75 lines
2.4 KiB
Python
75 lines
2.4 KiB
Python
|
import contextlib
|
||
|
|
import copy
|
||
|
|
import threading
|
||
|
|
from pathlib import Path
|
||
|
|
from typing import Any, Callable, Dict, Optional, Tuple, Union
|
||
|
|
|
||
|
|
|
||
|
|
class Shim: # pragma: no cover
|
||
|
|
"""Define a basic interface for external models. Users can create subclasses
|
||
|
|
of 'shim' to wrap external libraries. We provide shims for PyTorch.
|
||
|
|
|
||
|
|
The Thinc Model class treats Shim objects as a sort of special type of
|
||
|
|
sublayer: it knows they're not actual Thinc Model instances, but it also
|
||
|
|
knows to talk to the shim instances when doing things like using transferring
|
||
|
|
between devices, loading in parameters, optimization. It also knows Shim
|
||
|
|
objects need to be serialized and deserialized with to/from bytes/disk,
|
||
|
|
rather than expecting that they'll be msgpack-serializable.
|
||
|
|
"""
|
||
|
|
|
||
|
|
global_id: int = 0
|
||
|
|
global_id_lock: threading.Lock = threading.Lock()
|
||
|
|
cfg: Dict
|
||
|
|
_model: Any
|
||
|
|
_optimizer: Optional[Any]
|
||
|
|
|
||
|
|
def __init__(self, model: Any, config=None, optimizer: Any = None):
|
||
|
|
with Shim.global_id_lock:
|
||
|
|
Shim.global_id += 1
|
||
|
|
self.id = Shim.global_id
|
||
|
|
|
||
|
|
self.cfg = dict(config) if config is not None else {}
|
||
|
|
self._model = model
|
||
|
|
self._optimizer = optimizer
|
||
|
|
|
||
|
|
def __call__(self, inputs, is_train: bool) -> Tuple[Any, Callable[..., Any]]:
|
||
|
|
raise NotImplementedError
|
||
|
|
|
||
|
|
def predict(self, fwd_args: Any) -> Any:
|
||
|
|
Y, backprop = self(fwd_args, is_train=False)
|
||
|
|
return Y
|
||
|
|
|
||
|
|
def begin_update(self, fwd_args: Any) -> Tuple[Any, Callable[..., Any]]:
|
||
|
|
return self(fwd_args, is_train=True)
|
||
|
|
|
||
|
|
def finish_update(self, optimizer):
|
||
|
|
raise NotImplementedError
|
||
|
|
|
||
|
|
@contextlib.contextmanager
|
||
|
|
def use_params(self, params):
|
||
|
|
yield
|
||
|
|
|
||
|
|
def copy(self):
|
||
|
|
return copy.deepcopy(self)
|
||
|
|
|
||
|
|
def to_device(self, device_type: str, device_id: int):
|
||
|
|
raise NotImplementedError
|
||
|
|
|
||
|
|
def to_disk(self, path: Union[str, Path]):
|
||
|
|
bytes_data = self.to_bytes()
|
||
|
|
path = Path(path) if isinstance(path, str) else path
|
||
|
|
with path.open("wb") as file_:
|
||
|
|
file_.write(bytes_data)
|
||
|
|
|
||
|
|
def from_disk(self, path: Union[str, Path]) -> "Shim":
|
||
|
|
path = Path(path) if isinstance(path, str) else path
|
||
|
|
with path.open("rb") as file_:
|
||
|
|
bytes_data = file_.read()
|
||
|
|
return self.from_bytes(bytes_data)
|
||
|
|
|
||
|
|
def to_bytes(self):
|
||
|
|
raise NotImplementedError
|
||
|
|
|
||
|
|
def from_bytes(self, data) -> "Shim":
|
||
|
|
raise NotImplementedError
|