| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | """Factory for weight transfer engines with lazy loading.""" |
| 4 | |
| 5 | import importlib |
| 6 | from collections.abc import Callable |
| 7 | from typing import TYPE_CHECKING |
| 8 | |
| 9 | from vllm.distributed.weight_transfer.base import ( |
| 10 | TrainerWeightTransferEngine, |
| 11 | WeightTransferEngine, |
| 12 | ) |
| 13 | from vllm.logger import init_logger |
| 14 | |
| 15 | if TYPE_CHECKING: |
| 16 | import torch |
| 17 | |
| 18 | from vllm.config import VllmConfig |
| 19 | from vllm.config.weight_transfer import WeightTransferConfig |
| 20 | from vllm.distributed.weight_transfer.base import ( |
| 21 | TrainerInitInfo, |
| 22 | VLLMWeightSyncClient, |
| 23 | WeightSource, |
| 24 | ) |
| 25 | |
| 26 | logger = init_logger(__name__) |
| 27 | |
| 28 | |
| 29 | class WeightTransferEngineFactory: |
| 30 | """Factory for creating weight transfer engines with lazy loading. |
| 31 | |
| 32 | This factory implements a registry pattern that supports: |
| 33 | - Lazy loading: Engine modules are only imported when actually needed |
| 34 | - Extensibility: Custom engines can be registered at runtime |
| 35 | - Centralized registration: All built-in engines registered in one place |
| 36 | """ |
| 37 | |
| 38 | _registry: dict[str, Callable[[], type[WeightTransferEngine]]] = {} |
| 39 | |
| 40 | @classmethod |
| 41 | def register_engine( |
| 42 | cls, |
| 43 | name: str, |
| 44 | module_path_or_cls: str | type[WeightTransferEngine], |
| 45 | class_name: str | None = None, |
| 46 | ) -> None: |
| 47 | """Register an engine with lazy-loading or direct class reference. |
| 48 | |
| 49 | Supports two calling conventions: |
| 50 | 1. Lazy loading: register_engine(name, module_path, class_name) |
| 51 | 2. Direct class: register_engine(name, engine_cls) |
| 52 | |
| 53 | Args: |
| 54 | name: The name to register the engine under (e.g., "nccl") |
| 55 | module_path_or_cls: Either a module path string for lazy loading, |
| 56 | or the engine class directly |
| 57 | class_name: Name of the engine class (required if module_path is string) |
| 58 | |
| 59 | Raises: |
| 60 | ValueError: If an engine with the same name is already registered |
| 61 | |
| 62 | """ |
| 63 | if name in cls._registry: |
| 64 | raise ValueError(f"Weight transfer engine '{name}' is already registered.") |
| 65 | |
| 66 | if isinstance(module_path_or_cls, str): |
| 67 | # Lazy loading path |
| 68 | module_path = module_path_or_cls |
| 69 | if class_name is None: |
| 70 | raise ValueError( |
| 71 | "class_name is required when registering with module path" |
| 72 | ) |
| 73 | |
| 74 | def loader() -> type[WeightTransferEngine]: |
| 75 | module = importlib.import_module(module_path) |
| 76 | return getattr(module, class_name) |
| 77 | |
| 78 | cls._registry[name] = loader |
| 79 | else: |
| 80 | # Direct class registration |
| 81 | engine_cls = module_path_or_cls |
| 82 | cls._registry[name] = lambda: engine_cls |
| 83 | |
| 84 | @classmethod |
| 85 | def create_engine( |
| 86 | cls, |
| 87 | config: "WeightTransferConfig", |
| 88 | vllm_config: "VllmConfig", |
| 89 | device: "torch.device", |
| 90 | model: "torch.nn.Module", |
| 91 | ) -> WeightTransferEngine: |
| 92 | """Create a weight transfer engine instance. |
| 93 | |
| 94 | Args: |
| 95 | config: Weight transfer configuration containing the backend name |
| 96 | vllm_config: The full vLLM config (provides parallel/model config) |
| 97 | device: The device this worker's model lives on |
| 98 | model: The local model instance which will receive the weights |
| 99 | |
| 100 | Returns: |
| 101 | An initialized weight transfer engine instance |
| 102 | |
| 103 | Raises: |
| 104 | ValueError: If the backend is not registered |
| 105 | |
| 106 | """ |
| 107 | backend = config.backend |
| 108 | if backend not in cls._registry: |
| 109 | available = list(cls._registry.keys()) |
| 110 | raise ValueError( |
| 111 | f"Invalid weight transfer backend: {backend}. " |
| 112 | f"Available engines: {available}" |
| 113 | ) |
| 114 | engine_cls = cls._registry[backend]() |
| 115 | |
| 116 | logger.info( |
| 117 | "Creating weight transfer engine: %s", |
| 118 | engine_cls.__name__, |
| 119 | ) |
| 120 | |
| 121 | return engine_cls(config, vllm_config, device, model) |
| 122 | |
| 123 | |
| 124 | class WeightTransferTrainerFactory: |
| 125 | """Factory for creating trainer-side weight transfer engines. |
| 126 | |
| 127 | Parallel to `WeightTransferEngineFactory`, with its own lazy-import |
| 128 | registry. The trainer-side and worker-side registries are kept separate: |
| 129 | they share backend names by convention, but the trainer process never |
| 130 | instantiates a worker engine and vice versa, so unifying them would only |
| 131 | couple the import graphs. |
| 132 | """ |
| 133 | |
| 134 | _registry: dict[str, Callable[[], type[TrainerWeightTransferEngine]]] = {} |
| 135 | |
| 136 | @classmethod |
| 137 | def register_engine( |
| 138 | cls, |
| 139 | name: str, |
| 140 | module_path_or_cls: "str | type[TrainerWeightTransferEngine]", |
| 141 | class_name: str | None = None, |
| 142 | ) -> None: |
| 143 | """Register a trainer engine. Same conventions as |
| 144 | `WeightTransferEngineFactory.register_engine`.""" |
| 145 | if name in cls._registry: |
| 146 | raise ValueError( |
| 147 | f"Weight transfer trainer engine '{name}' is already registered." |
| 148 | ) |
| 149 | |
| 150 | if isinstance(module_path_or_cls, str): |
| 151 | module_path = module_path_or_cls |
| 152 | if class_name is None: |
| 153 | raise ValueError( |
| 154 | "class_name is required when registering with module path" |
| 155 | ) |
| 156 | |
| 157 | def loader() -> type[TrainerWeightTransferEngine]: |
| 158 | module = importlib.import_module(module_path) |
| 159 | return getattr(module, class_name) |
| 160 | |
| 161 | cls._registry[name] = loader |
| 162 | else: |
| 163 | engine_cls = module_path_or_cls |
| 164 | cls._registry[name] = lambda: engine_cls |
| 165 | |
| 166 | @classmethod |
| 167 | def trainer_init( |
| 168 | cls, |
| 169 | init_info: "TrainerInitInfo", |
| 170 | *, |
| 171 | client: "VLLMWeightSyncClient", |
| 172 | source: "WeightSource | None" = None, |
| 173 | ) -> TrainerWeightTransferEngine: |
| 174 | """Build and rendezvous a ready-to-send trainer engine. |
| 175 | |
| 176 | Called on every trainer rank (multi-rank trainers construct on all |
| 177 | ranks; the sender is resolved inside the engine's ``trainer_init``). |
| 178 | |
| 179 | The trainer side takes no `WeightTransferConfig` and no separate |
| 180 | `backend` argument: the backend is read from ``init_info.backend`` (a |
| 181 | `ClassVar` on each `TrainerInitInfo` subclass), and the static wire |
| 182 | params ride `init_info`. |
| 183 | |
| 184 | Args: |
| 185 | init_info: Backend-specific trainer init info. Its `backend` |
| 186 | selects the engine; it also carries the wire params (e.g. |
| 187 | `packed`). |
| 188 | client: Inference-side control-plane client. |
| 189 | source: `WeightSource` of `(name, tensor)` pairs to send each round, |
| 190 | for full-resync backends (NCCL, IPC). Sparse backend |
| 191 | omits it and passes its per-round payload to `send_weights`. |
| 192 | |
| 193 | Raises: |
| 194 | ValueError: If `init_info.backend` is not registered. |
| 195 | |
| 196 | """ |
| 197 | backend = init_info.backend |
| 198 | if backend not in cls._registry: |
| 199 | available = list(cls._registry.keys()) |
| 200 | raise ValueError( |
| 201 | f"Invalid weight transfer backend: {backend}. " |
| 202 | f"Available trainer engines: {available}" |
| 203 | ) |
| 204 | engine_cls = cls._registry[backend]() |
| 205 | |
| 206 | logger.info( |
| 207 | "Creating weight transfer trainer engine: %s", |
| 208 | engine_cls.__name__, |
| 209 | ) |
| 210 | |
| 211 | return engine_cls.trainer_init( |
| 212 | init_info=init_info, |
| 213 | client=client, |
| 214 | source=source, |
| 215 | ) |
| 216 | |
| 217 | |
| 218 | # Register built-in weight transfer engines here. |
| 219 | # Registration should be centralized to ensure lazy loading - |
| 220 | # engine modules are only imported when actually used. |
| 221 | |
| 222 | WeightTransferEngineFactory.register_engine( |
| 223 | "nccl", |
| 224 | "vllm.distributed.weight_transfer.nccl_engine", |
| 225 | "NCCLWeightTransferEngine", |
| 226 | ) |
| 227 | |
| 228 | WeightTransferEngineFactory.register_engine( |
| 229 | "ipc", |
| 230 | "vllm.distributed.weight_transfer.ipc_engine", |
| 231 | "IPCWeightTransferEngine", |
| 232 | ) |
| 233 | |
| 234 | WeightTransferEngineFactory.register_engine( |
| 235 | "sparse_nccl", |
| 236 | "vllm.distributed.weight_transfer.sparse_nccl_engine", |
| 237 | "SparseNCCLWeightTransferEngine", |
| 238 | ) |
| 239 | |
| 240 | WeightTransferEngineFactory.register_engine( |
| 241 | "sharded_rdt", |
| 242 | "vllm.distributed.weight_transfer.sharded_rdt_engine", |
| 243 | "ShardedRDTWeightTransferEngine", |
| 244 | ) |
| 245 | |
| 246 | WeightTransferEngineFactory.register_engine( |
| 247 | "nccl_m2n", |
| 248 | "vllm.distributed.weight_transfer.m2n_engine", |
| 249 | "M2NWeightTransferEngine", |
| 250 | ) |
| 251 | |
| 252 | |
| 253 | WeightTransferEngineFactory.register_engine( |
| 254 | "modelexpress", |
| 255 | "vllm.distributed.weight_transfer.modelexpress_engine", |
| 256 | "ModelExpressWeightTransferEngine", |
| 257 | ) |
| 258 | |
| 259 | |
| 260 | # Trainer-side engines, parallel to the worker registry above. |
| 261 | WeightTransferTrainerFactory.register_engine( |
| 262 | "nccl", |
| 263 | "vllm.distributed.weight_transfer.nccl_engine", |
| 264 | "NCCLTrainerWeightTransferEngine", |
| 265 | ) |
| 266 | |
| 267 | WeightTransferTrainerFactory.register_engine( |
| 268 | "ipc", |
| 269 | "vllm.distributed.weight_transfer.ipc_engine", |
| 270 | "IPCTrainerWeightTransferEngine", |
| 271 | ) |
| 272 | |
| 273 | WeightTransferTrainerFactory.register_engine( |
| 274 | "sparse_nccl", |
| 275 | "vllm.distributed.weight_transfer.sparse_nccl_engine", |
| 276 | "SparseNCCLTrainerWeightTransferEngine", |
| 277 | ) |
| 278 | |
| 279 | WeightTransferTrainerFactory.register_engine( |
| 280 | "sharded_rdt", |
| 281 | "vllm.distributed.weight_transfer.sharded_rdt_trainer", |
| 282 | "ShardedRDTTrainerWeightTransferEngine", |
| 283 | ) |
| 284 | |
| 285 | WeightTransferTrainerFactory.register_engine( |
| 286 | "nccl_m2n", |
| 287 | "vllm.distributed.weight_transfer.m2n_trainer", |
| 288 | "M2NTrainerWeightTransferEngine", |
| 289 | ) |