Ei~SrSSKrSSKrSSKrSSKJr SSKrSSKJrJ r J r J r J r J r JrJrJr SSjrSrS\S \4S jrSS \\S \S \\4Sjjrg)z A collection of utilities for ensuring that training can always occur. Heavily influenced by the [toma](https://github.com/BlackHC/toma) library. N)Optional) is_cuda_availableis_hpu_availableis_mlu_availableis_mps_availableis_musa_availableis_neuron_availableis_npu_availableis_sdaa_availableis_xpu_availablec:U(a[R"5 [5(a[RR 5 g[ 5(a[RR 5 g[5(a[RR 5 g[5(a[RR 5 g[5(a[RR 5 g[SS9(a[RR 5 g[!5(a[R"R 5 g[%5(ag['5(a[R(R 5 gg)z Clears the device cache by calling `torch.{backend}.empty_cache`. Can also run `gc.collect()`, but do note that this is a *considerable* slowdown and should be used sparingly. z2.0) min_versionN)gccollectr torchxpu empty_cachermlur sdaar musar npurmpsrcudarr neurongarbage_collections T/home/mande/repo/quber/.venv/lib/python3.13/site-packages/accelerate/utils/memory.pyclear_device_cacher(s                    e ,              " c[U[5(d [U5n[[U55HnSX'M [ SS9 U$)a Releases memory from `objects` by setting them to `None` and calls `gc.collect()` and `torch.cuda.empty_cache()`. Returned objects should be reassigned to the same variables. Args: objects (`Iterable`): An iterable of objects Returns: A list of `None` objects to replace `objects` Example: ```python >>> import torch >>> from accelerate.utils import release_memory >>> a = torch.ones(1000, 1000).cuda() >>> b = torch.ones(1000, 1000).cuda() >>> a, b = release_memory(a, b) ``` NTr) isinstancelistrangelenr)objectsis rrelease_memoryr(FsA, gt $ $w- 3w<  !$/ Nr exceptionreturnc^/SQn[T[5(a.[TR5S:Xa[ U4SjU55$g)z Checks if `exception` relates to CUDA out-of-memory, XPU out-of-memory, CUDNN not supported, or CPU out-of-memory Args: exception (`Exception`): An exception )z out of memory.z(cuDNN error: CUDNN_STATUS_NOT_SUPPORTED.z*DefaultCPUAllocator: can't allocate memoryz1FATAL ERROR :: MODULE:PT_DEVMEM Allocation failedrc3F># UHoTRS;v M g7f)rN)args).0errr)s r +should_reduce_batch_size..ssC{)..++{s!F)r" RuntimeErrorr%r-any)r) _statementss` rshould_reduce_batch_sizer5ds=K )\**s9>>/Ba/GC{CCC r functionstarting_batch_sizereduce_batch_size_fncj^^^Tc[R"[US9$UmTcU4SjmUUU4SjnU$)a A basic decorator that will try to execute `function`. If it fails from exceptions related to out-of-memory or CUDNN, the batch size is multiplied by 0.9 and passed to `function` `function` must take in a `batch_size` parameter as its first argument. Args: function (`callable`, *optional*): A function to wrap starting_batch_size (`int`, *optional*): The batch size to try and fit into memory Example: ```python >>> from accelerate.utils import find_executable_batch_size >>> @find_executable_batch_size(starting_batch_size=128) ... def train(batch_size, model, optimizer): ... ... >>> train(model, optimizer) ``` )r7c$>[TS-5mT$)Ng?)int) batch_sizesrr88find_executable_batch_size..reduce_batch_size_fnsZ#-.J r c @>[SS9 [[R"T5RR 55n[ U5[ U5S-:adSR[USSUSS5VVs/sH up4USU3PM snn5n[STRSTRSUS 35eTS :Xa [S 5eT"T/UQ70UD6$s snnf![a+n[U5(a[SS9 T "5mSnAO eSnAff=fM^) NTrrz, =zBatch size was passed into `zS` as the first argument when called.Remove this as the decorator already does so: `(z)`rz-No executable batch size found, reached zero.)rr#inspect signature parameterskeysr%joinzip TypeError__name__r2 Exceptionr5) r-kwargsparamsargvaluearg_strer<r6r8s r decorator-find_executable_batch_size..decorators3d3g''1<<AACD v;#d)a- (iiCqr TXYZY[T\D] ^D]jcC5%!1D] ^_G.x/@/@.ABBBJBSBSATTUV]U^^`b Q"#RSS    !;Qdee$J#  . r )F)NN)__doc__rRrrAtypingrrimportsrrrrr r r r r rr(rIboolr5callabler;rTr rr\s      #<< d($("/3Ax AA#8,Ar