+ "ݳj(^RIt^RIHt^RIHt^RIHt^RIHtH t H t ^RI H t H t ^RIHt^RIHtHt^RIHtR t!R R ]4t!R R ] 4t!RR] 4tR#)N)DTensor) DeviceMesh)ColwiseParallelPrepareModuleInputRowwiseParallel) ScalingType e4m3_dtype)tensor_already_casted_to_fp8)NoopFwToFloat8BwDynamichp_tensor_to_float8_dynamic) GemmInputRolecVP\P8H;'dVP\P8H#)N)scaling_type_inputrDYNAMICscaling_type_grad_output)ms&f/home/luca/screen-checking/.venv/lib/python3.14/site-packages/torchao/float8/float8_tensor_parallel.py(_float8_linear_supports_float8_allgatherr"s;  3 33 > > & &+*=*= =cbaa]tRt^*toRt]R4t]R4tV3RlV3RlltRt Vt V;t #)Float8ColwiseParallelzc Like `ColwiseParallel`, but with all-gather in float8. This currently assumes tensorwise scaling. cTV^,p\V\4'g\P!WTVRR7p\V4'gF\ VVP P PVP\PR7pW8wdVPVRR7pV#rF run_checkgemm_input_roleT placementsasync_op isinstancer from_localr r configcast_config_input target_dtypelinear_mm_configr INPUT redistribute input_layoutsdesired_input_layoutsmodinputs device_mesh input_tensors&&&&& r_prepare_input_fn'Float8ColwiseParallel._prepare_input_fn0s ay ,00"--=EL,L996 ,,99$$ - 3 3 L  1'44045LrcVPV8wdVPVRR7p\P!VVPVP P P4pV'dVP4#T#Tr rr)r applyr'r$cast_config_grad_outputr&to_localoutput_layoutsuse_local_outputr-outputsr/s&&&&&r_prepare_output_fn(Float8ColwiseParallel._prepare_output_fnJsu    /**)D+G *//   JJ . . ; ; &6w!B7BrcT<V^8dQhRS[PRS[RS[P/#moduler/returnnnModuler)format __classdict__s"r __annotate__"Float8ColwiseParallel.__annotate__\* 3 3RYY 3Z 3BII 3rc<^RIHp\W4'g\R\ V4 24h\W4'd\ V4'g \ R4h\SV`!W4#r Float8Linearz.Expecting module to be Float8Linear but found unsupported torchao.float8.float8_linearrOr" ValueErrortyperAssertionErrorsuper_applyselfrBr/rO __class__s&&& rrWFloat8ColwiseParallel._apply\c=&//@fO    :6BB / /w~f22r __name__ __module__ __qualname____firstlineno____doc__ staticmethodr1r=rW__static_attributes____classdictcell__ __classcell__rZrHs@@rrr*sC 2CC" 3 3 3rrcbaa]tRt^ktoRt]R4t]R4tV3RlV3RlltRt Vt V;t #)Float8RowwiseParallelzc Like `RowwiseParallel`, but with all-gather in float8. This currently assumes tensorwise scaling. cTV^,p\V\4'g\P!WTVRR7p\V4'gF\ VVP P PVP\PR7pW8wdVPVRR7pV#rr!r*s&&&&& rr1'Float8RowwiseParallel._prepare_input_fnqsay ,00"--=EL,L996 ,,99$$ - 3 3 L  1'44045LrcVPV8wdVPVRR7p\P!VVPVP P P4pV'dVP4#T#r4r5r9s&&&&&rr=(Float8RowwiseParallel._prepare_output_fnsp    /**nt*TG*//   JJ . . ; ; &6w!B7BrcT<V^8dQhRS[PRS[RS[P/#r@rD)rGrHs"rrI"Float8RowwiseParallel.__annotate__rKrc<^RIHp\W4'g\R\ V4 24h\W4'd\ V4'g \ R4h\SV`!W4#rMrQrXs&&& rrWFloat8RowwiseParallel._applyr\rr]r^rhs@@rrjrjksC .CC" 3 3 3rrjcaa]tRt^toRtRRRRRRRRRRR ]P R R/V3R lltR tV3R lV3Rllt Rt Vt V;t #)PrepareFloat8ModuleInputa Like `PrepareModuleInput`, but with all-gather in float8. This currently assumes tensorwise scaling. The only difference from `PrepareModuleInput` is that after we prepare the input DTensor, we cast the input to DTensor(Float8TrainingTensor) This is to ensure the float8 cast happens before the all-gather (i.e. Shard -> Replicate) so that if there are multiple float8 users of the input activation, we perform fp8 allgather only once. FP8 Args: float8_dtype (torch.dtype, optional): control what float8 dtype to cast to when prepare the module input, we currently only support torch.float8_e4m3fn. default: torch.float8_e4m3fn fwd_config_submodule_fqn (str, optional): the fqn of the submodule that contains the forward config used for the float8 cast. If not specified, we will search for the Float8Linear in the submodules and use the forward config from that module, in this case all module's forward config must be the same. r+Nr,input_kwarg_layoutsdesired_input_kwarg_layoutsr;F float8_dtypefwd_config_submodule_fqnc<\SV`VVVVVR7W`nRVnWpnVP\ P 8wd \R4hR#))r+r,rurvr;NzFPrepareFloat8ModuleInput only support casting to float8_e4m3fn for now)rV__init__rwr'rxtorch float8_e4m3fnNotImplementedError) rYr+r,rurvr;rwrxrZs &$$$$$$$rrz!PrepareFloat8ModuleInput.__init__se '"7 3(C-  ) $(@%    3 3 3%X  4rcVe\V\4'dTpMA\V\P4'gQR4h\P!WV3RR7p\ V\ VP\PR7pVeW48wdVPV3R7pVP'dVP4#T#V#)Nz%expecting input to be a torch.Tensor!Frr)r) r"rr{Tensorr#r r r'r r(r)r;r8)rYinputmesh input_layoutdesired_layoutdt_inps&&&&& r_prepare_input_arg+PrepareFloat8ModuleInput._prepare_input_args  #%))!%66;6!++,E1%% - 3 3 F )l.L,,8I,J(,(=(=(=6??$ I6 ILrcT<V^8dQhRS[PRS[RS[P/#r@rD)rGrHs"rrI%PrepareFloat8ModuleInput.__annotate__s*RYYZBIIrc<^RIHpVPeAVPVP4p\ WC4'gQhVP VnMmVP 4FYp\ WS4'gKVP fVP VnK7VP VP 8XdKTQR4h VP fQh\SV`!W4V#)rrNz?All the Float8Linear modules should have same linear_mm_config!) rRrOrx get_submoduler"r'modulesrVrW)rYrBr/rO fwd_linearr-rZs&&& rrWPrepareFloat8ModuleInput._applys=  ( ( 4--d.K.KLJj77 77$.$?$?D !~~'c00,,4030D0D-#448L8LL]L ($$000 v+ r)rwrxr') r_r`rarbrcr{r|rzrrWrerfrgrhs@@rrtrtso$#  !  %) (("&:6rrt)r{torch.nnrEtorch.distributed._tensorrtorch.distributed.device_meshr!torch.distributed.tensor.parallelrrrtorchao.float8.configrr torchao.float8.distributed_utilsr #torchao.float8.float8_scaling_utilsr r %torchao.float8.float8_training_tensorr rrrjrtr]rrrs^  -4 :I@>3O>3B<3O<3~_1_r