| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | |
| 4 | import json |
| 5 | import os |
| 6 | import struct |
| 7 | from functools import cache |
| 8 | from os import PathLike |
| 9 | from pathlib import Path |
| 10 | from typing import Any |
| 11 | |
| 12 | import vllm.envs as envs |
| 13 | from vllm.logger import init_logger |
| 14 | |
| 15 | logger = init_logger(__name__) |
| 16 | |
| 17 | |
| 18 | def is_s3(model_or_path: str) -> bool: |
| 19 | return model_or_path.lower().startswith("s3://") |
| 20 | |
| 21 | |
| 22 | def is_gcs(model_or_path: str) -> bool: |
| 23 | return model_or_path.lower().startswith("gs://") |
| 24 | |
| 25 | |
| 26 | def is_azure(model_or_path: str) -> bool: |
| 27 | return model_or_path.lower().startswith("az://") |
| 28 | |
| 29 | |
| 30 | def is_cloud_storage(model_or_path: str) -> bool: |
| 31 | return is_s3(model_or_path) or is_gcs(model_or_path) or is_azure(model_or_path) |
| 32 | |
| 33 | |
| 34 | def without_trust_remote_code(kwargs: dict[str, Any]) -> dict[str, Any]: |
| 35 | """Return kwargs without trust_remote_code without modifying original dict.""" |
| 36 | if "trust_remote_code" not in kwargs: |
| 37 | return kwargs |
| 38 | return {k: v for k, v in kwargs.items() if k != "trust_remote_code"} |
| 39 | |
| 40 | |
| 41 | def modelscope_list_repo_files( |
| 42 | repo_id: str, |
| 43 | revision: str | None = None, |
| 44 | token: str | bool | None = None, |
| 45 | ) -> list[str]: |
| 46 | """List files in a modelscope repo.""" |
| 47 | from modelscope.hub.api import HubApi |
| 48 | |
| 49 | api = HubApi() |
| 50 | api.login(token) |
| 51 | |
| 52 | # same as huggingface_hub.list_repo_files |
| 53 | return [ |
| 54 | file["Path"] |
| 55 | for file in api.get_model_files( |
| 56 | model_id=repo_id, revision=revision, recursive=True |
| 57 | ) |
| 58 | ] |
| 59 | |
| 60 | |
| 61 | def _maybe_json_dict(path: str | PathLike) -> dict[str, str]: |
| 62 | with open(path) as f: |
| 63 | try: |
| 64 | return json.loads(f.read()) |
| 65 | except Exception: |
| 66 | return dict[str, str]() |
| 67 | |
| 68 | |
| 69 | def _maybe_space_split_dict(path: str | PathLike) -> dict[str, str]: |
| 70 | parsed_dict = dict[str, str]() |
| 71 | with open(path) as f: |
| 72 | for line in f.readlines(): |
| 73 | try: |
| 74 | model_name, redirect_name = line.strip().split() |
| 75 | parsed_dict[model_name] = redirect_name |
| 76 | except Exception: |
| 77 | pass |
| 78 | return parsed_dict |
| 79 | |
| 80 | |
| 81 | @cache |
| 82 | def maybe_model_redirect(model: str) -> str: |
| 83 | """Use model_redirect to redirect the model name to a local folder. |
| 84 | |
| 85 | Args: |
| 86 | model: hf model name |
| 87 | |
| 88 | Returns: |
| 89 | maybe redirect to a local folder |
| 90 | |
| 91 | """ |
| 92 | model_redirect_path = envs.VLLM_MODEL_REDIRECT_PATH |
| 93 | |
| 94 | if not model_redirect_path: |
| 95 | return model |
| 96 | |
| 97 | if not Path(model_redirect_path).exists(): |
| 98 | return model |
| 99 | |
| 100 | redirect_dict = _maybe_json_dict(model_redirect_path) or _maybe_space_split_dict( |
| 101 | model_redirect_path |
| 102 | ) |
| 103 | if redirect_model := redirect_dict.get(model): |
| 104 | logger.info("model redirect: [ %s ] -> [ %s ]", model, redirect_model) |
| 105 | return redirect_model |
| 106 | |
| 107 | return model |
| 108 | |
| 109 | |
| 110 | def parse_safetensors_file_metadata(path: str | PathLike) -> dict[str, Any]: |
| 111 | with open(path, "rb") as f: |
| 112 | length_of_metadata = struct.unpack("<Q", f.read(8))[0] |
| 113 | metadata = json.loads(f.read(length_of_metadata).decode("utf-8")) |
| 114 | return metadata |
| 115 | |
| 116 | |
| 117 | def convert_model_repo_to_path(model_repo: str) -> str: |
| 118 | """When VLLM_USE_MODELSCOPE is True convert a model |
| 119 | repository string to a Path str.""" |
| 120 | if not envs.VLLM_USE_MODELSCOPE or Path(model_repo).exists(): |
| 121 | return model_repo |
| 122 | from modelscope.utils.file_utils import get_model_cache_root |
| 123 | |
| 124 | return os.path.join(get_model_cache_root(), model_repo) |