+ &j2^RIt^RIHtHt]],]P,tR.t RRlt RRlt RR lt ]'d^R I HtM]t!R R]4tR#) N)_is_tensorpipe_available constantsTensorPipeRpcBackendOptionscDV^8dQhR\R\P/#)devicereturn) DeviceTypetorchr)formats"u/Users/jameslopez/projects/CWCArchive/cwc-podcast/.venv/lib/python3.14/site-packages/torch/distributed/rpc/options.py __annotate__r szellc\P!V4pVPR8wd\RVP R24hV#)cudazA`set_devices` expect a list of CUDA devices, but got device type .)r rtype ValueError)rs&r _to_devicer sE \\& !F {{f !;;-q *   MrcV^8dQhR\\\3,R\\P\P3,/#)r device_mapr )dictr r r)r s"r rrs8Z+, %,, $%rc /p/pVP4FXwr4\P!V4\P!V4rCWB9d\RV RW$, RV 24hWAV&W2V&KZ V#)z9`device_map` only supports 1-to-1 mapping, trying to map  and  to )itemsr rr)rfull_device_map reverse_mapkvs& r _to_device_mapr!s9;O46K  "||A Q1 !!"5(8QCA A# rcpV^8dQhR\\,R\\P,/#)rdevicesr )listr r r)r s"r rr's(**T*-*$u||2D*rc4\\\V44#N)r$maprr#s&r _to_device_listr)'s J( ))r) _TensorPipeRpcBackendOptionsBasecaa]tRt^2toRtR]P R]PR]PRRRRRRR R/V3R lV3R lllt V3R lV3R llt V3RlRlt Rt Vt V;t#)ra The backend options for :class:`~torch.distributed.rpc.TensorPipeAgent`, derived from :class:`~torch.distributed.rpc.RpcBackendOptions`. Args: num_worker_threads (int, optional): The number of threads in the thread-pool used by :class:`~torch.distributed.rpc.TensorPipeAgent` to execute requests (default: 16). rpc_timeout (float, optional): The default timeout, in seconds, for RPC requests (default: 60 seconds). If the RPC has not completed in this timeframe, an exception indicating so will be raised. Callers can override this timeout for individual RPCs in :meth:`~torch.distributed.rpc.rpc_sync` and :meth:`~torch.distributed.rpc.rpc_async` if necessary. init_method (str, optional): The URL to initialize the distributed store used for rendezvous. It takes any value accepted for the same argument of :meth:`~torch.distributed.init_process_group` (default: ``env://``). device_maps (Dict[str, Dict], optional): Device placement mappings from this worker to the callee. Key is the callee worker name and value the dictionary (``Dict`` of ``int``, ``str``, or ``torch.device``) that maps this worker's devices to the callee worker's devices. (default: ``None``) devices (List[int, str, or ``torch.device``], optional): all local CUDA devices used by RPC agent. By Default, it will be initialized to all local devices from its own ``device_maps`` and corresponding devices from its peers' ``device_maps``. When processing CUDA RPC requests, the agent will properly synchronize CUDA streams for all devices in this ``List``. num_worker_threads rpc_timeout init_method device_mapsNr# _transports _channelsc<V^8dQhRS[RS[RS[RS[S[S[S[S[3,3,R,RS[S[,R,RS[R,RS[R,/#) rr,r-r.r/Nr#r0r1)intfloatstrrr r$)r __classdict__s"r r(TensorPipeRpcBackendOptions.__annotate__Ts~        #tJ $:;;3D3D3FG3F41!^A&&3FG  ")2og6N         HsAc<<V^8dQhRS[RS[S[S[3,/#)rtor)r5rr )r r6s"r rr7os%858585$z:7M2N85rc  <\V4p\SV`pW9d^VP4FIwrVWTV,9gKWdV,V,8wgK*\ RV RV RWA,V, 24h \SV`W4R#)a Set device mapping between each RPC caller and callee pair. This function can be called multiple times to incrementally add device placement configurations. Args: to (str): Callee name. device_map (Dict of int, str, or torch.device): Device placement mappings from this worker to the callee. This map must be invertible. Example: >>> # xdoctest: +SKIP("distributed") >>> # both workers >>> def add(x, y): >>> print(x) # tensor([1., 1.], device='cuda:1') >>> return x + y, (x + y).to(2) >>> >>> # on worker 0 >>> options = TensorPipeRpcBackendOptions( >>> num_worker_threads=8, >>> device_maps={"worker1": {0: 1}} >>> # maps worker0's cuda:0 to worker1's cuda:1 >>> ) >>> options.set_device_map("worker1", {1: 2}) >>> # maps worker0's cuda:1 to worker1's cuda:2 >>> >>> rpc.init_rpc( >>> "worker0", >>> rank=0, >>> world_size=2, >>> backend=rpc.BackendType.TENSORPIPE, >>> rpc_backend_options=options >>> ) >>> >>> x = torch.ones(2) >>> rets = rpc.rpc_sync("worker1", add, args=(x.to(0), 1)) >>> # The first argument will be moved to cuda:1 on worker1. When >>> # sending the return value back, it will follow the invert of >>> # the device map, and hence will be moved back to cuda:0 and >>> # cuda:1 on worker0 >>> print(rets[0]) # tensor([2., 2.], device='cuda:0') >>> print(rets[1]) # tensor([2., 2.], device='cuda:1') z=`set_device_map` only supports 1-to-1 mapping, trying to map rrN)r!r9r/rr_set_device_map)r;rArrcurr_device_mapsrr r>s&&& r set_device_map*TensorPipeRpcBackendOptions.set_device_maposZ)4 7. !'--/,,r6J16M1M$##$#T!E2B2Fq2I1JL0 4rc0<V^8dQhRS[S[,/#)rr#)r$r )r r6s"r rr7s 0 04 #3 0rc &\V4VnR#)a* Set local devices used by the TensorPipe RPC agent. When processing CUDA RPC requests, the TensorPipe RPC agent will properly synchronize CUDA streams for all devices in this ``List``. Args: devices (List of int, str, or torch.device): local devices used by the TensorPipe RPC agent. N)r)r#)r;r#s&&r set_devices'TensorPipeRpcBackendOptions.set_devicess'w/ rr()__name__ __module__ __qualname____firstlineno____doc__ rpc_contantsDEFAULT_NUM_WORKER_THREADSDEFAULT_RPC_TIMEOUT_SECDEFAULT_INIT_METHODr:rErI__static_attributes____classdictcell__ __classcell__)r>r6s@@r rr2sB #/"I"I *AA  (;;  GK  ,0 $( "&  68585t 0 0 0r)r rrrPr3r5rr __all__rr!r)torch._C._distributed_rpcr*objectrrr r\sS A3Y % ( )"*J'-$A0"BA0r