a HSic@s<UdZddlZddlZddlZddlZddlZddlZddlZddlZddl m Z m Z m Z m Z mZddlmZmZddlmZmZmZddlmZmZmZddlmZddlmZdd lmZej j!Z"ej j#Z$d=d d Z%d>ddZ&Gdddej'jZ(ddZ)ej*+dd Z,ej*+dd Z-ej*+dd Z.ej/d?ddZ0ej1dfddZ2ddZ3ddZ4Gd d!d!e5Z6e7d"d#Z8Gd$d%d%e9Z:e:;ej <d&d'Z=d(d)Z>d*d+Z?dddd,dd defd-d.Z@dZAe e e e feBd/<dddd,dd defd0d1ZCd2d3ZDGd4d5d5eZEGd6d7d7eEZFd8d9ZGd@d;d<ZHdS)AzTracing This module contains functionality to support the JIT's tracing frontend, notably: * torch.jit.trace * torch.jit.trace_module This is not intended to be imported directly; please use the exposed functionalities in `torch.jit`. N)AnyDictListOptionalSet) _python_cu_enabled) ScriptModule_CachedForwardscript)_qualified_name is_scriptingget_callable_argument_names)function)Module)default_tolerancescsfdd}|S)Ncst}|stdd}|dkr@|j}|s6td|d7}q|j}|j}|D]0\}}t|tj rT||urT|dkr||ndSqTdS)Nzfailed to inspect framerrzfailed to get frameself) inspect currentframe RuntimeErrorf_backf_locals f_globalsitems isinstancetorchTensor)varframeirrkv frames_upL/var/www/html/django/DPS/env/lib/python3.9/site-packages/torch/jit/_trace.py_get_interpreter_name_for_var"s  zI_create_interpreter_name_lookup_fn.._get_interpreter_name_for_varr&)r%r(r&r$r'"_create_interpreter_name_lookup_fn!s r)FcCsh|jdd}t|}t}|D]>\}}t||vr:q$|t||rV|||<q$|||<q$|S)NT keep_vars) state_dicttypesetridadddetach)moduler+r,Z filtered_dictZseen_idsr"r#r&r&r'_unique_state_dict9s    r3cs.eZdZdfdd ZejdddZZS) ONNXTracedModuleTFcs0tt|||_||_||_||_||_dSN)superr4__init__innerstrict_force_outplace_return_inputs_return_inputs_states)rr8r9force_outplace return_inputsZreturn_inputs_states __class__r&r'r7Ls zONNXTracedModule.__init__argscst|\ttdd}gggfdd}tj||tjj \}}j r||ddfSj r|ddfS|dfSdS)NTr*csg}ttD]*}t||tjs,td|||qt|}tdd|Dj rtt|j |j rd|fd<t \}}t|dkr|dSt|SdS)NzExpected Tensor argumentcss|]}|jtjdVqdS) memory_formatN)clonerpreserve_format).0xr&r&r' rz.wrapper..rr) rangelenrrrrappend _unflattentupler<r8_flatten)rBin_argsr!Z trace_inputsout_vars_Zin_descin_varsZ inputs_statesoutsZ ret_inputsrr&r'wrapperhs$   z)ONNXTracedModule.forward..wrapperr) rPlistr3valuesr_C_create_graph_by_tracingr)r9r:r;r<)rrBZ module_staterWgraphoutr&rTr'forward^s$ zONNXTracedModule.forward)TFFF)__name__ __module__ __qualname__r7rrr^ __classcell__r&r&r?r'r4Ks r4cs$fddtjdddd|S)Ncsh|dur dSt|tjrV|j|jr*dntjd|j}|j durR|j |_ |S|jtjdSdS)NrC) rrrr1rE is_mkldnnrFrequires_grad_ requires_gradgrad)ar# clone_inputr&r'ris   z"_clone_inputs..clone_inputcSs t|tjSr5rrrrHr&r&r'rJz_clone_inputs..tensors) condition_msg)r _nested_maprAr&rhr' _clone_inputss   rpZPYTORCH_JIT_TIMEZPYTORCH_JIT_DISABLEZPYTORCH_JIT_STATSTccsts|rtjsdVdStj}tjjdd}tjjdd}||z4dVW|||td ||| |n,|||td ||| |0dS)NT) enable_timingz{} {} time: {} ms) _JIT_TIMErcuda is_availablecurrent_streamEvent record_event synchronizeprintformat elapsed_time) trace_namenametimestreamstartendr&r&r'_times    rc sttjjstdttt|ts0|f}t|}rJt  }d fdd }tj j |dd.||dd\}}j |sJWd n1s0Yr|||dd \} } t|| t|| d S) aB Verify that a JIT compiled model has the same behavior as its uncompiled version along with its backwards pass. If your model returns multiple outputs, you must also specify a `loss_fn` to produce a loss for which the backwards will be computed. This function has side-effects (e.g., it executes your model / saves and loads parameters), so don't expect the model to come out exactly the same as what you passed in. Args: model (compiled torch.nn.Module or function): the module/function to be verified. The module/function definition MUST have been decorated with `@torch.jit.compile`. args (tuple or Tensor): the positional arguments to pass to the compiled function/module to be verified. A non-tuple is assumed to be a single positional argument to be passed to the model. loss_fn (function, optional): the loss function to be applied to the output of the model, before backwards is invoked. By default, we assume that a model returns a single result, and we :func:`torch.sum` before calling backwards; if this is inappropriate, you can pass your own loss function. Note that if a model returns a tuple of results, these are passed as separate positional arguments to `loss_fn`. devices (iterable of device IDs, optional): the GPU devices which the compiled module will be run on. This determines the RNG state we must save when running both compiled and uncompiled versions of the model. zICannot verify an uncompiled module. Add @torch.jit.compile to compile itFcsrtng}t||f\}}}|r4||r>|j}|}|r\|j|kr\tdt|tsl|f}tj krt |dkrt d t |t|\} }dd| D} |} tj | g|} dd| D} | | fS)Nz#failed to use the compiled functionrz_Model returns {} outputs, but default loss function (torch.sum) can only handle a single outputcSsg|]}|jtjdqSrCr1rErrFrGr#r&r&r' sz/verify..run_fwd_bwd..cSsg|]}|jtjdqSrrrr&r&r'rs)rX parametersrP clear_cachehitsrrrOrsumrL ValueErrorrzautogradrf)rB force_traceassert_compiledparamsrUrS compiled_fnrr]rRZ saved_outslossgradsZ saved_gradsZ is_moduleloss_fnmodelr&r' run_fwd_bwds:  zverify..run_fwd_bwdztorch.jit.verify)_callerT)rN)r)FF)rrrZZCompiledFunction TypeErrorrrOrpcopydeepcopyr,randomfork_rngZ has_trace_forload_state_dict _verify_equal) rrBrdevicesZ saved_args saved_staterZuncompiled_outsZuncompiled_gradsZ compiled_outsZcompiled_gradsr&rr'verifys&"  !,  rcCs6t||D]&\}}||dkr tdq dS)Ngư>z!JIT and real computation mismatch)zipsubabsmaxr)xsysrHyr&r&r'rsrcCsddd|DS)N cSsg|] }d|qS) r&)rGliner&r&r'rrJzindent..)join splitlines)sr&r&r'indentsrcseZdZdfdd ZZS)TracingCheckErrorNcsd|_|dur |j|d7_|durL|jd7_|jt|d7_|durx|jd7_|jt|d7_tt||jdS)NzTracing failed sanity checks! rz+ERROR: Graphs differed across invocations! zERROR: Tensor-valued Constant nodes differed in value across invocations. This often indicates that the tracer has encountered untraceable code. )messagerr6rr7)rZgraph_diff_errorZtensor_compare_error extra_msgr?r&r'r7#szTracingCheckError.__init__)N)r_r`rar7rbr&r&r?r'r"src sl|D]`}t|tjr|f}|ri} |D]\} } t| | | <q,tjjt|drX|jn|| d|||tj d} | j j |j }t|tjtfr|f}n tjj|t|d|||d} | fddddfd d } dgfd d fd d}| |d}| ||d}|||drF| |d}|||d}tdd|Drt|qdS)N__self__F) check_tracer9r: _module_class_compilation_unit)rr9r:rc stjj}tj|tj|t|}tdd|}tjj}tj|tj|t|}tdd|}d}||krxddl }| | d| d}dt d |d}t||D]\}}t|t|kr|d7}| t| dt| d} dt d | d} |} | rF| d t | d7} |} | rh| d t | d7} || 7}qxqd} t||D]:\}}||krq|d kr|s|s|d sq|d d ks|d d krq|d }|d }ztjj||ddWnttfy}zx| durTd} | dt t|d7} |}|r| dt |d7} | dt t|7} WYd}~qWYd}~n d}~00q|| fS)Nz___torch_mangle_[0-9]+\.rrTz Graph diff: rzFirst diverging operator: z Node diff: zTrace source location: zCheck source location: zprim::Constantvaluet) equal_nanzNode: zSource Location: zComparison exception: )rrZ_jit_pass_canonicalizer\_jit_pass_inline!_jit_pass_erase_shape_informationstrrerdifflibndiffrrrrnodes sourceRangekind mustBeNone hasAttributekindOfrtesting assert_closerAssertionError)Zmod_canonicalizedmod_strZcheck_canonicalizedZ check_strgraph_diff_errorsrZ graph_diffZn_modZn_checkZ node_diffZsource_printoutZ mod_stackZ check_stacktensor_compare_errorsZmod_tensor_valZcheck_tensor_valeZ compare_stack)check_mod_func traced_funcr&r'graph_diagnostic_infobs            *z+_check_trace..graph_diagnostic_infocSst|tr|S|fSr5)rrOrkr&r&r' wrap_retvalsz!_check_trace..wrap_retvalc sz$|t|}dd|D}|WStyz}z>\}}d|dtt|}t|||d|WYd}~n d}~00dS)NcSsg|]}t|tjr|qSr&rj)rGr]r&r&r'rrJzK_check_trace..run_mod_and_filter_tensor_outputs..z+encountered an exception while running the z with test inputs. Exception: )r)rp Exceptionrrr)modinputsZ running_whatrVrrrmsg)rrr&r'!run_mod_and_filter_tensor_outputss z7_check_trace..run_mod_and_filter_tensor_outputscs|dr dSdd<ddjD}t|dkrxd}|d7}|ddd|Ddd 7}|d 7}tj|td d dS) NrTcSsg|]}|r|qSr&)isNondeterministicrGopr&r&r'rszE_check_trace..maybe_warn_nondeterministic..z"Trace had nondeterministic nodes. z2Did you forget call .eval() on your model? Nodes: rcSsg|]}tt|qSr&)rrrr&r&r'rrJzp This may cause errors in trace checking. To disable trace checking, pass check_trace=False to torch.jit.trace()category stacklevel)r\rrLrwarningswarn TracerWarning)Z nondeterm_opsZnondeterministic_ops_warning) has_warnedrr&r'maybe_warn_nondeterministics( z1_check_trace..maybe_warn_nondeterministicc s.d}tt||D]\}\}}z|jr0|}|jr>|}|jrL|}|jrZ|}|sj|rtjj | tj | tj t ||dddn(tjj | | t ||dddWqty&}zFtjdt|dd|dt|tddd }WYd}~qd}~00q|S) NTr)rtolatolrz Output nr zH. of the traced function does not match the corresponding output of the z. Detailed error: rF) enumerater is_quantized dequantizercto_dense is_complexrrrtocdoublerdoublerrrrr)original referenceZ match_whatall_okr!origrefr)check_tolerancerr&r'compare_outputssZ       z%_check_trace..compare_outputstracezPython functionzrepeated tracecss|]}|duVqdSr5r&)rGinfor&r&r'rIrJz_check_trace..)rrrrrpjit trace_modulehasattrrrZCompilationUnit_c _get_methodr}dictranyr) check_inputsfuncrrr9r=Zis_trace_modulerrZ copied_dictr}dataZ check_modrrZ traced_outsZfn_outsZ check_outsZ diag_infor&)rrrrrrrr' _check_trace5sZ    T-   rc@seZdZeddZdS)rcCstjdtdddS)Nignorez torch.(?!jit))rr2)rfilterwarningsrr&r&r&r'ignore_lib_warningssz!TracerWarning.ignore_lib_warningsN)r_r`ra staticmethodrr&r&r&r'rsrcCs,t|tjtfr|fSt|ts(t|S|Sr5)rrrrrO)example_inputsr&r&r' make_tuple)s  r cCsVt|tr|Stj|r:tjjj}tjjj||dddS|durFt }|||dSdS)NFT share_types is_tracingr) rr r _jit_internalmodule_has_exportsr _recursive make_stubs_from_exported_methodscreate_script_moduleTopLevelTracedModule)rrrZinfer_methods_stubs_fnr&r&r' make_module2s   rcCs|dur dSdd|DS)NcSsg|] }d|iqS)r^r&)rGcr&r&r'rHrJz%wrap_check_inputs..r&)rr&r&r'wrap_check_inputsDsrgh㈵>c Csfts|S|durtdt|tjjr6td|St|tjjrdt |d|id|t ||||| St |drt|j tjjr|j dkrt |j d|id|t ||||| St|tjtfr|f}nt|tst|}td} t |drt|j tjjrtdt|} tj| ||| ||t|} |rb|durJt||| |||d|nt|g|| |||d|| S) a+ Trace a function and return an executable or :class:`ScriptFunction` that will be optimized using just-in-time compilation. Tracing is ideal for code that operates only on ``Tensor``\s and lists, dictionaries, and tuples of ``Tensor``\s. Using `torch.jit.trace` and `torch.jit.trace_module`, you can turn an existing module or Python function into a TorchScript :class:`ScriptFunction` or :class:`ScriptModule`. You must provide example inputs, and we run the function, recording the operations performed on all the tensors. * The resulting recording of a standalone function produces `ScriptFunction`. * The resulting recording of `nn.Module.forward` or `nn.Module` produces `ScriptModule`. This module also contains any parameters that the original module had as well. Warning: Tracing only correctly records functions and modules which are not data dependent (e.g., do not have conditionals on data in tensors) and do not have any untracked external dependencies (e.g., perform input/output or access global variables). Tracing only records operations done when the given function is run on the given tensors. Therefore, the returned `ScriptModule` will always run the same traced graph on any input. This has some important implications when your module is expected to run different sets of operations, depending on the input and/or the module state. For example, * Tracing will not record any control-flow like if-statements or loops. When this control-flow is constant across your module, this is fine and it often inlines the control-flow decisions. But sometimes the control-flow is actually part of the model itself. For instance, a recurrent network is a loop over the (possibly dynamic) length of an input sequence. * In the returned :class:`ScriptModule`, operations that have different behaviors in ``training`` and ``eval`` modes will always behave as if it is in the mode it was in during tracing, no matter which mode the `ScriptModule` is in. In cases like these, tracing would not be appropriate and :func:`scripting ` is a better choice. If you trace such models, you may silently get incorrect results on subsequent invocations of the model. The tracer will try to emit warnings when doing something that may cause an incorrect trace to be produced. Args: func (callable or torch.nn.Module): A Python function or `torch.nn.Module` that will be run with `example_inputs`. `func` arguments and return values must be tensors or (possibly nested) tuples that contain tensors. When a module is passed `torch.jit.trace`, only the ``forward`` method is run and traced (see :func:`torch.jit.trace ` for details). example_inputs (tuple or torch.Tensor): A tuple of example inputs that will be passed to the function while tracing. The resulting trace can be run with inputs of different types and shapes assuming the traced operations support those types and shapes. `example_inputs` may also be a single Tensor in which case it is automatically wrapped in a tuple. Keyword arguments: check_trace (``bool``, optional): Check if the same inputs run through traced code produce the same outputs. Default: ``True``. You might want to disable this if, for example, your network contains non- deterministic ops or if you are sure that the network is correct despite a checker failure. check_inputs (list of tuples, optional): A list of tuples of input arguments that should be used to check the trace against what is expected. Each tuple is equivalent to a set of input arguments that would be specified in ``example_inputs``. For best results, pass in a set of checking inputs representative of the space of shapes and types of inputs you expect the network to see. If not specified, the original ``example_inputs`` are used for checking check_tolerance (float, optional): Floating-point comparison tolerance to use in the checker procedure. This can be used to relax the checker strictness in the event that results diverge numerically for a known reason, such as operator fusion. strict (``bool``, optional): run the tracer in a strict mode or not (default: ``True``). Only turn this off when you want the tracer to record your mutable container types (currently ``list``/``dict``) and you are sure that the container you are using in your problem is a ``constant`` structure and does not get used as control flow (if, for) conditions. Returns: If `func` is `nn.Module` or ``forward`` of `nn.Module`, `trace` returns a :class:`ScriptModule` object with a single ``forward`` method containing the traced code. The returned `ScriptModule` will have the same set of sub-modules and parameters as the original ``nn.Module``. If ``func`` is a standalone function, ``trace`` returns `ScriptFunction`. Example (tracing a function): .. testcode:: import torch def foo(x, y): return 2 * x + y # Run `foo` with the provided inputs and record the tensor operations traced_foo = torch.jit.trace(foo, (torch.rand(3), torch.rand(3))) # `traced_foo` can now be run with the TorchScript interpreter or saved # and loaded in a Python-free environment Example (tracing an existing module):: import torch import torch.nn as nn class Net(nn.Module): def __init__(self): super(Net, self).__init__() self.conv = nn.Conv2d(1, 1, 3) def forward(self, x): return self.conv(x) n = Net() example_weight = torch.rand(1, 1, 3, 3) example_forward_input = torch.rand(1, 1, 3, 3) # Trace a specific method and construct `ScriptModule` with # a single `forward` method module = torch.jit.trace(n.forward, example_forward_input) # Trace a module (implicitly traces `forward`) and construct a # `ScriptModule` with a single `forward` method module = torch.jit.trace(n, example_forward_input) N]`optimize` is deprecated and has no effect. Use `with torch.jit.optimized_execution() insteadz`The input to trace is already a ScriptModule, tracing it is a no-op. Returning the object as is.r^rrzVtrace doesn't support compiling individual module's functions. Please use trace_moduleF)rrrrrrr nnrrrrrr_rrrOr)AttributeErrorr rZ_create_function_from_tracerr) rroptimizerrrr9r:rr var_lookup_fnr}Ztracedr&r&r'rKs       r_trace_module_mapc  s\ts|S|durtdtd} t|tjjs8tdt|t sJtdtj j j } zifdd|d<tj j _ |dt ||| } |D]\} }| d kr|}t|| }t|}nt|| }t|}t|}| j| ||| |||| j| }|r|dur$t||||||d |qt|g|||||d |qW| tj j _ n | tj j _ 0| S) a$ Trace a module and return an executable :class:`ScriptModule` that will be optimized using just-in-time compilation. When a module is passed to :func:`torch.jit.trace `, only the ``forward`` method is run and traced. With ``trace_module``, you can specify a dictionary of method names to example inputs to trace (see the ``inputs``) argument below. See :func:`torch.jit.trace ` for more information on tracing. Args: mod (torch.nn.Module): A ``torch.nn.Module`` containing methods whose names are specified in ``inputs``. The given methods will be compiled as a part of a single `ScriptModule`. inputs (dict): A dict containing sample inputs indexed by method names in ``mod``. The inputs will be passed to methods whose names correspond to inputs' keys while tracing. ``{ 'forward' : example_forward_input, 'method2': example_method2_input}`` Keyword arguments: check_trace (``bool``, optional): Check if the same inputs run through traced code produce the same outputs. Default: ``True``. You might want to disable this if, for example, your network contains non- deterministic ops or if you are sure that the network is correct despite a checker failure. check_inputs (list of dicts, optional): A list of dicts of input arguments that should be used to check the trace against what is expected. Each tuple is equivalent to a set of input arguments that would be specified in ``inputs``. For best results, pass in a set of checking inputs representative of the space of shapes and types of inputs you expect the network to see. If not specified, the original ``inputs`` are used for checking check_tolerance (float, optional): Floating-point comparison tolerance to use in the checker procedure. This can be used to relax the checker strictness in the event that results diverge numerically for a known reason, such as operator fusion. Returns: A :class:`ScriptModule` object with a single ``forward`` method containing the traced code. When ``func`` is a ``torch.nn.Module``, the returned :class:`ScriptModule` will have the same set of sub-modules and parameters as ``func``. Example (tracing a module with multiple methods):: import torch import torch.nn as nn class Net(nn.Module): def __init__(self): super(Net, self).__init__() self.conv = nn.Conv2d(1, 1, 3) def forward(self, x): return self.conv(x) def weighted_kernel_sum(self, weight): return weight * self.conv.weight n = Net() example_weight = torch.rand(1, 1, 3, 3) example_forward_input = torch.rand(1, 1, 3, 3) # Trace a specific method and construct `ScriptModule` with # a single `forward` method module = torch.jit.trace(n.forward, example_forward_input) # Trace a module (implicitly traces `forward`) and construct a # `ScriptModule` with a single `forward` method module = torch.jit.trace(n, example_forward_input) # Trace specific methods on a module (specified in `inputs`), constructs # a `ScriptModule` with `forward` and `weighted_kernel_sum` methods inputs = {'forward' : example_forward_input, 'weighted_kernel_sum' : example_weight} module = torch.jit.trace_module(n, inputs) Nrrz.expected torch.nn.Module as the first argumentz3expected a dictionary of (method_name, input) pairscs4|D]&\}}|d|}||<||qdS)N.)named_children)rprefixr}childZsubmod_qualnameregister_submodsZtrace_module_mapr&r'r#s z&trace_module..register_submodsZ__moduler^T)rrrr)rrrrrrr_tracerrrgetattrrr r_create_method_from_tracerr)rrrrrrr9r:rrrZold_module_mapr2 method_namerrZforward_methodargument_namesZcheck_trace_methodr&r"r'rDszV            rcCstr dStjS)z Returns ``True`` in tracing (if a function is called during the tracing of code with ``torch.jit.trace``) and ``False`` otherwise. F)r rrZ _is_tracingr&r&r&r'r sr csReZdZdZdfdd ZddZfddZfd d Zd d Zd dZ Z S) TracedModuleTNc stt|t|tjjs JtGdddtjj}tj t ||_ |}fdd}|j |_ |j D]"\}}|durp||j |<||qp|jD]"\}} | dur| |j|<|| q|jD]4\}} tj| r||j vr||jvrt||| q|jrtdt||jD],\}} | dur:q$t| tdd|j|<q$tjjj|ddd d d } t |j|jd <| |jd <dD]}t||qdS)Nc@s eZdZdS)z.TracedModule.__init__..QualnameWrapperN)r_r`rar&r&r&r'QualnameWrapper sr+cs|vrtd|dS)Nz=TracedModules don't support parameter sharing between modules)rr0)paramid_setr&r' check_uniques z+TracedModule.__init__..check_uniquez=Modules that have backward hooks assigned can't be compiled: r cSsdS)Nr&r&)r2r&r&r'rl9rJz'TracedModule.__init__..FTr _name_actual_script_module) _parameters_buffers_modulestraining)r6r*r7rrrrr.rr r-_jit_override_qualnamer5r2rr3__dict__rZ_jit_is_script_objectsetattr_backward_hooksrrr4rrrrr_delattr) rrr.rr+Z tmp_moduler/r}r,bufval submodule script_moduler?r-r'r7sZ         zTracedModule.__init__cOs tddS)Nz"Trace submodules cannot be called.)r)rrBkwargsr&r&r'r^AszTracedModule.forwardcs&d|jvrtt||St|j|SNr1)r7r6r* __getattr__r%r1)rattrr?r&r'rBDs zTracedModule.__getattr__cs.d|jvrtt|||St|j||dSrA)r7r6r* __setattr__r9r1)rrCrr?r&r'rDIs zTracedModule.__setattr__cCs|jSr5)r0rr&r&r' _get_nameNszTracedModule._get_namecCs d|jS)Nzoriginal_name={})rzr0rEr&r&r' extra_reprQszTracedModule.extra_repr)NN) r_r`ra_disable_script_metar7r^rBrDrFrGrbr&r&r?r'r*sD  r*c@seZdZeZddZdS)rcCs|jd|dS)z Re-construct an instance of TopLevelTracedModule using an instance of a C++ module. Args: cpp_module: The C++ module that this TopLevelTracedModule will be rebuilt around. r1N)r7 _reconstruct)r cpp_moduler&r&r'rIXsz!TopLevelTracedModule._reconstructN)r_r`rar r^rIr&r&r&r'rUsrcs(tfdd_d_S)Ncs,ts|i|Stj}||i|Sr5)r r __original_fn)rBr@rfnrWr&r'rWcs z#_script_if_tracing..wrapperT) functoolswrapsrK__script_if_tracing_wrapper)rMr&rLr'_script_if_tracingbs rQr&cCs:|dur i}t|ts|f}t||||||i|}|S)a .. warning:: This function is internal-only and should only be used by the ONNX exporter. If you are trying to get a graph through tracing, please go through the public API instead:: trace = torch.jit.trace(nn.LSTMCell(), (input, hidden)) trace_graph = trace.graph Trace a function or model, returning a tuple consisting of the both the *trace* of an execution, as well as the original return value. If return_inputs, also returns the trace inputs as part of the tuple Tracing is guaranteed not to change the semantics of the function/module that is traced. Args: f (torch.nn.Module or function): the function or module to be traced. args (tuple or Tensor): the positional arguments to pass to the function/module to be traced. A non-tuple is assumed to be a single positional argument to be passed to the model. kwargs (dict): the keyword arguments to pass to the function/module to be traced. Example (trace a cell): .. testcode:: trace = torch.jit.trace(nn.LSTMCell(), (input, hidden)) N)rrOr4)frBr@r9r:r>r<rVr&r&r'_get_trace_graphrs ! rS)r)F)T)r&NTFFF)I__doc__rros contextlibrNrrrtypingrrrrrZtorch.jit._staterrtorch.jit._scriptr r r torch._jit_internalr r rtorch.autogradrtorch.nnrZtorch.testing._comparisonrrZ _jit_flattenrP_jit_unflattenrNr)r3rr4rpenvirongetrrZ _JIT_DISABLEZ _JIT_STATScontextmanagerrrrrrrrno_gradrWarningrr_tracer_warn_use_pythonr rrrr__annotations__rr r*rrQrSr&r&r&r's      D \ e     w - [