bigscience/petals-api
18
1"""2Utilities for declaring and retrieving active model layers using a shared DHT.3"""4from __future__ import annotations5 6from functools import partial7from typing import Dict, List, Optional, Sequence, Union8 9from hivemind.dht import DHT, DHTNode, DHTValue10from hivemind.moe.client.remote_expert_worker import RemoteExpertWorker11from hivemind.p2p import P2P, PeerID12from hivemind.utils import DHTExpiration, MPFuture, get_dht_time, get_logger, use_hivemind_log_handler13 14import src15from src.data_structures import CHAIN_DELIMITER, UID_DELIMITER, ModuleUID, RemoteModuleInfo16 17use_hivemind_log_handler("in_root_logger")18logger = get_logger(__file__)19 20 21def declare_active_modules(22 dht: DHT,23 uids: Sequence[ModuleUID],24 expiration_time: DHTExpiration,25 throughput: Optional[float] = None,26 wait: bool = True,27) -> Union[Dict[ModuleUID, bool], MPFuture[Dict[ModuleUID, bool]]]:28 """29 Declare that your node serves the specified modules; update timestamps if declared previously30 31 :param uids: a list of module ids to declare32 :param wait: if True, awaits for declaration to finish, otherwise runs in background33 :param throughput: optionally specify your performance in terms of compute throughput34 :param expiration_time: declated modules will be visible for this many seconds35 :returns: if wait, returns store status for every key (True = store succeeded, False = store rejected)36 """37 if isinstance(uids, str):38 uids = [uids]39 if not isinstance(uids, list):40 uids = list(uids)41 for uid in uids:42 assert isinstance(uid, ModuleUID) and UID_DELIMITER in uid and CHAIN_DELIMITER not in uid43 return dht.run_coroutine(44 partial(_declare_active_modules, uids=uids, expiration_time=expiration_time, throughput=throughput),45 return_future=not wait,46 )47 48 49async def _declare_active_modules(50 dht: DHT,51 node: DHTNode,52 uids: List[ModuleUID],53 expiration_time: DHTExpiration,54 throughput: Optional[float] = None,55) -> Dict[ModuleUID, bool]:56 num_workers = len(uids) if dht.num_workers is None else min(len(uids), dht.num_workers)57 return await node.store_many(58 keys=uids,59 subkeys=[dht.peer_id.to_base58()] * len(uids),60 values=[throughput] * len(uids),61 expiration_time=expiration_time,62 num_workers=num_workers,63 )64 65 66def get_remote_module(67 dht: DHT,68 uid_or_uids: Union[ModuleUID, List[ModuleUID]],69 expiration_time: Optional[DHTExpiration] = None,70 return_future: bool = False,71) -> Union[List[Optional[src.RemoteTransformerBlock]], MPFuture[List[Optional[src.RemoteTransformerBlock]]]]:72 """73 :param uid_or_uids: find one or more modules with these ids from across the DHT74 :param expiration_time: if specified, return modules that expire no sooner than this (based on get_dht_time)75 :param return_future: if False (default), return when finished. Otherwise return MPFuture and run in background.76 :returns: a list of [RemoteTransformerBlock if found else None]77 """78 single_uid = isinstance(uid_or_uids, ModuleUID)79 uids = [uid_or_uids] if single_uid else uid_or_uids80 infos = dht.run_coroutine(81 partial(_get_remote_module_infos, uids=uids, expiration_time=expiration_time), return_future82 )83 84 if return_future:85 86 async def _unpack(infos_future: MPFuture, dht: DHT):87 p2p = await dht.replicate_p2p()88 modules = _create_remote_modules_from_infos(await infos_future, p2p)89 return modules[0] if single_uid else modules90 91 return RemoteExpertWorker.run_coroutine(_unpack(infos, dht), return_future)92 p2p = RemoteExpertWorker.run_coroutine(dht.replicate_p2p())93 modules = _create_remote_modules_from_infos(infos, p2p)94 return modules[0] if single_uid else modules95 96 97async def _get_remote_module_infos(98 dht: DHT, node: DHTNode, uids: List[ModuleUID], expiration_time: Optional[DHTExpiration]99) -> List[Optional[RemoteModuleInfo]]:100 if expiration_time is None:101 expiration_time = get_dht_time()102 num_workers = len(uids) if dht.num_workers is None else min(len(uids), dht.num_workers)103 found: Dict[ModuleUID, DHTValue] = await node.get_many(uids, expiration_time, num_workers=num_workers)104 105 modules: List[Optional[RemoteModuleInfo]] = [None] * len(uids)106 for i, uid in enumerate(uids):107 metadata = found[uid]108 if metadata is None or not isinstance(metadata.value, dict):109 if metadata is not None:110 logger.error(f"Incorrect metadata for {uid}: {metadata}")111 continue112 valid_entries = set()113 for maybe_peer_id, _unused_value in metadata.value.items():114 try:115 valid_entries.add(PeerID.from_base58(maybe_peer_id))116 except:117 logger.error(f"Incorrect peer entry for {uid}: {maybe_peer_id}")118 if valid_entries:119 modules[i] = RemoteModuleInfo(uid, valid_entries)120 return modules121 122 123def _create_remote_modules_from_infos(124 infos: Sequence[Optional[RemoteModuleInfo]], p2p: P2P125) -> List[Optional[src.RemoteTransformerBlock]]:126 modules: List[Optional[src.RemoteTransformerBlock]] = []127 for info in infos:128 if info is not None:129 modules.append(src.RemoteTransformerBlock(info, p2p))130 else:131 modules.append(None)132 return modules133 