product, usually the pre-computed gradients w.r.t. each of the outputs. If an output doesn't require_grad, then the gradient can be ``None``). .. note:: If you run any forward ops, create ``grad_outputs``, and/or call ``grad`` in a user-specified CUDA stream context, see :ref:`Stream semantics of backward passes`. .. note:: ``only_inputs`` argument is deprecated and is ignored now (defaults to ``True``). To accumulate gradient for other parts of the graph, please use ``torch.autograd.backward``. Args: outputs (sequence of Tensor): outputs of the differentiated function. inputs (sequence of Tensor): Inputs w.r.t. which the gradient will be returned (and not accumulated into ``.grad``). grad_outputs (sequence of Tensor): The "vector" in the vector-Jacobian product. Usually gradients w.r.t. each output. None values can be specified for scalar Tensors or ones that don't require grad. If a None value would be acceptable for all grad_tensors, then this argument is optional. Default: None. retain_graph (bool, optional): If ``False``, the graph used to compute the grad will be freed. Note that in nearly all cases setting this option to ``True`` is not needed and often can be worked around in a much more efficient way. Defaults to the value of ``create_graph``. create_graph (bool, optional): If ``True``, graph of the derivative will be constructed, allowing to compute higher order derivative products. Default: ``False``. allow_unused (bool, optional): If ``False``, specifying inputs that were not used when computing outputs (and therefore their grad is always zero) is an error. Defaults to ``False``. is_grads_batched (bool, optional): If ``True``, the first dimension of each tensor in ``grad_outputs`` will be interpreted as the batch dimension. Instead of computing a single vector-Jacobian product, we compute a batch of vector-Jacobian products for each "vector" in the batch. We use the vmap prototype feature as the backend to vectorize calls to the autograd engine so that this computation can be performed in a single call. This should lead to performance improvements when compared to manually looping and performing backward multiple times. Note that due to this feature being experimental, there may be performance cliffs. Please use ``torch._C._debug_only_display_vmap_fallback_warnings(True)`` to show any performance warnings and file an issue on github if warnings exist for your use case. Defaults to ``False``. .)rHr=r>rIrJr#zonly_inputs argument is deprecated and is ignored now (defaults to True). To accumulate gradient for other parts of the graph, please use torch.autograd.backward.rANc stjj|ddS)NFrB)r rFrG)ZgOrJr>r=Zt_inputsZ t_outputsr5r6vjpszgrad..vjprT)Zallow_none_pass_throughFrK)r rr(r)rr2rrr4rCrDr;rEr7rZ_vmapr rFrG) r!r@rHr=r>rIrJr#Zoverridable_argsZ grad_outputs_rMr5rLr6r4sD=&&  r4cCs tjSr:)r rFZis_checkpoint_validr5r5r5r6_is_checkpoint_valid&srNcOs tddS)NzItorch.autograd.variable(...) is deprecated, use torch.tensor(...) instead)r+)argskwargsr5r5r6variable*srQzautograd initialization failed) DeviceTypeProfilerActivity ProfilerStateProfilerConfig ProfilerEvent_enable_profiler_legacy_disable_profiler_legacy_profiler_enabled_enable_record_function_set_empty_test_observerkineto_available _record_function_with_args_enter_record_function_with_args_exit_supported_activities_add_metadata_json SavedTensor!_push_saved_tensors_default_hooks _pop_saved_tensors_default_hooks)_ProfilerResult _KinetoEvent _kineto_step_prepare_profiler_enable_profiler_disable_profiler)profilercCs$t|tstdtj||dS)Nzcls isn't a typeinfo object)r'r0r+r(_CZ_register_py_class_for_device)Zdeviceclsr5r5r6$_register_py_tensor_class_for_deviceCs rm)NNFNN)NNFTFF)N__doc__r(rCZ torch.typesrtypingrrrrrrr r rQr functionr rrrr rrrrZ anomaly_moderrZ overridesrrrrrrr__all__r)Z_OptionalTensorboolr7intr;rr4rNrkZ_autograd_initr+Ztorch._C._autogradrRrSrTrUrVrWrXrYrZr[r\r]r^r_r`rarbrcrdrerfrgrhrirjrmr5r5r5r6sz (        /  _ u P