+ j7 ^RIt^RIt^RIHt^RIHtHt^RIHtH t H t H t ^RI t ^RI Ht^RI Ht^RIHt.tRRltR R ltR R lt] !R ]]]4t] !R] P6]4t] RRl4t] RRl4tRtR/RRlltRRltRRlt ] !R4t!] !R]]]]"]]]4t#] RRl4t$] RRl4t$R t$R!R"lt%R0R#R$llt&R/R%R&llt'R'R(lt(R)R*lt)R+R,lt*R-R.lt+R#)1N) OrderedDict)Callable Container)AnyOptionaloverloadTypeVar)nn)PackedSequencec V^8dQhR\R\R\\\R3,\\R3,3,/#)argskwargsreturn.)rtuplestr)formats"X/home/luca/screen-checking/.venv/lib/python3.14/site-packages/torch/distributed/utils.py __annotate__rs<///s/uU38_eCQTHo5U/V/c.p\V4pVP4F'wrEVPV4VPV4K) \V4\V43#)aJ Turn argument list into separate key list and value list (unpack_kwargs does the opposite). Inspiration: https://github.com/facebookresearch/fairscale/blob/eeb6684/fairscale/internal/containers.py#L70 Usage:: kwarg_keys, flat_args = pack_kwargs(1, 2, a=3, b=4) assert kwarg_keys == ("a", "b") assert flat_args == (1, 2, 3, 4) args, kwargs = unpack_kwargs(kwarg_keys, flat_args) assert args == (1, 2) assert kwargs == {"a": 3, "b": 4} Returns: Tuple[Tuple[Any, ...], Tuple[str, ...]]: The first tuple element gives gives both positional args and kwarg values, where the positional args proceed kwarg values and kwarg values are ordered consistently with the kwarg keys. The second tuple element gives the kwarg keys. The second tuple element's length is at most the first tuple element's length. )listitemsappendr)rr kwarg_keys flat_argskvs*, r _pack_kwargsrsW(J:I !  U:. ..rc V^8dQhR\PR,R\R\R\\\3,/#)r dtypeNrrr)torchr!rr)rs"rrr.sGRR ;; R RR 38_ RrcPaSfW3#RV3Rllp\W14\W243#)z Cast floating point tensors in ``args`` and ``kwargs`` to ``input_dtype``. This respects the existing ``requires_grad`` on the tensors. cXV^8dQhR\PR\P/#)r xrr"Tensor)rs"rr*_cast_forward_inputs..__annotate__;s"5<<ELLrc<\P!V4'dVPS8XdV#VPS4#N)r"is_floating_pointr!to)r%r!s&rcast_fn%_cast_forward_inputs..cast_fn;s2&&q))QWW-=HttE{r)_apply_to_tensors)r!rrr-sf*, r_cast_forward_inputsr0.s4 }| g ,.?.P QQrc V^8dQhR\\R3,R\\R3,R\\\R3,\\\3,3,/#)r r.rr)rrrdict)rs"rrrDsN  S#X ,1#s(O  5c?DcN *+ rc \V4\V48d$\R\V4 R\V4 24h\V4^8XdV/3#VR\V4)p\\W\V4)R44pW#3#)zSee _pack_kwargs.ztoo many keys z vs. N)lenAssertionErrorr2zip)rrrrs&& r_unpack_kwargsr7Ds :Y'~c*o->eC NCSTUU :!"} 'J' (D #jS_,<,>"?@ AF <rSTc rV^8dQhR\R\PR\R\\,/#r inputs target_device!use_side_stream_for_tensor_copiesr)r8r"deviceboolr)rs"rrrVs4 #llOS !WrcR#r*r<r=r>s&&&r _recursive_torDUsrc rV^8dQhR\R\PR\R\\,/#r;)r9r"r?r@r)rs"rrr\s4 #llOS 1XrcR#r*rBrCs&&&rrDrD[src@aaaVVV3RloS!V4pRoV# Roi;i)z-Recursively moves input to the target_device.c<aa\S\P\34'Ed\S\4'dSPP M SP pVS 8XdS3#S 'gSP S 43#VPR8XdSP S 43#^RIH pV!S 4pV;_uu_4SP S 4pRRR4\PPS P4;_uu_4\PP4pVPV4\S\4'dXPPV4M<\X\P4'g \!R4hVPV4RRR4V3#^RIHpV 3RlpV!S4'dV!SSV3Rl4#\S\&4'd\)S4^8dV!SS\&4#\S\*4'd\)S4^8dV!SS\*4#\S\,4'dG\)S4^8d7\+SP/44oV!SSP14VV3Rl4#S.# +'giEL;i +'giX3#;i)cpu) _get_streamNzoutput must be a torch.Tensor_is_namedtuplec<\\SV44p\;QJdRV4F 'dK RM RM !RV44'dpVUu.uF qD^,NK pp\;QJd%R\WQ44F 'dK RM RM!R\WQ444'dV.#V!V4.#\V!Uu.uF qb!V4NK up#uupiuupi)z9Handle container types with object identity preservation.c3>"TFp\V4^8HxK R#5i)N)r4).0ms& r K_recursive_to..to_map.._handle_container..s/13q6Q;sFTc3,"TF wrWJxK R#5ir*rB)rPtos& rrRrSsE*D$!qv*Ds)rmapallr6)objelementsmake_containermappedrQ transformedrto_maps&&& r_handle_container8_recursive_to..to_map.._handle_containers#fh/0Fs//sss////-34VttV 43E#k*DE333E#k*DEEE5L&{34458&\B\TN4(\B B 5 Cs C C#c"<\S4!V!#r*)type)r%rYs&r/_recursive_to..to_map..s cArc:<\S4!\SV44#r*)rbr6)rkeysrYs&rrcrdsT#Ys4|-Dr) isinstancer"r'r datar?r,rbtorch.nn.parallel._functionsrJ accelerator device_indexindexcurrent_stream wait_stream record_streamr5 torch.nn.parallel.scatter_gatherrLrr4rr2rfvalues) rYr?rJstreamoutputrmrLr_rfr=r^r>s f @rr^_recursive_to..to_mapds cELL.9 : :(23(G(GSXX__SZZF&v 4}-//;;%'FF=133D%]3V VVM2F&&33M4G4GHH%*%6%6%E%E%GN"..v6"#~66 11.A)&%,,??"01P"QQ,,^<Iy C C #  $S#/FG G c5 ! !c#hl$S#u5 5 c4 SX\$S#t4 4 c4 SX\ #D$SZZ\#D u YVVIHy sJ% BJ9% J6 9 K NrB)r<r=r>resr^s&ff @rrDrDas*=@Vn Jsc@V^8dQhR\R\R\RR/#)r condsraise_assertion_errorrN)rrr@)rs"rrrs($$C$C$$$rcV'g6\V4\P!4V'd \V4hR#R#)zwAlternate to ``assert`` when in the backward context to print the error message ``s`` since otherwise, it is swallowed.N)print traceback print_stackr5)rwrxrys&&&r _p_assertr~s0  a  # # ! rc\V^8dQhR\PR\PRR/#)r tensorsizerN)r"r'Size)rs"rrrs)??5<<?uzz?d?rc\P!4;_uu_4\PPP 4'gVP 4P 4VP48HpV'g[VP 4P 4p\V^8HR4VP 4PVP44RRR4R# +'giR#;i)z Allocate storage for ``tensor`` with the given size. Returns: bool: ``True`` if this method allocated storage and ``False`` if the storage was already allocated. zCTensor storage should have been resized to be 0 but got PLACEHOLDERN) r"no_grad distributed_functional_collectivesis_torchdynamo_compiling_typed_storage_sizenumelr~_resize_)rralready_allocatedtensor_storage_sizes&& r_alloc_storagers   88QQSS & 5 5 7 = = ?4::< O $&,&;&;&=&C&C&E#'1,Y%%'00> sA%C(AC(( C9 c8V^8dQhR\P/#)r rr&)rs"rrrs44%,,4rc \P!4;_uu_4\PPP 4'gVP 4P 4^8HpV'gz\VP4^8HRVP4 RVP 4P 4 RVP 24VP 4P^4RRR4R# +'giR#;i)z Frees the underlying storage of ``tensor``. Returns: bool: ``True`` if the method freed the storage and ``False`` if the storage was already freed. zVFreeing a tensor's storage is unsafe when it is not the sole occupant storage offset: z storage size: z tensor shape: N) r"rrrrrrr~storage_offsetshaper)r already_freeds& r _free_storagers   88QQSS"11399;q@M ))+q0''-'<'<'>&?@%%+%:%:%<%B%B%D$EF%%+\\N4%%'003 sAC96A9C99 D QRcV^8dQhR\\P.\3,R\PR\/#r fn containerr)rr"r'r)rs"rrrs8  %,,"# 05   rcR#r*rBrrs&&rr/r/s rctV^8dQhR\\P.\3,R\R\/#r)rr"r'rr)rs"rrrs'PP(ELL>3#67PAP!PrcR#r*rBrs&&rr/r/sMPrc$aaVV3RloS!V4#)zFRecursively apply to all tensor in different kinds of container types.c <^RIHp\V\P4'd S !V4#\ VR4'dv\ P!V4p\ P!V4Uu/uF(q3PS !\W#P44bK* pp\ P!V3/VB#\V\4'd6VP4pVP4FwrgS !V4WV&K V#\V\4'dS !VP4V#\V\ 4'd,VP4UUu/uFwrgVS !V4bK upp#V!V4'dV 3RlV4p\#V4!V!#\V\$\&\(34'd\#V4!V 3RlV44#V#uupiuuppi)rrK__dataclass_fields__c34<"TF pS!V4xK R#5ir*rBrPelapplys& rrR3_apply_to_tensors..apply..s)q599qc34<"TF pS!V4xK R#5ir*rBrs& rrRrs1q599qr)rprLrgr"r'hasattr dataclassesreplacefieldsnamegetattrr __class__rr rhr2rbrrset) r%rLdcfchangesodkeyvaluerurrs & rr _apply_to_tensors..applysyC a & &a5L Q. / /$$Q'B * * !&&MH 4 89 B *#Cu% B B A  )q)C7C= D%- . .71q11 1H)Cs 0.GGrB)rrrsf&@rr/r/s:  rc*V^8dQhR\\R3,R\\\3,R,R\P R\ R\\\R3,\\\\3,R3,3,/#)r r<.rNr=r>r)rrr2rr"r?r@)rs"rrrsw44 #s(O4 cNT !4<<4(, 4  5c?E$sCx.#"56 67 4rc V'd \WV4M.pV'd \WV4M.p\V4\V48dETP\\V4\V4, 4Uu.uFpRNK up4M\\V4\V48dDTP\\V4\V4, 4Uu.uFp/NK up4\ V4\ V43#uupiuupi)NrB)rDr4extendranger)r<rr=r> moved_inputs moved_kwargs_s&&&& r _to_kwargsrs  f-NO   f-NO   <3|,,s R#)a Replace all keys that match a given old_prefix with a new_prefix (in-place). Usage:: state_dict = {"layer.xyz": torch.tensor(1)} replace_by_prefix_(state_dict, "layer.", "module.layer.") assert state_dict == {"module.layer.xyz": torch.tensor(1)} z*old_prefix and new_prefix must be distinctN) ValueErrorrrf startswithr4)rrrrnew_keys&&& r_replace_by_prefixrXsaEFFJOO%&~~j)) 3z?#455(o  O 'rcDV^8dQhR\PR\/#)r rr)r"r'r@)rs"rrrps33 33rcDVP4P4^8#)r)untyped_storagedata_ptr)rs&r_data_ptr_allocatedrps  ! ! # , , . 22rcV^8dQhR\\P,R\\P,/#)r modulesr)rr r)rs"rrrts*tBII4 ?rc.pVUu/uFq"\VP44bK ppVFTpRpVP4F wr&WBJ;'dWF9pV'gKRpM V'gKCVPV4KV V#uupi)z Returns the modules in ``modules`` that are root modules (i.e. parent-less) with respect to the set ``modules``. In other words, these are the modules in ``modules`` that are the not child of any other module in ``modules``. TF)rrrr)r root_modulesrmodule_to_modulescandidate_moduleis_root_module_modulesis_child_modules& r_get_root_modulesrts%'L4;:4;&FNN$%%G:$ 1 7 7 9 F .OO3C3O !& !: >    0 1$ :s"B)Tr*),rr| collectionsrcollections.abcrrtypingrrrr r"torch.distributedrrr torch.nn.utils.rnnr __all__rr0r7r2rrr8r'r9rDr~rrrrrr/rrrrrrrrBrrrs#/33 - /:R,  CtU# C~.      GT$?(4, CL CtUCncJ     P P F4.PW:  03r