2024-05-11 15:06:26 +00:00
|
|
|
from __future__ import annotations
|
|
|
|
from abc import ABC, ABCMeta, abstractmethod
|
|
|
|
|
|
|
|
import logging
|
|
|
|
from typing import Any, Callable
|
|
|
|
from collections import deque
|
|
|
|
|
|
|
|
import numpy as np
|
|
|
|
from numpy.typing import DTypeLike
|
|
|
|
|
|
|
|
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
|
|
|
|
|
|
class LazyMeta(ABCMeta):
|
|
|
|
|
|
|
|
def __new__(cls, name: str, bases: tuple[type, ...], namespace: dict[str, Any], **kwargs):
|
2024-07-07 19:04:39 +00:00
|
|
|
def __getattr__(self, name: str) -> Any:
|
|
|
|
meta_attr = getattr(self._meta, name)
|
2024-05-11 15:06:26 +00:00
|
|
|
if callable(meta_attr):
|
|
|
|
return type(self)._wrap_fn(
|
2024-07-07 19:04:39 +00:00
|
|
|
(lambda s, *args, **kwargs: getattr(s, name)(*args, **kwargs)),
|
2024-05-11 15:06:26 +00:00
|
|
|
use_self=self,
|
|
|
|
)
|
|
|
|
elif isinstance(meta_attr, self._tensor_type):
|
|
|
|
# e.g. self.T with torch.Tensor should still be wrapped
|
2024-07-07 19:04:39 +00:00
|
|
|
return type(self)._wrap_fn(lambda s: getattr(s, name))(self)
|
2024-05-11 15:06:26 +00:00
|
|
|
else:
|
|
|
|
# no need to wrap non-tensor properties,
|
|
|
|
# and they likely don't depend on the actual contents of the tensor
|
|
|
|
return meta_attr
|
|
|
|
|
|
|
|
namespace["__getattr__"] = __getattr__
|
|
|
|
|
|
|
|
# need to make a builder for the wrapped wrapper to copy the name,
|
|
|
|
# or else it fails with very cryptic error messages,
|
|
|
|
# because somehow the same string would end up in every closures
|
|
|
|
def mk_wrap(op_name: str, *, meta_noop: bool = False):
|
|
|
|
# need to wrap the wrapper to get self
|
|
|
|
def wrapped_special_op(self, *args, **kwargs):
|
|
|
|
return type(self)._wrap_fn(
|
|
|
|
getattr(type(self)._tensor_type, op_name),
|
|
|
|
meta_noop=meta_noop,
|
|
|
|
)(self, *args, **kwargs)
|
|
|
|
return wrapped_special_op
|
|
|
|
|
|
|
|
# special methods bypass __getattr__, so they need to be added manually
|
|
|
|
# ref: https://docs.python.org/3/reference/datamodel.html#special-lookup
|
|
|
|
# NOTE: doing this from a metaclass is very convenient
|
|
|
|
# TODO: make this even more comprehensive
|
|
|
|
for binary_op in (
|
|
|
|
"lt", "le", "eq", "ne", "ge", "gt", "not"
|
|
|
|
"abs", "add", "and", "floordiv", "invert", "lshift", "mod", "mul", "matmul",
|
|
|
|
"neg", "or", "pos", "pow", "rshift", "sub", "truediv", "xor",
|
|
|
|
"iadd", "iand", "ifloordiv", "ilshift", "imod", "imul", "ior", "irshift", "isub", "ixor",
|
|
|
|
"radd", "rand", "rfloordiv", "rmul", "ror", "rpow", "rsub", "rtruediv", "rxor",
|
|
|
|
):
|
|
|
|
attr_name = f"__{binary_op}__"
|
|
|
|
# the result of these operators usually has the same shape and dtype as the input,
|
|
|
|
# so evaluation on the meta tensor can be skipped.
|
|
|
|
namespace[attr_name] = mk_wrap(attr_name, meta_noop=True)
|
|
|
|
|
|
|
|
for special_op in (
|
|
|
|
"getitem", "setitem", "len",
|
|
|
|
):
|
|
|
|
attr_name = f"__{special_op}__"
|
|
|
|
namespace[attr_name] = mk_wrap(attr_name, meta_noop=False)
|
|
|
|
|
|
|
|
return super().__new__(cls, name, bases, namespace, **kwargs)
|
|
|
|
|
|
|
|
|
|
|
|
# Tree of lazy tensors
|
|
|
|
class LazyBase(ABC, metaclass=LazyMeta):
|
|
|
|
_tensor_type: type
|
|
|
|
_meta: Any
|
|
|
|
_data: Any | None
|
|
|
|
_lazy: deque[LazyBase] # shared within a graph, to avoid deep recursion when making eager
|
|
|
|
_args: tuple
|
|
|
|
_func: Callable[[tuple], Any] | None
|
|
|
|
|
|
|
|
def __init__(self, *, meta: Any, data: Any | None = None, lazy: deque[LazyBase] | None = None, args: tuple = (), func: Callable[[tuple], Any] | None = None):
|
|
|
|
super().__init__()
|
|
|
|
self._meta = meta
|
|
|
|
self._data = data
|
|
|
|
self._lazy = lazy if lazy is not None else deque()
|
|
|
|
self._args = args
|
|
|
|
self._func = func
|
|
|
|
assert self._func is not None or self._data is not None
|
|
|
|
if self._data is None:
|
|
|
|
self._lazy.append(self)
|
|
|
|
|
|
|
|
def __init_subclass__(cls) -> None:
|
|
|
|
if "_tensor_type" not in cls.__dict__:
|
|
|
|
raise TypeError(f"property '_tensor_type' must be defined for {cls!r}")
|
|
|
|
return super().__init_subclass__()
|
|
|
|
|
|
|
|
@staticmethod
|
|
|
|
def _recurse_apply(o: Any, fn: Callable[[Any], Any]) -> Any:
|
|
|
|
# TODO: dict and set
|
|
|
|
if isinstance(o, (list, tuple)):
|
|
|
|
L = []
|
|
|
|
for item in o:
|
|
|
|
L.append(LazyBase._recurse_apply(item, fn))
|
|
|
|
if isinstance(o, tuple):
|
|
|
|
L = tuple(L)
|
|
|
|
return L
|
|
|
|
elif isinstance(o, LazyBase):
|
|
|
|
return fn(o)
|
|
|
|
else:
|
|
|
|
return o
|
|
|
|
|
|
|
|
@classmethod
|
2024-05-13 18:10:51 +00:00
|
|
|
def _wrap_fn(cls, fn: Callable, *, use_self: LazyBase | None = None, meta_noop: bool | DTypeLike | tuple[DTypeLike, Callable[[tuple[int, ...]], tuple[int, ...]]] = False) -> Callable[[Any], Any]:
|
2024-05-11 15:06:26 +00:00
|
|
|
def wrapped_fn(*args, **kwargs):
|
|
|
|
if kwargs is None:
|
|
|
|
kwargs = {}
|
|
|
|
args = ((use_self,) if use_self is not None else ()) + args
|
|
|
|
|
|
|
|
meta_args = LazyBase._recurse_apply(args, lambda t: t._meta)
|
|
|
|
|
|
|
|
if isinstance(meta_noop, bool) and not meta_noop:
|
|
|
|
try:
|
|
|
|
res = fn(*meta_args, **kwargs)
|
|
|
|
except NotImplementedError:
|
|
|
|
# running some operations on PyTorch's Meta tensors can cause this exception
|
|
|
|
res = None
|
|
|
|
else:
|
|
|
|
# some operators don't need to actually run on the meta tensors
|
|
|
|
assert len(args) > 0
|
|
|
|
res = args[0]
|
|
|
|
assert isinstance(res, cls)
|
|
|
|
res = res._meta
|
2024-05-13 18:10:51 +00:00
|
|
|
# allow operations to override the dtype and shape
|
2024-05-11 15:06:26 +00:00
|
|
|
if meta_noop is not True:
|
2024-05-13 18:10:51 +00:00
|
|
|
if isinstance(meta_noop, tuple):
|
|
|
|
dtype, shape = meta_noop
|
|
|
|
assert callable(shape)
|
|
|
|
res = cls.meta_with_dtype_and_shape(dtype, shape(res.shape))
|
|
|
|
else:
|
|
|
|
res = cls.meta_with_dtype_and_shape(meta_noop, res.shape)
|
2024-05-11 15:06:26 +00:00
|
|
|
|
|
|
|
if isinstance(res, cls._tensor_type):
|
2024-07-07 19:04:39 +00:00
|
|
|
class CollectSharedLazy:
|
|
|
|
# emulating a static variable
|
|
|
|
shared_lazy: None | deque[LazyBase] = None
|
2024-05-11 15:06:26 +00:00
|
|
|
|
2024-07-07 19:04:39 +00:00
|
|
|
@staticmethod
|
|
|
|
def collect_replace(t: LazyBase):
|
|
|
|
if CollectSharedLazy.shared_lazy is None:
|
|
|
|
CollectSharedLazy.shared_lazy = t._lazy
|
|
|
|
else:
|
|
|
|
CollectSharedLazy.shared_lazy.extend(t._lazy)
|
|
|
|
t._lazy = CollectSharedLazy.shared_lazy
|
2024-05-11 15:06:26 +00:00
|
|
|
|
2024-07-07 19:04:39 +00:00
|
|
|
LazyBase._recurse_apply(args, CollectSharedLazy.collect_replace)
|
2024-05-11 15:06:26 +00:00
|
|
|
|
2024-07-07 19:04:39 +00:00
|
|
|
shared_lazy = CollectSharedLazy.shared_lazy
|
2024-05-11 15:06:26 +00:00
|
|
|
|
|
|
|
return cls(meta=cls.eager_to_meta(res), lazy=shared_lazy, args=args, func=lambda a: fn(*a, **kwargs))
|
|
|
|
else:
|
|
|
|
del res # not needed
|
|
|
|
# non-tensor return likely relies on the contents of the args
|
|
|
|
# (e.g. the result of torch.equal)
|
|
|
|
eager_args = cls.to_eager(args)
|
|
|
|
return fn(*eager_args, **kwargs)
|
|
|
|
return wrapped_fn
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
def to_eager(cls, t: Any) -> Any:
|
|
|
|
def simple_to_eager(_t: LazyBase) -> Any:
|
|
|
|
def already_eager_to_eager(_t: LazyBase) -> Any:
|
|
|
|
assert _t._data is not None
|
|
|
|
return _t._data
|
|
|
|
|
|
|
|
while _t._data is None:
|
|
|
|
lt = _t._lazy.popleft()
|
|
|
|
if lt._data is not None:
|
2024-05-13 18:10:51 +00:00
|
|
|
# Lazy tensor did not belong in the lazy queue.
|
|
|
|
# Weirdly only happens with Bloom models...
|
|
|
|
# likely because tensors aren't unique in the queue.
|
|
|
|
# The final output is still the same as in eager mode,
|
|
|
|
# so it's safe to ignore this.
|
|
|
|
continue
|
2024-05-11 15:06:26 +00:00
|
|
|
assert lt._func is not None
|
|
|
|
lt._args = cls._recurse_apply(lt._args, already_eager_to_eager)
|
|
|
|
lt._data = lt._func(lt._args)
|
|
|
|
# sanity check
|
2024-07-07 19:04:39 +00:00
|
|
|
assert lt._data is not None
|
2024-05-11 15:06:26 +00:00
|
|
|
assert lt._data.dtype == lt._meta.dtype
|
|
|
|
assert lt._data.shape == lt._meta.shape
|
|
|
|
|
|
|
|
return _t._data
|
|
|
|
|
|
|
|
# recurse into lists and/or tuples, keeping their structure
|
|
|
|
return cls._recurse_apply(t, simple_to_eager)
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
def eager_to_meta(cls, t: Any) -> Any:
|
2024-05-13 18:10:51 +00:00
|
|
|
return cls.meta_with_dtype_and_shape(t.dtype, t.shape)
|
2024-05-11 15:06:26 +00:00
|
|
|
|
|
|
|
# must be overridden, meta tensor init is backend-specific
|
|
|
|
@classmethod
|
|
|
|
@abstractmethod
|
2024-05-13 18:10:51 +00:00
|
|
|
def meta_with_dtype_and_shape(cls, dtype: Any, shape: Any) -> Any: pass
|
2024-05-11 15:06:26 +00:00
|
|
|
|
|
|
|
@classmethod
|
|
|
|
def from_eager(cls, t: Any) -> Any:
|
|
|
|
if type(t) is cls:
|
|
|
|
# already eager
|
|
|
|
return t
|
|
|
|
elif isinstance(t, cls._tensor_type):
|
|
|
|
return cls(meta=cls.eager_to_meta(t), data=t)
|
|
|
|
else:
|
|
|
|
return TypeError(f"{type(t)!r} is not compatible with {cls._tensor_type!r}")
|
|
|
|
|
|
|
|
|
|
|
|
class LazyNumpyTensor(LazyBase):
|
|
|
|
_tensor_type = np.ndarray
|
|
|
|
|
|
|
|
@classmethod
|
2024-05-22 18:29:50 +00:00
|
|
|
def meta_with_dtype_and_shape(cls, dtype: DTypeLike, shape: tuple[int, ...]) -> np.ndarray[Any, Any]:
|
2024-05-11 15:06:26 +00:00
|
|
|
# The initial idea was to use np.nan as the fill value,
|
|
|
|
# but non-float types like np.int16 can't use that.
|
|
|
|
# So zero it is.
|
|
|
|
cheat = np.zeros(1, dtype)
|
2024-05-13 18:10:51 +00:00
|
|
|
return np.lib.stride_tricks.as_strided(cheat, shape, (0 for _ in shape))
|
2024-05-11 15:06:26 +00:00
|
|
|
|
|
|
|
def astype(self, dtype, *args, **kwargs):
|
2024-05-13 18:10:51 +00:00
|
|
|
meta = type(self).meta_with_dtype_and_shape(dtype, self._meta.shape)
|
2024-05-11 15:06:26 +00:00
|
|
|
full_args = (self, dtype,) + args
|
|
|
|
# very important to pass the shared _lazy deque, or else there's an infinite loop somewhere.
|
|
|
|
return type(self)(meta=meta, args=full_args, lazy=self._lazy, func=(lambda a: a[0].astype(*a[1:], **kwargs)))
|
|
|
|
|
|
|
|
def tofile(self, *args, **kwargs):
|
|
|
|
eager = LazyNumpyTensor.to_eager(self)
|
|
|
|
return eager.tofile(*args, **kwargs)
|
|
|
|
|
|
|
|
# TODO: __array_function__
|