| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | |
| 4 | from __future__ import annotations |
| 5 | |
| 6 | import importlib |
| 7 | |
| 8 | from torch import nn |
| 9 | |
| 10 | from vllm.config import ModelConfig, VllmConfig |
| 11 | from vllm.config.load import LoadConfig |
| 12 | from vllm.model_executor.model_loader.base_loader import BaseModelLoader |
| 13 | from vllm.tracing import instrument |
| 14 | |
| 15 | _MODELEXPRESS_LOADER_MODULE = "modelexpress.engines.vllm.loader" |
| 16 | _MISSING_MODELEXPRESS_MODULES = frozenset( |
| 17 | { |
| 18 | "modelexpress", |
| 19 | "modelexpress.engines", |
| 20 | "modelexpress.engines.vllm", |
| 21 | _MODELEXPRESS_LOADER_MODULE, |
| 22 | } |
| 23 | ) |
| 24 | |
| 25 | |
| 26 | def _missing_modelexpress_error() -> ImportError: |
| 27 | return ImportError( |
| 28 | "The 'modelexpress' load format requires the ModelExpress Python package. " |
| 29 | "Install it with `pip install modelexpress`." |
| 30 | ) |
| 31 | |
| 32 | |
| 33 | class ModelExpressModelLoader(BaseModelLoader): |
| 34 | """Thin vLLM loader wrapper for ModelExpress.""" |
| 35 | |
| 36 | def __init__(self, load_config: LoadConfig): |
| 37 | super().__init__(load_config) |
| 38 | self._loader = self._load_modelexpress_loader(load_config) |
| 39 | |
| 40 | @staticmethod |
| 41 | def _load_modelexpress_loader(load_config: LoadConfig) -> BaseModelLoader: |
| 42 | try: |
| 43 | module = importlib.import_module(_MODELEXPRESS_LOADER_MODULE) |
| 44 | except ModuleNotFoundError as exc: |
| 45 | if exc.name not in _MISSING_MODELEXPRESS_MODULES: |
| 46 | raise |
| 47 | raise _missing_modelexpress_error() from exc |
| 48 | |
| 49 | ModelExpressVllmLoader = module.MxModelLoader |
| 50 | return ModelExpressVllmLoader(load_config) |
| 51 | |
| 52 | def download_model(self, model_config: ModelConfig) -> None: |
| 53 | self._loader.download_model(model_config) |
| 54 | |
| 55 | def load_weights(self, model: nn.Module, model_config: ModelConfig) -> None: |
| 56 | self._loader.load_weights(model, model_config) |
| 57 | |
| 58 | @instrument(span_name="Load model") |
| 59 | def load_model( |
| 60 | self, |
| 61 | vllm_config: VllmConfig, |
| 62 | model_config: ModelConfig, |
| 63 | prefix: str = "", |
| 64 | ) -> nn.Module: |
| 65 | model = self._loader.load_model( |
| 66 | vllm_config=vllm_config, |
| 67 | model_config=model_config, |
| 68 | prefix=prefix, |
| 69 | ) |
| 70 | return model.eval() |