o + i@sZddlmZddlZddlZddlZddlmZddlmZddl Z ddl m Z ddl mZddlmZddlmZd d lmZerLdd lmZmZmZd aejejejejejejgZ id d dd ddddddddddddddddddddddddddddddddddZ!gd Z"d!d"Z#ee#Z$d#d$Z%d%d&Z&dS)') annotationsN)reduce) TYPE_CHECKING)_C_ops)DataType)wrap_decorator)size_args_decorator_patch)Value) DTypeLike PlaceLike ShapeLikeFfloat16Zhalfbfloat16float32floatfloat64doubleint8charint16Zshortint32intint64longbool complex64 complex128)rZcfloatZcdouble)__add____radd____sub____rsub____mul____rmul____mod____rmod____div____rdiv__ __truediv__ __rtruediv__ __floordiv__ __rfloordiv____pow____rpow____eq____ne____lt____le____gt____ge__csfdd}|S)Ncstdjdjd)N'a' only can be called by `paddle.Tensor` in dynamic graph mode. Suggestions: 1. If you are in static graph mode, you can switch to dynamic graph mode by turning off `paddle.enable_static()` or calling `paddle.disable_static()`. 2. If you are using `@paddle.jit.to_static`, you can call `paddle.jit.enable_to_static(False)`. If you have to translate dynamic graph to static graph, please use other API to replace 'z'.)AssertionError__name__argskwargsfuncd/home/app/PaddleOCR-VL-test/.venv_paddleocr/lib/python3.10/site-packages/paddle/pir/math_op_patch.py__impl__is  z'_fake_interface_only_..__impl__r<)r;r>r<r:r=_fake_interface_only_hs r?c Cst|tsJt|}d}g}t|jD]\}}|dkr.|dkr(|}||q|dq||q|dks:Jddlm}t ||||||| }d|_ |S)Nrr )coreT) isinstancer r enumerateshapeappendZpaddle.frameworkrArZfull_batch_size_likePlace stop_gradient) Zref_varvaluedtypeZ batch_dimZ out_shapeidrAoutr<r<r=create_tensor_with_batchsizews$     rMc8sdddd}ddd}tdd }d d }d d }tdd}dd}dd}dfdd }fddfdd} fdd} dd} fd d!} fd"d#} fd$d%}fd&d'}fd(d)}fd*d+}fd,d-} . dfd/d0 }tfd1d2}td3d4}td5d6}ddd.d.d7dfdCdD}tddd.d.d7dfdEdF}tddd.d.d7dfdGdH}tddd.d.d7dfdIdJ}dKdL}dMdN}dOdP}dQdR}fdSdT}tdUdV} dWdX}!fdYdZ}"d[d\}#d]d^}$d_d`}%fdadb}&dcdd}' dfdedf }(fdgdh})tdidj}*tdkdl}+tdmdn},tddpdq}-|-jddtdq}-dudlgdv|fdw|fdx|fdy|fdz|fd{|fd||fd}|fd~|fdfd| fd| fd| fd8|fd|fd|fd|fd|fd|fd|fd@|-fd|fd| fd|!fd|"fd|&fd|'fd|#fd|%fd|$fd|(fd|)fd|+fd|*fd|,fd|djjd.| fd|djjd.| fd|djjd.| fd|djjd|fd|djjd.|fd|djjd.|fd|djj d.|fd|djj d.|fd|djj ddfd|djj ddfd|djj d.dfd|djj ddfd|djj d.dfd|djj ddfd|djj d.dfd|djj ddfd|djj d.dfd|djj ddfd|fd|fd|djjd.dfd|djjd.dfd|djjd.dfd|djjd.dfd|djjd.dfd|djjd.dfd|fd|fd|fd|f}.| }/|.|/tsW|.D]}0|0du}1|0d}2tt|1|2qdudljjD]}1tt|1rqtj|1d}2|2r)tt|1|2qjjD]\}3}4tj|4d}5|5rCtt|3|5q/ddlm}6m}7|6t_ |7t_!dadSdS)NcSsz|j}W|Std)NzCannot get data type from var)rI ValueError)varrIr<r<r=safe_get_dtypes z*monkey_patch_value..safe_get_dtypecSs t|dS)a In dy2static, Value also needs cpu() and cuda() interface. But, the underneath operator has only forward op but not backward one. Returns: The tensor which has copied to cpu place. Examples: In Static Graph Mode: .. code-block:: python >>> import paddle >>> paddle.enable_static() >>> x = paddle.static.data(name="x", shape=[2,2], dtype='float32') >>> y = x.cpu() r)rmemcpyselfr<r<r=cpus zmonkey_patch_value..cpuTcSs0|dur td|durtdt|dS)a In dy2static, Value also needs cpu() and cuda() interface. But, the underneath operator has only forward op but not backward one. Args: self(Value): The variable itself. device_id(int, optional): The destination GPU device id. Default: None, means current device. We add this argument for dy2static translation, please do not use it. blocking(bool, optional): Whether blocking or not, Default: True. We add this argument for dy2static translation, please do not use it. Returns: The tensor which has copied to cuda place. Examples: In Static Graph Mode: .. code-block:: python >>> import paddle >>> paddle.enable_static() >>> x = paddle.static.data(name="x", shape=[2,2], dtype='float32') >>> y = x.cpu() >>> z = y.cuda() Nz3device_id is not supported, and it will be ignored.T2blocking is not supported, and it will be ignored.r )warningswarnrrQ)rSZ device_idblockingr<r<r=cudas    z monkey_patch_value..cudacStddS)z Value don't have 'place' interface in static graph mode But this interface can greatly facilitate dy2static. So we give a warning here and return None. zaValue do not have 'place' interface for pir graph mode, try not to use it. None will be returned.NrVrWrRr<r<r=placesz!monkey_patch_value..placecSstd|S)z Value don't have 'contiguous' interface in static graph mode But this interface can greatly facilitate dy2static. So we give a warning here and return None. ziValue do not have 'contiguous' interface for static graph mode, try not to use it. self will be returned.r[rRr<r<r= contiguousz&monkey_patch_value..contiguouscSrZ)z Value don't have 'is_contiguous' interface in static graph mode But this interface can greatly facilitate dy2static. So we give a warning here and return None. zlValue do not have 'is_contiguous' interface for static graph mode, try not to use it. True will be returned.Tr[rRr<r<r= is_contiguousr^z)monkey_patch_value..is_contiguouscS t|jS)a Returns the dimension of current Value Returns: the dimension Examples: .. code-block:: python >>> import paddle >>> paddle.enable_static() >>> # create a static Value >>> x = paddle.static.data(name='x', shape=[3, 2, 1]) >>> # print the dimension of the Value >>> print(x.ndim) 3 lenrDrRr<r<r=_ndims z!monkey_patch_value.._ndimcSr`)a Returns the dimension of current Value Returns: the dimension Examples: .. code-block:: python >>> import paddle >>> paddle.enable_static() >>> # create a static Value >>> x = paddle.static.data(name='x', shape=[3, 2, 1]) >>> # print the dimension of the Value >>> print(x.ndimension()) 3 rarRr<r<r= ndimension  z&monkey_patch_value..ndimensioncSr`)a Returns the dimension of current Value Returns: the dimension Examples: .. code-block:: python >>> import paddle >>> paddle.enable_static() >>> # create a static Value >>> x = paddle.static.data(name='x', shape=[3, 2, 1]) >>> # print the dimension of the Value >>> print(x.dim()) 3 rarRr<r<r=dim"rezmonkey_patch_value..dimr8rc s|r |s dSddlm}||}t|dkr!tdd|nd}d}t|dkrE z3monkey_patch_value.._item..r zUonly one element tensors can be converted to Python scalars when no input coordinateszindex z is out of bounds for size z&incorrect number of indices for Tensorzargument (position z) must be long, but got z is out of bounds for axis z with size )Zis_distZ_is_initializedZpaddle.jit.dy2staticrgrbrrBpirr rNranger TypeErrortypeflatten) rSr8rgdimsnumeloffsetstridesrJindexpaddler<r=_item8sL       z!monkey_patch_value.._itemcs2t|ts jj|}|j|kr|St||S)a **Notes**: Convert a value to a specified data type if it differs from the current dtype; otherwise, return the original value. Args: self(Value): The source Value dtype: The target data type Returns: Value: Value with new dtype Examples: In Static Graph Mode: .. code-block:: python >>> import paddle >>> paddle.enable_static() >>> startup_prog = paddle.static.Program() >>> main_prog = paddle.static.Program() >>> with paddle.static.program_guard(startup_prog, main_prog): ... original_value = paddle.static.data(name = "new_value", shape=[2,2], dtype='float32') ... new_value = original_value.astype('int64') ... print(f"new value's dtype is: {new_value.dtype}") ... new Value's dtype is: paddle.int64 )rBrrmrAZconvert_np_dtype_to_dtype_rIrcast)rSrIrwr<r=astypeqs "  z"monkey_patch_value..astypecsF|r|d}|dS|r|d}|dS|dS)Nruint8)Zis_floating_pointZ is_complexreal)rStensorr}r{r<r=bytes    z monkey_patch_value..bytecsVg}tD]"\}}fdd}||}||_d|d|d|_|||fq|S)z? Batch create all data type conversion methods cfdd}|S)Ncs |Srhr<rRr{rIr<r=conversion_method zwmonkey_patch_value.._create_dtype_conversion_methods..make_conversion_method..conversion_methodr<)rIrrrIr=make_conversion_methodsz\monkey_patch_value.._create_dtype_conversion_methods..make_conversion_methodz Cast a Value to z data type if it differs from the current dtype; otherwise, return the original Value. Returns: Value: a new Value with z dtype )_supported_dtype_conversionsitemsr6__doc__rE)methods method_nameZ target_dtyper method_implrr<r= _create_dtype_conversion_methodss  z._create_dtype_conversion_methodscSs ||jSrhr)rSotherr<r<r=type_as z#monkey_patch_value..type_asc|d|SN?scalerOrHrwr<r= _scalar_add_z(monkey_patch_value.._scalar_add_cs|d| Srrrrwr<r= _scalar_sub_sz(monkey_patch_value.._scalar_sub_cr)Nrrrwr<r= _scalar_rsub_rz)monkey_patch_value.._scalar_rsub_cs||dS)Nrrrwr<r= _scalar_mul_rz(monkey_patch_value.._scalar_mul_cs|d|dS)Nrrrrrwr<r= _scalar_div_sz(monkey_patch_value.._scalar_div_cs|ddS)NrrrrOrwr<r= _scalar_neg_rz(monkey_patch_value.._scalar_neg_c |Srh)absrrwr<r= _scalar_abs_rz(monkey_patch_value.._scalar_abs_Fcs&fdd}d|_||_|S)Ncst|tr|jtvr|tj}dur||Sn2t|trKt|}jkr4|jtvr4|tj}|jj krA|tj }durJ||Sn |}t|t s}rt|j D]}|dkrht |||}n qZjj|j ||}n jjg||}r|}|}|}||}|S)Nr)rBrrI_supported_int_dtype_rZFLOAT32rdividerzrINT64r rDrMr~ZcreationZ fill_constant)rSZ other_varZ lhs_dtypeelemtmprL)r{rx python_apireverserP scalar_methodr<r=r>sX           z>monkey_patch_value.._binary_creator_..__impl__z Args: self(Value): left hand Value other_var(Value|float|int): right hand Value Returns: Value )rr6)rrrrr>r{rxrP)rrrr=_binary_creator_sHz,monkey_patch_value.._binary_creator_cr)a Returns the number of elements for current Value, which is a int64 Value with shape [] . Returns: Value, the number of elements for current Value Examples: .. code-block:: python >>> import paddle >>> paddle.enable_static() >>> startup_prog = paddle.static.Program() >>> main_prog = paddle.static.Program() >>> with paddle.static.program_guard(startup_prog, main_prog): ... x = paddle.assign(np.random.rand(2, 3, 4).astype("float32")) ... (output_x,) = exe.run(main_program, fetch_list=[x.size]) ... print(f"value's size is: {output_x}") ... value's size is: 24 )rsrRrwr<r=_size_/s z"monkey_patch_value.._size_cSs4t|jdkr |Stttt|j}t||S)aZ Permute current Value with its dimensions reversed. If `n` is the dimensions of `x` , `x.T` is equivalent to `x.transpose([n-1, n-2, ..., 0])`. Examples: .. code-block:: python >>> import paddle >>> paddle.enable_static() >>> x = paddle.ones(shape=[2, 3, 5]) >>> x_T = x.T >>> exe = paddle.static.Executor() >>> x_T_np = exe.run(paddle.static.default_main_program(), fetch_list=[x_T])[0] >>> print(x_T_np.shape) (5, 3, 2) r )rbrDlistreversedrnr transposerSpermr<r<r=_T_Gs zmonkey_patch_value.._T_cSs\t|jdkrtdt|jdttt|j}|d|d|d<|d<t||S)al Permute current Value with its last two dimensions reversed. If `n` is the dimensions of `x` , `x.mT` is equivalent to `x.transpose([0, 1, ..., n-1, n-2])`. Examples: .. code-block:: python >>> import paddle >>> paddle.enable_static() >>> x = paddle.ones(shape=[2, 3, 5]) >>> x_mT = x.mT >>> exe = paddle.static.Executor() >>> x_mT_np = exe.run(paddle.static.default_main_program(), fetch_list=[x_mT])[0] >>> print(x_mT_np.shape) (2, 5, 3) z Tensor.ndim(z/) is required to be greater than or equal to 2.r@)rbrDrNrrnrrrr<r<r=_mT_ds z monkey_patch_value.._mT_rIdevice requires_grad pin_memorysizer fill_valuebool | float | paddle.TensorrIDTypeLike | NonerPlaceLike | Nonerrrcs2|dur|j}|dur|j}j||||||dS)a$ Returns a Tensor of size ``size`` filled with ``fill_value``. By default, the returned Tensor has the same dtype and place as this tensor. Examples: .. code-block:: python >>> import paddle >>> paddle.enable_static() >>> x = paddle.ones(shape=[2, 3, 5]) >>> x_new = x.new_full([2, 3], 3.14, dtype="float64", device="cpu") >>> exe = paddle.static.Executor() >>> x_new_np = exe.run(paddle.static.default_main_program(), fetch_list=[x_new])[0] >>> print(x_new_np.shape) (2, 5, 3) >>> print(str(x_new_np.dtype)) 'paddle.float64' >>> print(x_new_np.place) Place(cpu) NrrIr\full)rSrrrIrrrrwr<r= _new_full_!z&monkey_patch_value.._new_full_cs0|dur|j}|dur|j}j|||||dS)a  Returns a Tensor of size ``size`` filled with uninitialized data. By default, the returned Tensor has the same dtype and place as this tensor. Examples: .. code-block:: python >>> import paddle >>> paddle.enable_static() >>> x = paddle.ones(shape=[2, 3, 5]) >>> x_new = x.new_empty([2, 3], dtype="float64", device="cpu") >>> exe = paddle.static.Executor() >>> x_new_np = exe.run(paddle.static.default_main_program(), fetch_list=[x_new])[0] >>> print(x_new_np.shape) (2, 3) >>> print(str(x_new_np.dtype)) 'paddle.float64' >>> print(x_new_np.place) Place(cpu) Nr)rIr\emptyrSrrIrrrrwr<r= _new_empty_s!z'monkey_patch_value.._new_empty_c2|dur|j}|dur|j}j|d||||dS)a Returns a Tensor of size ``size`` filled with ``1``. By default, the returned Tensor has the same dtype and place as this tensor. Examples: .. code-block:: python >>> import paddle >>> paddle.enable_static() >>> x = paddle.ones(shape=[2, 3, 5]) >>> x_new = x.new_ones([2, 3], dtype="float64", device="cpu") >>> exe = paddle.static.Executor() >>> x_new_np = exe.run(paddle.static.default_main_program(), fetch_list=[x_new])[0] >>> print(x_new_np.shape) (2, 3) >>> print(str(x_new_np.dtype)) 'paddle.float64' >>> print(x_new_np.place) Place(cpu) Nr rrrrwr<r= _new_ones_rz&monkey_patch_value.._new_ones_cr)a Returns a Tensor of size ``size`` filled with ``0``. By default, the returned Tensor has the same dtype and place as this tensor. Examples: .. code-block:: python >>> import paddle >>> paddle.enable_static() >>> x = paddle.ones(shape=[2, 3, 5]) >>> x_new = x.new_zeros([2, 3], dtype="float64", device="cpu") >>> exe = paddle.static.Executor() >>> x_new_np = exe.run(paddle.static.default_main_program(), fetch_list=[x_new])[0] >>> print(x_new_np.shape) (2, 3) >>> print(str(x_new_np.dtype)) 'paddle.float64' >>> print(x_new_np.place) Place(cpu) Nrrrrrwr<r= _new_zeros_rz'monkey_patch_value.._new_zeros_cSd}tt|)Na int(Tensor) is not supported in static graph mode. Because it's value is not available during the static mode. It's usually triggered by the logging implicitly, for example: >>> logging.info("The value of x is: {int(x)}") ^ `x` is Tensor, `int(x)` triggers int(Tensor) There are two common workarounds available: If you are logging Tensor values, then consider logging only at dynamic graphs, for example: Modify the following code >>> logging.info("The value of x is: {int(x)}") to >>> if paddle.in_dynamic_mode(): ... logging.info("The value of x is: {int(x)}") If you need to convert the Tensor type, for example: Modify the following code >>> x = int(x) to >>> x = x.astype("int64") rotextwrapdedentrS error_msgr<r<r=_int_@sz!monkey_patch_value.._int_cSr)Na float(Tensor) is not supported in static graph mode. Because it's value is not available during the static mode. It's usually triggered by the logging implicitly, for example: >>> logging.info("The value of x is: {float(x)}") ^ `x` is Tensor, `float(x)` triggers float(Tensor) There are two common workarounds available: If you are logging Tensor values, then consider logging only at dynamic graphs, for example: Modify the following code >>> logging.info("The value of x is: {float(x)}") to >>> if paddle.in_dynamic_mode(): ... logging.info("The value of x is: {float(x)}") If you need to convert the Tensor type, for example: Modify the following code >>> x = float(x) to >>> x = x.astype("float64") rrr<r<r=_float_Yz#monkey_patch_value.._float_cSr)Na bool(Tensor) is not supported in static graph mode. Because it's value is not available during the static mode. If you haven't call bool(Tensor) explicitly, it's usually triggered by the control flow implicitly, for example: >>> if x > 0: ^ `x` is Tensor, `x` > 0 is also a Tensor, `if x > 0` triggers bool(Tensor) ... y = y + 1 There are two common workarounds available: If you are checking for Tensor values, then consider checking only at dynamic graphs, for example: Modify the following code >>> if x > 0: ... raise ValueError("x should be positive") to >>> if paddle.in_dynamic_mode() and x < 0: >>> raise ValueError("x should be positive") If you need to control the flow of execution based on the value of the Tensor, then you need to rewrite the code as a control flow, for example: Modify the following code >>> if x < y: ... y = y + 1 ... else: ... y = y - 1 to >>> pred = paddle.less_than(x=x, y=y, name=None) >>> y = paddle.static.nn.cond(pred, lambda: y + 1, lambda: y - 1) For more info, please refer to https://www.paddlepaddle.org.cn/documentation/docs/zh/api/paddle/static/nn/cond_cn.html rrr<r<r=_bool_qsz"monkey_patch_value.._bool_cSr)Na( complex(Tensor) is not supported in static graph mode. Because it's value is not available during the static mode. It's usually triggered by the logging implicitly, for example: >>> logging.info("The value of x is: {complex(x)}") ^ `x` is Tensor, `complex(x)` triggers complex(Tensor) There are two common workarounds available: If you are logging Tensor values, then consider logging only at dynamic graphs, for example: Modify the following code >>> logging.info("The value of x is: {complex(x)}") to >>> if paddle.in_dynamic_mode(): ... logging.info("The value of x is: {complex(x)}") If you need to convert the Tensor type, for example: Modify the following code >>> x = complex(x) to >>> x = x.astype("complex64") rrr<r<r= _complex_rz%monkey_patch_value.._complex_cr)a Returns a new static Value, which is the clone of the original static Value. It remains in the current graph, that is, the cloned Value provides gradient propagation. Calling ``out = tensor.clone()`` is same as ``out = assign(tensor)`` . Returns: Value, The cloned Value. Examples: .. code-block:: python >>> import paddle >>> paddle.enable_static() >>> # create a static Value >>> x = paddle.static.data(name='x', shape=[3, 2, 1]) >>> # create a cloned Value >>> y = x.clone() )ZassignrRrwr<r=clones z!monkey_patch_value..clonecSdS)aJ **Notes**: **1. This API is ONLY available in Dygraph mode** **2. Use it only Value has gradient, normally we use this for Parameters since other temporal Value will be deleted by Python's GC** Clear (set to ``0`` ) the Gradient of Current Value Returns: None Examples: .. code-block:: python >>> import paddle >>> import numpy as np >>> x = np.ones([2, 2], np.float32) >>> inputs2 = [] >>> for _ in range(10): >>> tmp = paddle.to_tensor(x) >>> tmp.stop_gradient=False >>> inputs2.append(tmp) >>> ret2 = paddle.add_n(inputs2) >>> loss2 = paddle.sum(ret2) >>> loss2.retain_grads() >>> loss2.backward() >>> print(loss2.gradient()) >>> loss2.clear_gradient() >>> print("After clear {}".format(loss2.gradient())) 1.0 After clear 0.0 Nr<rRr<r<r=clear_gradients"z*monkey_patch_value..clear_gradientcSs<|s td|ddlm}m}|||||ddS)zL Notes: The type of Value must be Tensor Array. zGOnly Value with DenseTensorArray support `append` method, but received r) array_length array_write)rirJarrayN)is_dense_tensor_array_typeroZpaddle.tensor.arrayrr)rSrOrrr<r<r=rEs z"monkey_patch_value..appendcs>|s td|t|dkrd}n|d}j||S)a The type of Value must be Tensor Array. When self is TensorArray, calling pop is similar to Python's pop on list. This interface is used to simplify dygraph to static graph operations. Args: self(Value): The source variable, which must be DenseTensorArray *args: optional, a int means index. Returns: Value: self[index] zDOnly Value with DenseTensorArray support `pop` method, but received rr@)rrorbZ_pir_opsZ array_pop)rSr8idxrwr<r=pops  zmonkey_patch_value..popcS t|Srh)rZsparse_to_denserRr<r<r=to_dense rz$monkey_patch_value..to_densecSrrh)rZ sparse_valuesrRr<r<r=valuesrz"monkey_patch_value..valuescSrrh)rZsparse_indicesrRr<r<r=indicesrz#monkey_patch_value..indicescsNjjjs Jd|s|r#j||}||dSt d)Nz3We only support call 'set_shape' in to_static mode.zBCurrently, we can only set shape for dense and selected_row tensor) baseZdygraphZin_to_static_modeZis_dense_tensor_typeZis_selected_row_typermZcreate_shaped_typerpset_typerN)rSrDrprwr<r= set_shapesz%monkey_patch_value..set_shapecSs tt|Srh)hashidrRr<r<r= value_hash"rz&monkey_patch_value..value_hashcs|dur|dur|durS|dur6t|trj|}nt|jjjjjfr-n t dt |j |dur=d}n t|t sFJdfdd}||||S)Nzdevice value error, must be str, paddle.CPUPlace(), paddle.CUDAPlace(), paddle.CUDAPinnedPlace(), paddle.XPUPlace() or paddle.CustomPlace(), but the type of device is Tz5blocking value error, must be the True, False or Nonecs|dur|j}|}||jkr.jjj|jd|j|d}Wdn1s(wYn|}t|jr>|j|d}|St|j rU|durMt dt d}|St|jra|}|S|}|S)N)r\r)rXTrUr)rIrZ frameworkZ_dygraph_place_guardr\rzrB CUDAPlacerYCUDAPinnedPlacerVrWrrQCPUPlacerT)trrIrXZt_usedZt_castedZnew_t)rxrSr<r= transformJs4      z2monkey_patch_value.._to..transform) rBstrrZ_convert_to_placerArFrrrrNrpr6r)rSrrIrXrrwrRr=_to%s4   zmonkey_patch_value.._toc sJt|}t|}||dks||dkrtdhd}t||}t|dkr4tdtt|d#ddd$d d d#d d  Gd dd fdd fdd fddd%fdd d&fdd fdd f dd }||i|}|d!d}|dur|d!|j|d"<|j |j d'i|S)(a Performs Tensor dtype and/or device conversion. A paddle.dtype and place are inferred from the arguments of ``self.to(*args, **kwargs)``.There are three ways to call `to`: 1. to(dtype, blocking=True) 2. to(device, dtype=None, blocking=True) 3. to(other, blocking=True) Returns: Tensor: self Examples: .. code-block:: python >>> import paddle >>> x = paddle.to_tensor([1,2,3]) >>> print(x) Tensor(shape=[3], dtype=int64, place=Place(gpu:0), stop_gradient=True, [1, 2, 3]) >>> x = x.to("cpu") >>> print(x.place) Place(cpu) >>> x = x.to("float32") >>> print(x.dtype) paddle.float32 >>> x = x.to("gpu", "int16") >>> print(x) Tensor(shape=[3], dtype=int16, place=Place(gpu:0), stop_gradient=True, [1, 2, 3]) >>> y = paddle.to_tensor([4,5,6]) >>> y Tensor(shape=[3], dtype=int64, place=Place(gpu:0), stop_gradient=True, [4, 5, 6]) >>> y = y.to(x) >>> print(y) Tensor(shape=[3], dtype=int16, place=Place(gpu:0), stop_gradient=True, [4, 5, 6]) rato() received too many arguments - expected one of: * (Union[str, paddle.CPUPlace(), paddle.CUDAPlace(), paddle.CUDAPinnedPlace(), paddle.XPUPlace(), paddle.CustomPlace()] device, Union[str, paddle.dtype, numpy.dtype] dtype, bool blocking) * (Union[str, paddle.dtype, numpy.dtype] dtype, bool blocking) * (paddle.Tensor other, bool blocking) >rrrIrXz(to() got an unexpected keyword argument NcSdSrhr<)rIrXr<r<r=dtype_first_sigz7monkey_patch_value..to..dtype_first_sigcSrrhr<)rrIrXr<r<r=device_first_sigrz8monkey_patch_value..to..device_first_sigcSrrhr<)rrXr<r<r=tensor_like_first_sigrz=monkey_patch_value..to..tensor_like_first_sigc@s eZdZdS)z.monkey_patch_value..to.._NoArgN)r6 __module__ __qualname__r<r<r<r=_NoArgrlrcs0gd}t|jtjfpt|to||vS)N) rrrrrrrrr|rrr)rBrInprlower)argZ valid_dtypesrwr<r=is_dtypesz0monkey_patch_value..to..is_dtypecs|dup t|jjtfSrh)rBrArFrrrwr<r= is_devicesz1monkey_patch_value..to..is_devicecst|jjSrh)rBrmr rrwr<r= is_tensorrz1monkey_patch_value..to..is_tensorpositionrcr)Ncst|kr |SSrh)rbr7)rrr<r=extract_positional_args zgmonkey_patch_value..to..create_positional_arg_extractor..extract_positional_argr<)rrrrr=create_positional_arg_extractorszGmonkey_patch_value..to..create_positional_arg_extractorkeyrcsfdd}|S)Ncs6|vrt|krt|kr|SSrh)rbrkeysr7)rrrr<r=extract_keyword_args  zamonkey_patch_value..to..create_keyword_arg_extractor..extract_keyword_argr<)rrrrrrr=create_keyword_arg_extractors zDmonkey_patch_value..to..create_keyword_arg_extractorcr)Ncs,D]}t|||}s|SqSrh)rB)r8r9 extractorr)r extractorsr<r=chains zGmonkey_patch_value..to..chain_extractors..chainr<)r r r)r r=chain_extractorssz8monkey_patch_value..to..chain_extractorsc sdddddfdddddf dddddfi}|D](\}\}}t|||}sX||rXt|j|i|}||jSq0td)NrrrIrrrzNo matching signature found.)rrBinspect signaturebindapply_defaults argumentsrN)r8r9Z signature_mapsigr  conditionrZ bound_args rr r rrrrrrrr<r=dispatch_to_signaturesB    z=monkey_patch_value..to..dispatch_to_signaturerrIrh)NN)rr)rrrrr<) rbrosetrnextitergetrrIr\r) rSr8r9Z size_argsZ size_kwargsZ valid_keysZ invalid_keysrrrwrr=tojs@,          #   zmonkey_patch_value..tocSr)am **Notes**: **This API is ONLY available in Dygraph mode** Returns a numpy array shows the value of current :ref:`api_guide_Variable_en` Returns: ndarray: The numpy value of current Variable. Returns type: ndarray: dtype is same as current Variable Examples: .. code-block:: python >>> import paddle >>> import paddle.base as base >>> from paddle.nn import Linear >>> import numpy as np >>> data = np.random.uniform(-1, 1, [30, 10, 32]).astype('float32') >>> with base.dygraph.guard(): ... linear = Linear(32, 64) ... data_tensor = paddle.to_tensor(data) ... x = linear(data_tensor) ... print(x.numpy()) Nr<rRr<r<r=numpyz!monkey_patch_value..numpycSr)a **Notes**: **This API is ONLY available in Dygraph mode** Returns a Python list that contains the elements of current :ref:`api_guide_Variable_en` Returns: list: The Python list containing the elements of current Variable. Returns type: list: Elements have the same dtype as current Variable Examples: .. code-block:: python >>> import paddle >>> import paddle.base as base >>> import numpy as np >>> data = np.random.uniform(-1, 1, [2, 3]).astype('float32') >>> with base.dygraph.guard(): ... x = paddle.to_tensor(data) ... print(x.tolist()) # Convert tensor to Python list Nr<rRr<r<r=tolist/rz"monkey_patch_value..tolistcSr)z Value don't have 'register_hook' interface in static graph mode But this interface can greatly facilitate dy2static. So we give a error here. Nr<)rShookr<r<r= register_hookIsz)monkey_patch_value..register_hookreturncSs|j S)a! Whether this Tensor requires gradient computation. This is a convenience property that returns the opposite of stop_gradient. Setting requires_grad=True is equivalent to setting stop_gradient=False. Examples: .. code-block:: python >>> import paddle >>> x = paddle.randn([2, 3]) >>> print(x.requires_grad) # False by default >>> >>> x.requires_grad = False >>> print(x.stop_gradient) # True )rGrRr<r<r=rRsz)monkey_patch_value..requires_gradrHNonecSs(t|tstdt|| |_dS)z Set whether this Tensor requires gradient computation. Args: value (bool): True to enable gradient computation, False to disable. z$requires_grad must be bool, but got N)rBrrorprG)rSrHr<r<r=rfs   rrTrYr\r]r_itemrfrdndimr{rr|rTZmTZnew_fullZ new_emptyZnew_onesZ new_zerosrrrErr__hash__rrrrrrrr rrr r!r"r#r&r(r'r)r,r-r*r+r$r% __matmul__ __rmatmul____neg____abs__r.r/r0r1r2r3 __float____int____bool__ __complex__r r)_getitem_static_setitem_static)NT)r8r)FN) rr rrrIrrrrrrr) rr rIrrrrrrr)NNN)r!r)rHrr!r")"propertyrfake_interface_onlysetterrxr~addsubtractmultiplyrpowZ floor_divide remaindermatmulequal not_equal less_thanZ less_equal greater_thanZ greater_equalextend_already_patch_valuesetattrr Z paddle.tensorZtensor_method_funchasattrgetattrZmagic_method_funcZbase.variable_indexr/r0 __getitem__ __setitem__)8rTrYr\r]r_rcrdrfryrrrrrrrrrrrrrrrrrrrrrrrrrErrrrrrrrrrr rZ value_methodsZdtype_conversion_methodsmethodrrZ magic_methodZ origin_methodimplr/r0r<rr=monkey_patch_values $     9 *        Y  %/-..  #    E,            !"#&+ 1 7 = D J P VZ^bf l rvz~   !"$     rG)' __future__rrrrV functoolsrtypingrrrrxrZpaddle.base.libpaddlerZpaddle.base.wrapped_decoratorrZpaddle.utils.decorator_utilsrr Zpaddle._typingr r r r?ZBOOLZUINT8ZINT8ZINT16ZINT32rrrZSUPPORT_PROMOTION_OPSr?r2rMrGr<r<r<r=s