+ jDbaRdt10t^RIt^RIt^RIt^RIt^RIHtHt^RIH t ^RI H t ^RI H t Ht^RIHt^RIt^RIHt.ReOt!RR]4tRtRtRtRt!RR 4t!RR 4t]!R .RfO4t!RR ]4tRRltRRlt /RR]!3bRR ]!3bR!R!]PD3bR"R"]PD3bR#R#]!3bR$R%]#3bR&R']#3bR(R)]#3bR*R*]PD3bR+R+]!3bR,R-]!3bR.R/]!3bR0R0]!3bR1R1]!3bR2R2]!3bR3R3]!3bR4R4]!3b/R5R6]#3bR7R8]3bR9R9]3bR:R;]!3bRR?]3bR@RA]3bRBRC]!3bRDRE]!3bRFRG]!3bRHRI]3bRJRK]3bRLRM]3bRNRO]!3bRPRQ]!3bRRRS]!3bRTRU]!3bCRVRW] 3/Ct$]^kRXt%!RYR ]4t&!RZR]4t'!R[R]4t(!R\R4t)R]t*R^t+R_t,R`t-RgRalt.RhRblt/Rct0R#)iN) defaultdict namedtuple)Callable) attrgetter)Any NamedTuple) deprecated) DeviceType EventList EventMetadataFormattedTimesMixinIntervalKernel FunctionEventFunctionEventAvg StringTable MemRecordsAcccaa]tRt^toRtV3RltRtRtRtRt Rt ] R4t RR lt R tR tV3R lR ltRRltRtRtVtV;t#)r ar A list of profiling events with helper methods for analysis and visualization. EventList extends the standard Python list to provide specialized methods for working with profiling events (FunctionEvent or FunctionEventAvg objects). It includes utilities for aggregating statistics, formatting output tables, and exporting profiling data. This class is typically returned by profiler methods and should not be instantiated directly by users. Args: *args: Standard list arguments. use_device (str, optional): Device type for profiling ("cuda", "xpu", etc.). profile_memory (bool, optional): Whether memory profiling was enabled. Default: False. with_flops (bool, optional): Whether to include FLOP counts. Default: False. Attributes: _use_device (str): Device type being profiled. _profile_memory (bool): Whether memory profiling is enabled. _with_flops (bool): Whether FLOP counting is enabled. _tree_built (bool): Whether the event tree structure has been built. Key Methods: table(...): Format events as a table string for display. export_chrome_trace(path): Export to Chrome tracing format. export_stacks(path, metric): Export stack traces with metrics. key_averages(...): Compute averaged statistics grouped by operation name. total_average(): Compute aggregate totals across all events (sums, not averages). Properties: self_cpu_time_total: Sum of self CPU time across all events. Example:: import torch from torch.profiler import profile, ProfilerActivity with profile(activities=[ProfilerActivity.CPU]) as prof: x = torch.randn(100, 100) y = torch.matmul(x, x) # EventList is returned by prof.events() events = prof.events() # Display as formatted table print( events.table( sort_by="cpu_time_total", row_limit=20, top_level_events_only=False ) ) # Export to Chrome tracing format events.export_chrome_trace("trace.json") # Get averaged statistics avg_events = events.key_averages() print(avg_events.table()) # Export stack traces events.export_stacks("stacks.txt", "self_cpu_time_total") See Also: - :class:`FunctionEvent`: Individual profiling event - :class:`FunctionEventAvg`: Averaged profiling statistics - :meth:`table`: Format events as a readable table - :meth:`key_averages`: Aggregate events by operation name c<VPRR4pVPRR4pVPRR4p\SV`!V/VBW0nW@nRVnWPnR#) use_deviceNprofile_memoryF with_flops)popsuper__init__ _use_device_profile_memory _tree_built _with_flops)selfargskwargsrrr __class__s&*, ]/home/luca/screen-checking/.venv/lib/python3.14/site-packages/torch/autograd/profiler_util.pyrEventList.__init__bs`ZZ d3 $4e<ZZ e4  $)&)%- %ctVP4VP4VP4RVnR#TN)_populate_cpu_children_remove_dup_nodes_set_backward_stacktracesrr s&r$ _build_treeEventList._build_treems. ##%   &&(r&c"VP4#N)tabler,s&r$__str__EventList.__str__sszz|r&c\4p\\V44EFpW,PfKW,PPW,P8XgKM\W,PP 4^8XgKyW,P W,PnW,P W,PnW,P FpW,PVnK VPV4EK \V4^8XdR#\V4UUu.uFwrEWA9gK VNK pppVP4VPV4EKuuppir() setrangelen cpu_parentname cpu_childrenkernelsadd enumerateclearextend)r to_deleteidxchindevnew_evtss& r$r*EventList._remove_dup_nodesvsISY'I((4 ,,11TY^^CDI00==>!C8< 8N8NDI((53793D3DDI((0"i44(, (<(< 5MM#&(9~"*3D/R/wsS=Q/HR JJL KK ! Ss F F c VUu.uF:pVP'dKVP\P8XgK8VNK< pp\ V\ R4R7p\ P!VRR7pVEFwrV\ VRR7p.pVFp \V4^8dVR,p V PPV PP8g0V PPV PP8dVP4KV PV 4V Pe\RV P 24hV P#V 4VP%V 4K EK R#uupi)aPopulate child events into each underlying FunctionEvent object. One event is a child of another if [s1, e1) is inside [s2, e2). Where s1 and e1 would be start and end of the child event's interval. And s2 and e2 start and end of the parent event's interval Example: In event list [[0, 10], [1, 3], [3, 4]] would make [0, 10] be a parent of two other intervals. If for any reason two intervals intersect only partially, this function will not record a parent child relationship between then. threadkeyc2VPVP3#r0)rHnode_idevents&r$2EventList._populate_cpu_children..su||U]]&Cr&c\VPPVPP).#r0) time_rangestartendrMs&r$rOrPs$5#3#3#9#9E.bw_parents*{a  00r&Nz1Expected fwd_thread to be set for backward parent)stack sequence_nrrH fwd_threadr]get)r fwd_stacksr`tprms& @r$r+#EventList._set_backward_stacktracess 1 C~%#))*?__cjj1&$'IIJM  C#A}<<'(K]]ALL1&NN1b1 r&c&\RV44#)c38"TFqPxK R#5ir0)self_cpu_time_total.0rNs& r$ 0EventList.self_cpu_time_total..s?$,,$)sumr,s&r$ryEventList.self_cpu_time_totals?$???r&c  V\VVVVVVVVPVPVVR7 #)a/Print an EventList as a nicely formatted table. Args: sort_by (str, optional): Attribute used to sort entries. By default they are printed in the same order as they were registered. Valid keys include: ``cpu_time``, ``cuda_time``, ``xpu_time``, ``cpu_time_total``, ``cuda_time_total``, ``xpu_time_total``, ``cpu_memory_usage``, ``cuda_memory_usage``, ``xpu_memory_usage``, ``self_cpu_memory_usage``, ``self_cuda_memory_usage``, ``self_xpu_memory_usage``, ``count``. top_level_events_only(bool, optional): Boolean flag to determine the selection of events to display. If true, the profiler will only display events at top level like top-level invocation of python `lstm`, python `add` or other functions, nested events like low-level cpu/cuda/xpu ops events are omitted for profiler result readability. time_unit(str, optional): A time unit to be used for all values in the table. Valid options are: ``s``, ``ms`` and ``us``. Returns: A string containing the table. ) sort_by row_limitmax_src_column_widthmax_name_column_widthmax_shapes_column_widthheaderrrtop_level_events_only time_unit) _build_tablerr) r rrrrrrrrs &&&&&&&&&r$r1EventList.tables?@ !5"7$;//''"7  r&c ^RIpVP'gRM VPp\VR4;_uu_4p^pVPR4VEFpVPfK\ P !VP4pTPRPTVPPVPP4VP'g VPMRVP RVP R244VPFKpVPR V R VPP R VP R V R V R2 4V^, pKM EK \V4^8dBVP!VP#4^, VP$4VP'4VPR4RRR4R# +'giR#;i)zExport an EventList as a Chrome tracing tools file. The checkpoint can be later loaded and inspected under ``chrome://tracing`` URL. Args: path (str): Path where the trace will be written. Ncudaw[za{{"name": {}, "ph": "X", "ts": {}, "dur": {}, "tid": {}, "pid": "CPU functions", "args": {{}}}}, z " node_id:z , thread_id:z "z {"name": z, "ph": "s", "ts": z , "tid": z , "pid": "CPU functions", "id": z, "cat": "cpu_to_z", "args": {}}, ])osropenwrite trace_namejsondumpsformatrRrS elapsed_us is_remoterHrLr;r7seektellSEEK_SETtruncate) r pathr device_namefnext_idr` name_json_s && r$export_chrome_traceEventList.export_chrome_trace s $($4$4$4f$:J:J $__G GGCL>>) JJs~~6 ' (.v!,,113"}}} )#++l3::,bQ ( AGG$YK0!!$!5!5 67""%**.!!( ***57( ( qLG%)F4y1}qvvx!|R[[1 GGCLY___sB GC3G G( c .RO#)ry)ryself_cuda_time_totalself_xpu_time_totalself_privateuse1_time_totalr,s&r$supported_export_stacks_metrics)EventList.supported_export_stacks_metricsYs   r&c&<V^8dQhRS[RS[/#)rmetricstr)r __classdict__s"r$ __annotate__EventList.__annotate__as22#2s2r&c W P49d*\R\VP44,4h\PRR4p\ VR4;_uu_4pVFpVP 'gK\ VP 4^8gK3\VVPRR4PRR4PRR44p\V4^8gKRp\VP 4F#pWxPV4, pVR , pK% VR R R ,\\V44,pVPVR ,4K R R R 4R # +'giR #;i)zmetric should be one of: z ; ____rrdevicexpu privateuse1;N  rU) r ValueErrorr maketransrror7getattrreplaceintreversed translater) r rrtranslate_tablerr` metric_value stack_strentrys &&& r$ export_stacksEventList.export_stacksas+ ==? ?+d::<=> --&9 $__999SYY!!3#*vx8 1 9 $L <(1,$& %-cii%8E%)III%,I&9%.crNS$83sV^8dQhR\\R3,/#)rreturn.)tupler)rs"r$r,EventList.key_averages..__annotate__s   38_ r&c<\VP4\VP4\VP4\VP4\VP 4.pV'dVP SP4V'd%VP \VP44V^8dW@PRV, p\V4#rN) rrJrLrW is_legacyis_user_annotationr_ overload_name input_shapesror)rNgroup_by_input_shapesgroup_by_stack_ngroup_by_overload_namerJr`s&&&& r$get_key'EventList.key_averages..get_keysEIIEMM"E%%&EOO$E,,- C& 3,,-$ 3u1123!#{{#4$455: r&rrrNr)rr]rris_python_functionr<r valuesrrrrorr) r rrrinclude_python_functionsstatsravg_listr`s &&&&& @r$ key_averagesEventList.key_averagesys> G :EEU9V  $C%%%.F .BX c#h LLN''//''   C "3#34CI(#% ))$&!  r&c X\4pVFpW, pRVnK RVnV#)aCompute aggregate statistics across all events. Accumulates statistics from all events into a single FunctionEventAvg object. This is primarily useful for computing total metrics (total CPU time, total memory usage, etc.) across the entire profiling session, regardless of operation type. Note: This sums up times and counts across ALL different operations, so the "average" metrics (like cpu_time) represent the average time per operation call across the entire session, mixing all operation types together. For per-operation averages, use :meth:`key_averages` instead. Returns: FunctionEventAvg: A single aggregate object with key="Total" containing accumulated statistics. NTotal)rrJ)r total_statr`s& r$ total_averageEventList.total_averages4&&' C  J!JN! r&)rrrr)NdK7PNFN)FrFF)__name__ __module__ __qualname____firstlineno____doc__rr-r2r*r)r+propertyryr1rrrrr__static_attributes____classdictcell__ __classcell__)r#rs@@r$r r snBH & "0D-L24@@, \7r 220L\r&c^RpRpW8d W, R R2#W8d W, R R2#VR R2#)+Define how to format time in FunctionEvent.@@.3fsmsus.Ar)time_us US_IN_SECONDUS_IN_MSs& r$ _format_timersM"LH(-Q//$S),,c]" r&cfV^8XdV^8wd\RV 24hR#VR,V, R R2#)rzExpected time_us == 0 but got NaNgY@.2f%)r])r total_time_uss&&r$_format_time_sharers@ a< #A'!KL Lo -c 2! 44r&c&RpRV,pRV,p\V4V8dVR,V, R R2#\V4V8dVR,V, R R2#\V4V8dVR,V, R R2#\V4R,#)z&Return a formatted memory size string.i?rz GBz MBz KBz B)absr)nbytesKBMBGBs& r$_format_memoryrs B B B 6{b3,#C(,, V 3,#C(,, V 3,#C(,,6{T!!r&c"a\V3Rl4#)c.<\\VS44#r0)rr)r r9s&r$rO!_attr_formatter..sgdD.A!Br&)rr9sfr$_attr_formatterrs B CCr&ca]tRtRtoRt]!R4t]!R4t]!R4t]!R4t ]!R4t ]!R4t ] R 4t ] R 4t] ]!R ]R 7R 44tRtVtR#)r i zsHelpers for FunctionEvent and FunctionEventAvg. The subclass should define `*_time_total` and `count` attributes. cpu_time device_timecpu_time_totaldevice_time_totalryself_device_time_totalcpVP^8XdR#RVP,VP, #rgr)countrr,s&r$rFormattedTimesMixin.cpu_times+jjAosQ31D1D+Dtzz+QQr&cpVP^8XdR#RVP,VP, #r)rrr,s&r$rFormattedTimesMixin.device_times+jjAosT31G1G+G$**+TTr&z<`cuda_time` is deprecated, please use `device_time` instead.categorycVP#r0)rr,s&r$ cuda_timeFormattedTimesMixin.cuda_times r&rN)rrrrrr cpu_time_strdevice_time_strcpu_time_total_strdevice_time_total_strself_cpu_time_total_strself_device_time_total_strrrrr FutureWarningr#rrrs@r$r r s #:.L%m4O()9:+,?@-.CD!01I!J RRUUF   r&c,a]tRtRtoRtRtRtVtR#)ri'cWnW nR#r0)rSrT)r rSrTs&&&r$rInterval.__init__(s  r&c <VPVP, #)z$ Returns the length of the interval rTrSr,s&r$rInterval.elapsed_us,sxx$**$$r&r1N)rrrrrrrrr,s@r$rr's%%r&c,a]tRtRtoV3RltRtVtR#)r i6c<V^8dQh/S[R,;R&S[R,;R&S[S[,R,;R&S[S[,R,;R&S[R,;R&S[R,;R&S[R,;R&S[S[S[3,R,;R &S[R,;R &S[R,;R &S[R,;R &S[R,;R &S[R,;R&S[R,;R&S[R,;R&S[R,;R&S[R,;R&S[R,;R&S[R,;R&S[R,;R&S[R,;R&S[R,;R&S[R,;R&S[R,;R&S[R,;R&S[R,;R&S[R,;R&S[R,;R&S[R,;R&S[R,;R&S[R,;R &S[R,;R!&S[R,;R"&S[R,;R#&S[R,;R$&#)%rNregisters_per_thread shared_memorygridblockpriority blocks_per_sm warps_per_sm occupancyest_occupancy_pctqueuedgraph_id graph_node_idstreamcontextchannel channel_typebytesbandwidth_gb_scollective_namedtype in_msg_nelemsout_msg_nelems in_split_sizeout_split_sizeglobal_rank_startglobal_rank_stride group_sizeprocess_group_nameprocess_group_desc group_ranksranksrc_rankdst_rankseqrV)rlistfloatdictrrbool)rrs"r$rEventMetadata.__annotate__6sv*$: s)d   9t   Dj 4<$,CH~$$t|# $JDj: $J4Z 4Z!"*#& :'(DL ),4Z-. :/0:12$J34:56$J78Tz!9:d ";<d =>d "?@d "ABtCD *EFDjGHDjIJ tOKLTkMr&rN)rrrr__annotate_func__rrr,s@r$r r 6sr&c0V^8dQhR\R\/#rvrr)rs"r$rr_sssr&c$VPR4#)")stripr_s&r$_to_strrd_s 773<r&c0V^8dQhR\R\/#r^)rrZ)rs"r$rrcsr&c VR9#)1)rgtruerrcs&r$_to_boolrics r&zregisters per threadr5z shared memoryr6r7r8r9z blocks per SMr:z warps per SMr;zest. achieved occupancy %r=r<r>zgraph idr?z graph node idr@rArBrCrDrEzmemory bandwidth (GB/s)rFzCollective namerGrHz In msg nelemsrIzOut msg nelemsrJz In split sizerKzOut split sizerLzGlobal rank startrMzGlobal rank striderNz Group sizerOzProcess Group NamerPzProcess Group DescriptionrQzProcess Group RanksrRRankrSzSrc RankrTzDst RankrUSeqrVzIs asynchronized oprVc/pRp\P4F/wpwrEVPV4pVeV!V4W&RpK+RW&K1 V'd \R/VB#R#)FNTr)_EVENT_METADATA_KEYSitemsrrr ) extra_metafields any_populated kineto_key field_nameconvertr_s& r$_build_metadataruskFM-A-G-G-I) )Z NN: & =!(F  M!%F  .J'4= "6 "==r&c%a]tRtRtoRtRRRR^R^^RRRR]P ^RRRRRRRRRRRRR^^RRRRRRRR3%RltRtRt Rt ] R 4t ] R 4t ] ]!R ]R 7R 44t] ]!R]R 7R44t] R4t] R4t] R4t] ]!R]R 7R44t] R4t] ]!R]R 7R44t] R4tRtRtVtR#)riaProfiling information about a single function. FunctionEvent records the execution of a single operation during profiling. These events are obtained from the profiler/kineto and contain detailed timing and memory usage information. .. note:: FunctionEvent objects are typically created by the profiler/kineto and should not be instantiated directly by users. Access them through the profiler's output. Attributes: id (int): Unique identifier for this event. node_id (int): Node identifier for distributed profiling (-1 if not applicable). name (str): Name of the profiled function/operator. overload_name (str): Overload name for the operator (requires _ExperimentalConfig(capture_overload_names=True) set). trace_name (str): Same as name, just changes ProfilerStep* to ProfilerStep# time_range (Interval): Time interval containing start and end timestamps in microseconds. thread (int): Thread ID where the operation started. fwd_thread (int): Thread ID of the corresponding forward operation. kernels (List[Kernel]): List of device kernels launched by this operation. count (int): Number of times this event was called (usually 1). cpu_children (List[FunctionEvent]): Direct CPU child operations. cpu_parent (FunctionEvent): Direct CPU parent operation. input_shapes (List[List[int]]): Shapes of input tensors (requires record_shapes=True). For plain tensor inputs, each entry is a list of dimensions (e.g. ``[16, 16]``). TensorList inputs are represented as an empty list ``[]``; use ``structured_input_shapes`` to get per-element shapes for TensorList inputs. concrete_inputs (List[Any]): Concrete input values (requires record_shapes=true). kwinputs (Dict[str, Any]): Keyword arguments (requires record_shapes=true). stack (List[str]): Python stack trace where the operation was called (requires with_stack=true). scope (int): at::RecordScope identifier (0=forward, 1=backward, etc.). use_device (str): Device type being profiled ("cuda", "xpu", etc.). cpu_memory_usage (int): CPU memory allocated in bytes. device_memory_usage (int): Device memory allocated in bytes. is_async (bool): Whether this is an asynchronous operation. is_remote (bool): Whether this operation occurred on a remote node. sequence_nr (int): Sequence number for autograd operations. device_type (DeviceType): Type of device (CPU, CUDA, XPU, PrivateUse1, etc.). device_index (int): Index of the device (e.g., GPU 0, 1, 2). device_resource_id (int): Resource ID on the device (ie. stream ID). is_legacy (bool): Whether this is from the legacy profiler. flops (int): Estimated floating point operations. is_user_annotation (bool): Whether this is a user-annotated region. metadata (Dict[str, Any]): Additional metadata keyed by the field names used in exported traces. Use ``_ExperimentalConfig(expose_kineto_event_metadata=True)`` to expose Kineto activity metadata. Available fields vary by activity and backend. metadata_json (str): Deprecated. Use event_metadata instead. event_metadata (EventMetadata): Additional metadata in structured format. structured_input_shapes (List[List[int] | List[List[int]]]): Like ``input_shapes`` but distinguishes TensorList inputs. Plain tensor inputs are ``List[int]``; TensorList inputs are ``List[List[int]]`` containing one shape per tensor in the list. Matches the ``"Input Dims"`` field in the Chrome trace JSON. structured_input_strides (List[List[int] | List[List[int]]]): Strides of input tensors in the same format as ``structured_input_shapes`` (requires record_shapes=True). Properties: cpu_time_total (float): Total CPU time in microseconds. device_time_total (float): Total device (CUDA/XPU/etc) time in microseconds. self_cpu_time_total (float): CPU time excluding child operations. self_device_time_total (float): Device time excluding child operations. self_cpu_memory_usage (int): CPU memory usage excluding child operations. self_device_memory_usage (int): Device memory usage excluding child operations. cpu_time (float): Average CPU time per call. device_time (float): Average device time per call. key (str): Key used for grouping events (usually same as name). See Also: - :class:`torch.profiler.profile`: Context manager for profiling - :class:`EventList`: List container for FunctionEvent objects with helper methods - :class:`FunctionEventAvg`: Averaged statistics over multiple FunctionEvent objects NFc+WnVVnW nW`nVVn\ WE4VnW0nWpn.Vn ^Vn .Vn RVn Wn VVnVVnWnWnWnWnWnWnWnVVnVVnVVnVfTMTVnVVnVVnVVnVVnVVnRVn RVn!RVn"VVn#VVn$VVn%V Vn&V!Vn'V"Vn(V*Vn)V#'d \UV#4MRVn+V$Vn,V%Vn-V&Vn.V'Vn/V(Vn0V)Vn1R#NrU)2idrLr9rrrrRrHrqr;rr:r8rconcrete_inputskwinputsrorlrcpu_memory_usagedevice_memory_usagerVrrprW device_indexdevice_resource_idrflopsrr activity_typeself_cpu_percenttotal_cpu_percenttotal_device_percent_metadata_jsonflow_id flow_type flow_start external_idlinked_correlation_idmetadataruevent_metadatastructured_input_shapesstructured_input_strides input_dtypes python_idpython_parent_idpython_module_id)+r rzr9rHstart_usend_usrrqrrorlrr}r~rVrrprLrWrrrrrr{r|rrr metadata_jsonrrrrrrorrrrrrtyped_metadatas+&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&r$rFunctionEvent.__init__suZ#  "/)$,X$>! &0%'  1304-9*9(0   &0%5(; & ( +'2!-(0F6H  )!& /A(:)6 "!#$&!+#* %.'1 +*?"/= +5OJ '4 .E$.F%'3'%5%5r&cVP\P8wd \R4hVPP \ WV44R#)Expected device_type to be CPUN)rWr rXr]r;r_r)r r9rdurations&&&&r$ append_kernelFunctionEvent.append_kernelRs9   z~~ - !AB B F4:;r&c &VP\P8wd \R4h\ V\ 4'g \R4hVP\P8wd \R4hVP PV4R#)zAppend a CPU child of type FunctionEvent. One is supposed to append only direct children to the event to have correct self cpu time being reported. rz$Expected child to be a FunctionEventz$Expected child device_type to be CPUN)rWr rXr] isinstancerr:r_)r childs&&r$r\FunctionEvent.append_cpu_childWsm   z~~ - !AB B%// !GH H     . !GH H   'r&c VP\P8wd \R4h\ V\ 4'g \R4hVP\P8wd \R4hWnR#)aSet the immediate CPU parent of type FunctionEvent. One profiling FunctionEvent should have only one CPU parent such that the child's range interval is completely inside the parent's. We use this connection to determine the event is from top-level op or not. rz%Expected parent to be a FunctionEventz%Expected parent device_type to be CPUN)rWr rXr]rrr8)r rhs&&r$r^FunctionEvent.set_cpu_parentes^   z~~ - !AB B&-00 !HI I    / !HI I r&cVP'g VP\P8wd^#VP\ RVP 44, #)rc38"TFqPxK R#5ir0)r}r{rs& r$r|6FunctionEvent.self_cpu_memory_usage..zs+ 0Au " "0Ar~)rVrWr rXr}rr:r,s&r$self_cpu_memory_usage#FunctionEvent.self_cpu_memory_usagevsL ===D,, >$$s+ 040A0A+ (   r&cVP'g VP\P8wd^#VP\ RVP 44, #)rc38"TFqPxK R#5ir0)r~rs& r$r|9FunctionEvent.self_device_memory_usage..s. 3D% % %3Dr~)rVrWr rXr~rr:r,s&r$self_device_memory_usage&FunctionEvent.self_device_memory_usage~sL ===D,, >''#. 373D3D. +   r&z<`metadata_json` is deprecated. Use `event_metadata` instead.r cVP#r0)rr,s&r$rFunctionEvent.metadata_jsons """r&zO`self_cuda_memory_usage` is deprecated. Use `self_device_memory_usage` instead.cVP#r0rr,s&r$self_cuda_memory_usage$FunctionEvent.self_cuda_memory_usages ,,,r&cxVP\P8XdVPP 4#^#)r)rWr rXrRrr,s&r$rFunctionEvent.cpu_time_totals*   z~~ -??--/ /r&cVP'g VP\P8wd^#VP\ RVP 44, #)rc38"TFqPxK R#5ir0)rrs& r$r|4FunctionEvent.self_cpu_time_total..s) .?U .?r~)rVrWr rXrrr:r,s&r$ry!FunctionEvent.self_cpu_time_totalsL ===D,, >""S) .2.?.?) &   r&c~VP'gVP'g^#VP\P8XdlVP 'g>\ RVP44\ RVP44,#\ RVP44#VP\P\P\P\P\P39d\RVP 24hVPP!4#)rc38"TFqPxK R#5ir0rr{kinfos& r$r|2FunctionEvent.device_time_total..D|e>>|r~c38"TFqPxK R#5ir0r)r{rBs& r$r|rsK3DR((3Dr~c38"TFqPxK R#5ir0rrs& r$r|rrr~HExpected device_type to be CUDA, PrivateUse1, MTIA, HPU or XPU, but got )rVrrWr rXrrr;r:CUDA PrivateUse1MTIAHPUXPUr]rRrr,s&r$rFunctionEvent.device_time_totals ===   z~~ ->>>Dt||DDsK373D3DKH Dt||DDD&& (%^_c_o_o^pq??--/ /r&zA`cuda_time_total` is deprecated. Use `device_time_total` instead.cVP#r0rr,s&r$cuda_time_totalFunctionEvent.cuda_time_totals %%%r&cVP'gVP'g^#VP\P8Xd.VP \ RVP44, #VP\P\P\P\P\P39d\RVP 24hVP #)rc38"TFqPxK R#5ir0rrs& r$r|7FunctionEvent.self_device_time_total..s05FE''5Fr~r)rVrrWr rXrrr:rrrrrr]r,s&r$r$FunctionEvent.self_device_time_totals ===   z~~ -))C0595F5F0- && (%^_c_o_o^pq)) )r&zK`self_cuda_time_total` is deprecated. Use `self_device_time_total` instead.cVP#r0rr,s&r$r"FunctionEvent.self_cuda_time_totals ***r&cVP#r0rr,s&r$rJFunctionEvent.keys yyr&c VPpVPpVPpRP.RNVP NRNVP NRNVP NRNVP NRNVP NRNVP NRNVPP NRNVPP NR N\VPUu.uFqDPNK up4 NR NV NR NV NRNVP NR NVP NR N\VP 4 NRNVP" NR NV NRNV NRNVP$ NRNVP& NRNVP( NRNVP* NRN4#uupi)rz)rr&r~joinrzr9rrWrLr%rRrSrTrr:rHrr}rVrrpr)r rrr~rs& r$__repr__FunctionEvent.__repr__soo ** "66 y y  y  y y { y/ y$J\J\I] y^ y++, y,5 y6:ll^ yCM yNRN_N_M` ya y--. y.6 y7;7J7J6K yL y t7H7H I7He7H IJK yLM yNYM yZ` yal_l ym yII;  y'  y(,{{m  y4B  yCFdFWFWBXAY  yZ  y !% 5 56  y 78  y 9D}  y ES  y TgRg  y h  y    y '2  y 37..1A  y BJ  y KOJZJZI[  y \g  y hlgugufv  y wx  y !JsF )0rrr{rr:r}r8rr~rrWrrrrrrrqrzrrrVrrrrr;r|rrr9rLrrrrrlrrprorrrHrRrrrrrU)rrrrrr rXrrr\r^rrrr r+rrrryrrrrrJrrrr,s@r$rrsHbNN   $!%Wi6V< ( !"    F#  #Y-  -   002K&  &**(U+  +   r&cHa]tRtRtoRtV3RlRltRtRtRtRt Vt R #) ria Averaged profiling statistics over multiple FunctionEvent objects. FunctionEventAvg aggregates statistics from multiple FunctionEvent objects with the same key (typically same operation name). This is useful for getting average performance metrics across multiple invocations of the same operation. This class is typically created by calling :meth:`EventList.key_averages()` on a profiler's event list. Attributes: key (str): Grouping key for the events (typically operation name). count (int): Total number of events aggregated. node_id (int): Node identifier for distributed profiling (-1 if not applicable). is_async (bool): Whether the operations are asynchronous. is_remote (bool): Whether the operations occurred on a remote node. use_device (str): Device type being profiled ("cuda", "xpu", etc.). cpu_time_total (int): Accumulated total CPU time in microseconds. device_time_total (int): Accumulated total device time in microseconds. self_cpu_time_total (int): Accumulated self CPU time (excluding children) in microseconds. self_device_time_total (int): Accumulated self device time (excluding children) in microseconds. input_shapes (List[List[int]]): Input tensor shapes (requires record_shapes=true). overload_name (str): Operator overload name (requires _ExperimentalConfig(capture_overload_names=True) set). stack (List[str]): Python stack trace where the operation was called (requires with_stack=true). scope (int): at::RecordScope identifier (0=forward, 1=backward, etc.). cpu_memory_usage (int): Accumulated CPU memory usage in bytes. device_memory_usage (int): Accumulated device memory usage in bytes. self_cpu_memory_usage (int): Accumulated self CPU memory usage in bytes. self_device_memory_usage (int): Accumulated self device memory usage in bytes. cpu_children (List[FunctionEvent]): CPU child events. cpu_parent (FunctionEvent): CPU parent event. device_type (DeviceType): Type of device (CPU, CUDA, XPU, PrivateUse1, etc.). is_legacy (bool): Whether from legacy profiler. flops (int): Total floating point operations. is_user_annotation (bool): Whether this is a user-annotated region. Properties: cpu_time (float): Average CPU time per invocation. device_time (float): Average device time per invocation. See Also: - :class:`EventList.key_averages`: Method that creates FunctionEventAvg objects - :class:`FunctionEvent`: Individual profiling event - :class:`EventList`: Container for profiling events c<V^8dQhRR/#)rrNr)rrs"r$rFunctionEventAvg.__annotate__%s$r&cdRVn^Vn^VnRVnRVnRVn^Vn^Vn^Vn^Vn RVn RVn RVn RVn ^Vn^Vn^Vn^VnRVnRVn\(P*VnRVn^VnR#)NF)rJrrLrVrrrrryrrrrorlr}r~rrr:r8r rXrWrrr,s&r$rFunctionEventAvg.__init__%s#  # $&*#$&'() +,#48)-"& !% %&() *+"-.%8<04'1~~$ r&cVPfVPVnVPVnVPVnVPVnVPVnVP VnVP VnVPVnVPVnVPVn VPVn VPVn VPVn VPVn \V\\ 34'g \#R4hVPVP8wd&\#RVP RVP 24hV;P$VP$, unV;P&VP&, unV;P(VP(, unV;P*VP*, unV;P,VP,, unV;P.VP., unV;P0VP0, unV;P2VP2, unV;P4VP4, unVP6fVP6VnV#VP6e$V;P6VP6, unV#)Nz8Expected other to be a FunctionEvent or FunctionEventAvgz Expected keys to match, but got z vs )rJrLrVrr8r:rrrorlrWrrrrrrr]rrryrr}r~rrrrr others&&r$r<FunctionEventAvg.add>s  88 yyDH ==DL!NNDM"__DN#..DO % 2 2D !&!4!4D  % 2 2D DJDJ$00D "__DN#..DO&+&>&>D #%-1A!BCC J  99  2599+T$((L  u333 %"9"99   E$=$==  ##u'C'CC# !7!77   E$=$==  ""e&A&AA" %%)G)GG% ekk! :: DJ [[ $ JJ%++ %J r&c$VPV4#r0)r<rs&&r$__iadd__FunctionEventAvg.__iadd__ksxxr&cHVP'gRM VPpVPpVPpVPpRVP RVP RVP RV RV RV RV R\VP4 RVP RV R V R 2#) rzZ>Z=[[efjfwfwexy M(8'9;-vk]Zhilmqm~m~iiAA $ 5 56a }NS`Raab d r&)rr:r}r8rr~rrWrrrVrrrrJrLrrlrryrrrorN) rrrrrrr<rrrrr,s@r$rrs*+Z2+Z   r&c&a]tRtRtoRtRtVtR#)rizcx\V4^8d \PPV4MTW&W,#)ry)r7torch_C _demangle)r rJs&&r$ __missing__StringTable.__missing__{s.033x!|EHH&&s+ yr&rN)rrrrrrrr,s@r$rrzsr&c0a]tRtRtoRtRtRtRtVtR#)riz=Acceleration structure for accessing mem_records in interval.cWn.Vn.Vn\V4^8dX\ \ V4UUu.uFwr#V^,P 4V3NK upp4p\V!wVnVnR#R#uuppir) _mem_records _start_nses_indicesr7rYr=start_nszip)r mem_recordsirtmps&& r$rMemRecordsAcc.__init__sp'&(#% { a 9[;QR;Q411Q4==?A.;QRSC.13i +D dm Rs$A> c# "\P!VPV4p\P!VPV4p\ W44F)pVP VP V,,xK+ R#5i)z* Return all records in the given interval N)bisect bisect_leftr bisect_rightr6rr)r rend_ns start_idxend_idxrs&&& r$ in_intervalMemRecordsAcc.in_intervals`&&t'7'7B %%d&6&6?y*A##DMM!$45 5+sA=A?)rrrN) rrrrrrr rrr,s@r$rrsG866r&ca.ROp\;QJdV3RlV4F 'dK R# R#!V3RlV44#)autograd/__init__c3l<"TF)q^,S9;'d V^,S9'*xK+ R#5irr)r{rrs& r$r|&_filter_stack_entry..s/O>NaDEM33adem44>Ns44FT))r _make_grads)rbackward)z torch/tensorr)_internal/common_utils prof_callable)rprof_func_call)rprof_meth_call)all)rfiltered_entriessf r$_filter_stack_entryrs8 3O>NO33O3O3O>NO OOr&z[memory]z [OutOfMemory]c.\\RRRRRR.pW9#)z profiler::_record_function_enterz$profiler::_record_function_enter_newzprofiler::_record_function_exitz aten::is_leafzaten::output_nrzaten::_version)MEMORY_EVENT_NAMEOUT_OF_MEMORY_EVENT_NAME)r9filtered_out_namess& r$ _filter_namers/  *.)   %%r&cl\4pW ,pV'dVPR4'dRpV#)z ProfilerStep#z ProfilerStep*)r startswith)r9 with_wildcard string_tables&& r$ _rewrite_namer$s-=L  D ??? + +"D Kr&c aa0a1a2a3a4\V4^8XdR#\;QJdRV4F 'gK RM RM !RV44p \;QJdRV4F 'gK RM RM !RV44p V^,Pp V 'gV 'd \R4h\;QJdRV4F 'gK RM RM !RV44p\;QJdRV4F 'gK RM RM !RV44pSe \ \ VV3R lRR 7V VVR 7p\ R V44^,pVe \VV4p\ RV44^,pVe \VV4p^ pTpR pVUu.uF;pVPfK\VP4^8gK/VPNK= pp\V4^8pV'd*\ RV44^,pVe \VV4pR.pV'dVPR4V.R/O, pV eV P4MRpV 'd#VPRV 2RV R2V R2V R2.4V'd>VPRR.4V 'd#V 'dVPV R2RV R2.4VPR4\;QJdRV4F 'gK RM RM !RV44pV'dVPR4^o0R.o4R.o1S0).o2R0V0V1V2V43RllpRpV!V4V'd V!V4V^V,R F pV!V4K V'dVPR4V!V4V'dVPR 4V!VR!R"7V'drVUu.uF"pVP^8gKVPNK$ pp\V4^8wd2V!\V44wpp VPR#V 24V!V4MRpS4^,p!S1^,p"S2^,p#R p.o3V33R$lp$^p%^p&VFpV%VP, p%VP\P 8Xd(VP"'dV&VP$, p&K\VP\P&\P(\P*\P,39gKVP.'dKV&VP$, p&K VeV$!R%V#,4V$!V4V 'dV$!R%V#,4V$!R&4V$!V"4V$!V!P0!V!4V$!V"4R'p'R(p(^p)VEFDpV)V8XdEM;V 'dVP2eK&V)^, p)VP4p*Ve+\V*4V^, 8dV*R V^, R),p*\7VPV%4VnVP:'g\7VP<V%4M^VnV*.p+V'dEVP@p,Ve+\V,4V^, 8dV,R V^, R),p,V+V,., p+V+VP8V(!VPVPBV 4VP>V(!VP<VPDV 4V(!VPFVPHV 4., p+V 'd\7VP$V&4Vn%V+PV(!VP$VPLV 4VPJV(!VPNVPPV 4V(!VPRVPTV 4.4V'dV+P\WVPX4\WVPZ4.4V 'dBV 'd:V+P\WVP\4\WVP^4.4V+PVP`4V'dV+PVPb4V'd(V+P\eVPf4R V4V'dHVP^8:dV+PR*4M$V+PVPX,R+ 4V'dHRp-\VP4^8dV'!VP^,V4p-V+PV-4V$!V!P0!V+!4V'gEKR.\V4^, ,p.VPR,,F(p/V$!V!P0!V.V'!V/V4.,!4K* V.PR4V$!V!P0!V.!4EKG V$!V"4V$!R-V(!V%\iV%4V 4 24V 'd4T$!RV eV P4MR R.V(!V&\iV&4V 4 24RPkS34#uupiuupi)1zUPrint a summary of events (which can be a list of FunctionEvent or FunctionEventAvg).rc3>"TFqP^8xK R#5irrrzs& r$r|_build_table..sOu66:TFc3>"TFqP^8xK R#5irrrzs& r$r|r'sP77!;r(z9use_device is None, but there is device performance data.c3~"TF3pVPRJ;'d\VP4^8xK5 R#5ir0)rr7rzs& r$r|r's=E   4 ' G GC0B0B,Ca,G G=!=c3~"TF3pVPRJ;'d\VP4^8xK5 R#5ir0)rr7rzs& r$r|r's=E   D ( I IS1D1D-E-I Ir+Nc|<\VSPRR4PRR4PRR44#)rrrr)rr)r`rs&r$rO_build_table..s3OOFH5WUH-W]H5 !r&)rJreverserc3L"TFp\VP4xK R#5ir0)r7rJr{r`s& r$r|r's;FSCLLFs"$c3^"TF#p\\VP44xK% R#5ir0)r7rrr1s& r$r|r's#KFSc#c&6&6"788Fs+-c3F"TFp\RV44xK R#5i)c38"TFp\V4xK R#5ir0r7)r{rs& r$r|)_build_table... s2E5CJJEr~N)max)r{ros& r$r|r' sGu2E222s!Namez Overload NameNonezSelf z %z totalz time avgzCPU Memz Self CPU Memz Memz # of Callsc3>"TFqPR8gxK R#5irx)rLr1s& r$r|r'5s=fs*fr(zNode IDc<S^;;,RV,\V4,R,RS,,, uu&S^;;,RV,RS,,, uu&S^;;,VS,, uu&R#)rz{: }r-Nr)paddingtext_dir SPACING_SIZEheader_sep_lstline_length_lstrow_format_lsts&&r$ add_column _build_table..add_column>shq H s7| +c 1S<5G H  qS7]cL.@AAg 44r&c .ROpV^8:d\RV 24h\^\\P!V4^, \ \ V4^, 444pV^8dV\ V48g\R\ V4 RV 24h\^ \P!V4R,4V\V4,3#)FLOPsz'Expected flops to be positive, but got z&Expected log_flops to be in range [0, z ), but got )rGKFLOPsMFLOPsGFLOPsTFLOPsPFLOPsg) r]r7minmathlog10rXr7powfloorr)r flop_headers log_flopss& r$auto_scale_flops&_build_table..auto_scale_flopsEs  A: #J5'!RS S3tzz%014eC .appendws a dr&=z1This report only display top-level ops statisticsc\V4V8d7\V4V, pWRp\V4^8dRVR,,pV#)N...:NNr5)rsrc_column_widthoffsets&& r$ trim_path_build_table..trim_pathsE t9' 'Y!11F=D4y1}tBx' r&ctRpRpVR8Xd W, R R2#VR8Xd W, R R2#VR8XdVR R2#V#)rrrrrrr)r default_strrrrs&&& r$override_time_unit(_build_table..override_time_units_&   ,S13 3 $ (-R0 0 $ c]"% % r&r\z--z8.3f:ryNNzSelf CPU time total: z time total: )z Self CPU %zSelf CPUz CPU total %z CPU totalz CPU time avg)r)6r7anyr RuntimeErrorr rYr7rMror_upperr?rryrWr rXrrrrrrrrr8rJrrrVrrrr)r'rr%rr*rr(rr&rr}rr~rrrLrrrr)5rbrrrrrrrrrrhas_device_timehas_device_memrhas_input_shapeshas_overload_namesname_column_widthshapes_column_widthDEFAULT_COLUMN_WIDTHflops_column_widthr^r`stacks has_stackheadersrappend_node_idrDrTr raw_flops flops_scale flops_header row_format header_sep line_lengthr_sum_self_cpu_time_totalsum_self_device_time_totalr`rd event_limitr9 row_valuesr src_field empty_headersrr@rArBrXrCs5&f&&&&&&&&& @@@@@r$rrs{  6{acOOcccOOOOSPPSSSPPPN%%J /VWWssss     ")!  ;F;;a?( 13HIKFKKaO*!"57NO-##csyy SYYRSAS  V F aI GG G! K  +"#35IJ hG' G)3(>*""$FK }% }B'-v&-y)     . NN"m4(K=-   NN< S=f=SSS=f==Ny!LTNTN$}oO55W& !$% Q++- .'(/~&&'()#c2*0B&3CIIMYSYY& B y>Q *:3y>*J '[, NNVL>2 3 ) *J"J"J!!$KJ F !"3#:#:: ??jnn , &#*D*D D & OO&&  *** '#*D*D D &!&s[ !vs[ !BC : :  g &' : K ) #  S^^%?  1 Kww ,T>SVW>W1W50146>D1  # #%<  <<< s113J K  V --M%1 &*?!*CC -.K1F1J Lu T =/ )J  '')D)Di   ! ! ""C$:$:I   c..     '9**,F(C $   &2266! ,,&--s/H/H)')<)'>?&s'C'CD   II     ckk *    c#"2"234H5HI J yyA~!!$'!!SYY%.Os!L/r&rIzevent=z node=z stack_trace=>r^~rU)rrsplitrbr5issubsetr_sortr) rbevents_with_tracesrNrrrrlines source_linesr`s & r$_canonicalize_profiler_eventsr/s^ YYvr* &M%%k26 Fm'' r: %0$5$5d$;I$;qwwy$;I#(Q5aA0P5 Q*6l2&B !!joY{eiia0   2 9:E! S&'vc+.>-?}SQ^M_L` a " 99U /JQs E-<E-$E2?E2cV^8dQh/^\9d>\\\\\\.\ 3,3,3,;R&#)rrm)__conditional_annotations__rYrrrr)rs"r$rrs= L$$d3c8SE3J+?&? @@A$Mr&) r r r rrrrrr)r9rr)F) NNrrrrFFFN)2rrrZrrN collectionsrrcollections.abcroperatorrtypingrrtyping_extensionsr rtorch.autogradr __all__rWr rrrrr rrr rdrirloadsrXrmrurrrrrrrrr$rrr)rs@r$rs  /$"( % BBJ5 "D  < % % H< =&J&R $E3S9$Eos+$E VTZZ $E gtzz " $E S! $E ou- $E^U+$E "5u!=$E+tzz*$E xo$ES!$Eos+$E xo$E 3$E 3$E ^S)!$E" gs^#$E$ 0%8%$E&)73'$E( gw )$E*os++$E,'--$E.ow//$E0'11$E2-s33$E4/55$E6<%7$E8/99$E: "6!@;$E<M73=$E> VSM?$E@S!A$EBS!C$ED E3X 'X v @ *@ F+66* P* &&gV )r&