Utilities for working with the local dataset cache.
CACHE_ROOT = Path(os.getenv("ALLENNLP_CACHE_ROOT", Path.home() / ".allennlp"))
CACHE_DIRECTORY = str(CACHE_ROOT / "cache")
DEPRECATED_CACHE_DIRECTORY = str(CACHE_ROOT / "datasets")
DATASET_CACHE = CACHE_DIRECTORY
class FileLock(_FileLock): | def __init__( | self, | lock_file: Union[str, PathLike], | timeout=-1, | read_only_ok: bool = False | ) -> None
This is just a subclass of the
FileLock class from the
filelock library, except that
it adds an additional argument to the
By default this flag is
False, which an exception will be thrown when a lock
can't be acquired due to lack of write permissions.
But if this flag is set to
True, a warning will be emitted instead of an error when
the lock already exists but the lock can't be acquired because write access is blocked.
class FileLock(_FileLock): | ... | @overrides | def acquire(self, timeout=None, poll_interval=0.05)
def filename_to_url( filename: str, cache_dir: Union[str, Path] = None ) -> Tuple[str, str]
Return the url and etag (which may be
None) stored for
filename or its stored metadata do not exist.
def check_tarfile(tar_file: tarfile.TarFile)
Tar files can contain files outside of the extraction directory, or symlinks that point outside the extraction directory. We also don't want any block devices fifos, or other weird file types extracted. This checks for those issues and throws an exception if there is a problem.
def cached_path( url_or_filename: Union[str, PathLike], cache_dir: Union[str, Path] = None, extract_archive: bool = False, force_extract: bool = False ) -> str
Given something that might be a URL or local path, determine which. If it's a remote resource, download the file and cache it, and then return the path to the cached file. If it's already a local path, make sure the file exists and return the path.
For URLs, "http://", "https://", "s3://", "gs://", and "hf://" are all supported. The latter corresponds to the HuggingFace Hub.
For example, to download the PyTorch weights for the model
on HuggingFace, you could do:
For paths or URLs that point to a tarfile or zipfile, you can also add a path
to a specific file to the
url_or_filename preceeded by a "!", and the archive will
be automatically extracted (provided you set
returning the local path to the specific file. For example:
A URL or path to parse and possibly download.
Union[str, Path], optional (default =
The directory to cache downloads.
bool, optional (default =
True, then zip or tar.gz archives will be automatically extracted. In which case the directory is returned.
bool, optional (default =
Trueand the file is an archive file, it will be extracted regardless of whether or not the extracted directory already exists.
Use this flag with caution! This can lead to race conditions if used from multiple processes on the same file.
def is_url_or_existing_file( url_or_filename: Union[str, Path, None] ) -> bool
Given something that might be a URL (or might be a local path), determine check if it's url or an existing file path.
class TensorCache(MutableMapping[str, Tensor], ABC): | def __init__( | self, | filename: Union[str, PathLike], | *, | map_size: int = 1024 * 1024 * 1024 * 1024, | read_only: bool = False | ) -> None
This is a key-value store, mapping strings to tensors. The data is kept on disk, making this class useful as a cache for storing tensors.
TensorCache is also safe to access from multiple processes at the same time, so
you can use it in distributed training situations, or from multiple training
runs at the same time.
class TensorCache(MutableMapping[str, Tensor], ABC): | ... | @property | def read_only(self) -> bool
class TensorCache(MutableMapping[str, Tensor], ABC): | ... | def __iter__(self)
It is not hard to implement this, but we have not needed it so far.
class CacheFile: | def __init__( | self, | cache_filename: Union[PathLike, str], | mode: str = "w+b", | suffix: str = ".tmp" | ) -> None
This is a context manager that makes robust caching easier.
__enter__, an IO handle to a temporarily file is returned, which can
be treated as if it's the actual cache file.
__exit__, the temporarily file is renamed to the cache file. If anything
goes wrong while writing to the temporary file, it will be removed.
class LocalCacheResource: | def __init__( | self, | resource_name: str, | version: str, | cache_dir: str = CACHE_DIRECTORY | ) -> None
This is a context manager that can be used to fetch and cache arbitrary resources locally
using the same mechanisms that
cached_path uses for remote resources.
It can be used, for example, when you want to cache the result of an expensive computation.
with LocalCacheResource("long-computation", "v1") as cache: if cache.cached(): with cache.reader() as f: # read from cache else: with cache.writer() as f: # do the computation # ... # write to cache
class LocalCacheResource: | ... | def cached(self) -> bool
class LocalCacheResource: | ... | @contextmanager | def writer(self, mode="w")
class LocalCacheResource: | ... | @contextmanager | def reader(self, mode="r")
def get_from_cache( url: str, cache_dir: Union[str, Path] = None ) -> str
Given a URL, look for the corresponding dataset in the local cache. If it's not there, download it. Then return the path to the cached file.
def read_set_from_file(filename: str) -> Set[str]
Extract a de-duped collection (set) of text from a file. Expected file format is one item per line.
def get_file_extension(path: str, dot=True, lower: bool = True)
def open_compressed( filename: Union[str, PathLike], mode: str = "rt", encoding: Optional[str] = "UTF-8", **kwargs )
def text_lines_from_file( filename: Union[str, PathLike], strip_lines: bool = True ) -> Iterator[str]
def json_lines_from_file( filename: Union[str, PathLike] ) -> Iterable[Union[list, dict]]
def remove_cache_entries( patterns: List[str], cache_dir: Union[str, Path] = None ) -> int
Remove cache entries matching the given patterns.
Returns the total reclaimed space in bytes.
def inspect_cache( patterns: List[str] = None, cache_dir: Union[str, Path] = None )
Print out useful information about the cache directory.