+ j(l^RIt^RIt^RIHt^RIt^RIHt^RIHtH t ^RI H t R.t !RR4t R#)NCallable)register_multi_grad_hook)register_module_forward_hook register_module_forward_pre_hook) tree_flatten ModTrackerca]tRt^toRtRtRt]R4tRt RV3RlRllt R t R t R t R tR tRtRtRtV3RltRtVtR#)r a ``ModTracker`` is a context manager that tracks the nn.Module hierarchy during execution so that other system can query which Module is currently being executed (or its backward is being executed). You can access the ``parents`` attribute on this context manager to get the set of all the Modules currently being executed via their fqn (fully qualified name, also used as the key within the state_dict). You can access the ``is_bw`` attribute to know if you are currently running in backward or not. Note that ``parents`` is never empty and always contains the "Global" key. The ``is_bw`` flag will remain ``True`` after the forward until another Module is executed. If you need it to be more accurate, please submit an issue requesting this. Adding a map from fqn to the module instance is possible but not done yet, please submit an issue requesting this if you need it. Example usage .. code-block:: python mod = torch.nn.Linear(2, 2) with ModTracker() as tracker: # Access anything during the forward pass def my_linear(m1, m2, bias): print(f"Current modules: {tracker.parents}") return torch.mm(m1, m2.t()) + bias torch.nn.functional.linear = my_linear mod(torch.rand(2, 2)) cR0Vn/Vn\P!4Vn\P !4VnRVn.VnRVn RVn RVn RVn R#GlobalFN) parents_active_module_cntweakrefWeakKeyDictionary_known_modulesWeakSet _seen_modules _has_callback_post_bw_callbacks_to_enqueue_user_pre_fw_hook_user_post_fw_hook_user_pre_bw_hook_user_post_bw_hookselfs&e/home/luca/screen-checking/.venv/lib/python3.14/site-packages/torch/distributed/_tools/mod_tracker.py__init__ModTracker.__init__9sc z "$9@9R9R9T.5oo.?"=?*!%"&!%"&caSP'dR#\SP4F6p\PP P PV4K8 SPP4V3Rlp\PP P PV4RSnR#)Nc&<R0SnRSnR#r )rrrsrcallback7ModTracker._maybe_set_engine_callback..callbackNs$:DL!&D r T) rreversedrtorchautogradVariable_execution_enginequeue_callbackclear)rpost_bw_callbackr#sf r_maybe_set_engine_callback%ModTracker._maybe_set_engine_callbackEs      ()K)K L  NN # # 5 5 D DEU V!M **002 ' 11@@J!r c D\PP4R8g#)zP A boolean marking if this is currently running during the backward pass or not )r&_C_current_graph_task_idrs&ris_bwModTracker.is_bwUs xx..0B66r c :VPPVR4#)z_ Return the fqn for the given module if it is known to the ``ModTracker``, otherwise ``None``. N)rget)rmods&&r get_known_fqnModTracker.get_known_fqn\s""&&sD11r Nc j<V^8dQhRS[R,RS[R,RS[R,RS[R,/#) pre_fw_hookN post_fw_hook pre_bw_hook post_bw_hookr)format __classdict__s"r __annotate__ModTracker.__annotate__bsC3 3 _3 o3 _ 3 o 3 r c RpV!WPR4VnV!W PR4VnV!W0PR4VnV!W@PR4VnR#)a Registers user-specified hooks to be called before/after the forward/backward pass for each module tracked by the ``ModTracker``. One or more can be ``None``. Args: pre_fw_hook (Callable, optional): A hook to be called before the forward pass for the module. It should have the following signature: pre_fw_hook (module, input) -> None post_fw_hook (Callable, optional): A hook to be called after the forward pass for the module. It should have the following signature: post_fw_hook (module, input, output) -> None pre_bw_hook (Callable, optional): A multi-grad hook to be called on all the outputs of the module that require gradients. It should have the following signature: pre_bw_hook (module, grad_output) -> None post_bw_hook (Callable, optional): A multi-grad hook to be called on all the inputs of the module that require gradients. It should have the following signature: post_bw_hook (module, grad_input) -> None Raises: AssertionError: If a new hook is provided when one is already registered. Note: If the module is not alive during the backward pass, the pre_bw_hook and post_bw_hook will receive None as the module argument. The module fqn will be present in the ``parents`` attribute when each of the hooks is called. Hooks are intended to be used as markers only not to modify the inputs/outputs. c4VeVe\RV R24hV#)Nz Only one zq can be registered at a time Clear the existing hook by calling ``clear_user_hooks`` before registering a new one)AssertionError)hook user_hook hook_names&&&rset_hook0ModTracker.register_user_hooks..set_hooks4I$9$ {+lmKr r<r=r>r?Nrrrr)rr<r=r>r?rJs&&&&& rregister_user_hooksModTracker.register_user_hooksbsp@ "* //" #+ 11># "* //" #+ 11># r c >RVnRVnRVnRVnR#)zI Clears the user specified hooks registered with ``register_user_hooks`` NrLrs&rclear_user_hooksModTracker.clear_user_hookss$"&"&!%"&r cbWP9d#\V4PVPV&VPV,pWP9dZVP 4F*wr4V RV 2VPV&VP V4K, VPP V4V#).)rtype__name__rnamed_children _get_mod_nameadd)rr7mod_namenamesubmods&& rrWModTracker._get_mod_names )) )'+Cy'9'9D   $&&s+ (( ( # 2 2 4 19 !D6.B##F+""6*!5    " "3 'r c aaaaVVVV3RlpV#)c<S'dSP4SSP9d;SP'g)RRlpV\n\P !R^R7SSP9d,^SP S&SPPS4MSP S;;,^, uu&SPe$S'dSPS!4V4R#R#R#)Nc2V RV RVP RV R2#):z: z )rU)msgcategoryfilenamelinenolines&&&&&rcustom_formatwarningCModTracker._get_append_fn..fn..custom_formatwarnings(&Zq83D3D2ERuCPPr zbThe module hierarchy tracking maybe be messed up. Please file a bug to PyTorch, if it is the case.) stacklevelN) r-rr3warnings formatwarningwarnrrXr)argsrfr3rZrw_mods* rfn%ModTracker._get_append_fn..fns//1t||#DJJJQ*>& H  4<<'01''-   &''-2-%%1e&&uw57<1r rrnrZr3rosffff r_get_append_fnModTracker._get_append_fns 6 60 r c aaaaVVVV3RlpV#)cj<SPe S'dSPS!4V4SSP9dWSPS;;,^,uu&SPS,^8XdSPPS4R#R#SP'g \ R4hR#)Nz?The Module hierarchy tracking is wrong. Report a bug to PyTorch)rrrremover3 RuntimeError)rmr3rZrrns*rro"ModTracker._get_pop_fn..fns&&2u''6t||#''-2-**40A5LL''-6ZZZ#U r rqrrsffff r _get_pop_fnModTracker._get_pop_fns   r c\PPP4'dR#VP V4p\ P !V4pVPWCR4!4VPeVPW4\V4wrVVUu.uF;p\V\P4'gK%VP'gK9VNK= ppVP'gSV'd\WPWCR44R#VP P#VPWCR44R#R#uupi)NFT)r&_dynamo eval_frame_is_in_optimized_modulerWrrefrsrr isinstanceTensor requires_gradr3rrzrappend) rr7inputrZrnrm_atensorss &&& r _fw_pre_hookModTracker._fw_pre_hooks == # # ; ; = = !!#& C  E/1  ! ! -  " "3 .u%"VdjELL&A1aoo11dVzzz(2B2B5PT2UV2299$$U$7 Ws"EEEcJ\PPP4'dR#VP V4p\ P !V4pVPeVPWV4VPWTR4!4\V4wrgVUu.uF;p\V\P4'gK%VP'gK9VNK= p pVP'g*V 'd \WPWTR4RR7R#R#R#uupi)NFTany)mode)r&r}r~rrWrrrrzrrrrr3rrs) rr7routputrZrnrmrrrs &&&& r _fw_post_hookModTracker._fw_post_hooks == # # ; ; = = !!#& C   " " .  # #C 7 e,.v&"VdjELL&A1aoo11dVzzzg $,,U$?e &zWs "D D D cr\VP4Vn\VPRR7VnV#)T) always_call)rr_fw_pre_handlerr_fw_post_handlers&r __enter__ModTracker.__enter__s4>t?P?PQ;   D  r cnVPP4VPP4R#ri)rrwr)rrms&*r__exit__ModTracker.__exit__s& ""$ ##%r c6<V^8dQh/S[S[,;R&#)r;r)setstr)r@rAs"rrBrCsDXEr ) rrrrrrrrrrrr)NNNN)rU __module__ __qualname____firstlineno____doc__rr-propertyr3r8rMrPrWrsrzrrrr__annotate_func____static_attributes____classdictcell__)rAs@rr r snD '" 77 2 3 3 j' 6"& &_r )rjrcollections.abcrr&torch.autograd.graphrtorch.nn.modules.modulerrtorch.utils._pytreer__all__r rqr rrs3$ 9- .q&q&r