U 9%e@sddlZddlmZddlmZddlZddlmZdddd d d gZd a d dZ ddddZ dddd Z edddgZ Gddde Zdddd ZGdd d eZejjZGdddeZdS)N) namedtuple)Any)_DecoratorContextManagerUnpackedDualTensorenter_dual_levelexit_dual_level make_dual unpack_dual dual_levelcCs&tj}|tdkrtd|a|S)aFunction that can be used to enter a new forward grad level. This level can be used to make and unpack dual Tensors to compute forward gradients. This function also updates the current level that is used by default by the other functions in this API. rzoEntering a new forward AD level but the current level is not valid. Make sure you did not modified it directly.)torch_CZ_enter_dual_level_current_level RuntimeError)Z new_levelrX/var/www/html/Darija-Ai-API/env/lib/python3.8/site-packages/torch/autograd/forward_ad.pyrs  levelcCs6|dkr t}|tkrtdtjj|d|dadS)a-Function that can be used to exit a forward grad level. This function deletes all the gradients associated with this level. Only deleting the latest entered level is allowed. This function also updates the current level that is used by default by the other functions in this API. NzdTrying to exit a forward AD level that was not the last one that was created. This is not supported.rr)rrr rZ_exit_dual_levelrrrrr)s cCstjdddkrddlm}|dkr*t}|dkr:td|sZ|sZt d|j |sz|szt d |j t j j |||d S) aAssociates a tensor value with a forward gradient, the tangent, to create a "dual tensor", which is used to compute forward AD gradients. The result is a new tensor aliased to :attr:`tensor` with :attr:`tangent` embedded as an attribute as-is if it has the same storage layout or copied otherwise. The tangent attribute can be recovered with :func:`unpack_dual`. This function is backward differentiable. Given a function `f` whose jacobian is `J`, it allows one to compute the Jacobian-vector product (`jvp`) between `J` and a given vector `v` as follows. Example:: >>> # xdoctest: +SKIP("Undefined variables") >>> with dual_level(): ... inp = make_dual(x, v) ... out = f(inp) ... y, jvp = unpack_dual(out) Please see the `forward-mode AD tutorial `__ for detailed steps on how to use this API. Z PYTORCH_JIT1Tr)decompositions_for_jvpNziTrying to create a dual Tensor for forward AD but no level exists, make sure to enter_dual_level() first.z:Expected primal to be floating point or complex, but got: z;Expected tangent to be floating point or complex, but got: r)osenvirongetZ torch._decomprrrZis_floating_pointZ is_complex ValueErrorZdtyper _VFZ _make_dual)tensortangentrrrrrr =s"(   _UnpackedDualTensorprimalrc@seZdZdZdS)rzNamedtuple returned by :func:`unpack_dual` containing the primal and tangent components of the dual tensor. See :func:`unpack_dual` for more details.N)__name__ __module__ __qualname____doc__rrrrrscCs<|dkr t}|dkrt|dStjj||d\}}t||S)aUnpacks a "dual tensor" to get both its Tensor value and its forward AD gradient. The result is a namedtuple ``(primal, tangent)`` where ``primal`` is a view of :attr:`tensor`'s primal and ``tangent`` is :attr:`tensor`'s tangent as-is. Neither of these tensors can be dual tensor of level :attr:`level`. This function is backward differentiable. Example:: >>> # xdoctest: +SKIP("Undefined variables") >>> with dual_level(): ... inp = make_dual(x, x_t) ... out = f(inp) ... y, jvp = unpack_dual(out) ... jvp = unpack_dual(out).tangent Please see the `forward-mode AD tutorial `__ for detailed steps on how to use this API. Nrr)rrr rZ _unpack_dual)rrrZdualrrrr s  c@s,eZdZdZddZeeeddddZdS)r aContext-manager that enables forward AD. All forward AD computation must be performed in a ``dual_level`` context. .. Note:: The ``dual_level`` context appropriately enters and exit the dual level to controls the current forward AD level, which is used by default by the other functions in this API. We currently don't plan to support nested ``dual_level`` contexts, however, so only a single forward AD level is supported. To compute higher-order forward grads, one can use :func:`torch.func.jvp`. Example:: >>> # xdoctest: +SKIP("Undefined variables") >>> x = torch.tensor([1]) >>> x_t = torch.tensor([1]) >>> with dual_level(): ... inp = make_dual(x, x_t) ... # Do computations with inp ... out = your_fn(inp) ... _, grad = unpack_dual(out) >>> grad is None False >>> # After exiting the level, the grad is deleted >>> _, grad_after = unpack_dual(out) >>> grad is None True Please see the `forward-mode AD tutorial `__ for detailed steps on how to use this API. cCstSN)rselfrrr __enter__szdual_level.__enter__Nexc_type exc_value tracebackreturncCs tdSr$)rr&r)r*r+rrr__exit__szdual_level.__exit__)r r!r"r#r'rr.rrrrr s"c@s>eZdZeddddZddddZeeeddd d ZdS) _set_fwd_grad_enabledN)moder,cCst|_tj|dSr$)_is_fwd_grad_enabledprevr rr/)r&r0rrr__init__sz_set_fwd_grad_enabled.__init__)r,cCsdSr$rr%rrrr'sz_set_fwd_grad_enabled.__enter__r(cCstj|jdSr$)r rr/r2r-rrrr.sz_set_fwd_grad_enabled.__exit__)r r!r"boolr3r'rr.rrrrr/sr/)r collectionsrtypingrr Z grad_moder__all__rrrr rrr r rr1r/rrrrs*    ?+