[lazyinit] add correctness verification (#3147)

* [lazyinit] fix shared module

* [tests] add lazy init test utils

* [tests] add torchvision for lazy init

* [lazyinit] fix pre op fn

* [lazyinit] handle legacy constructor

* [tests] refactor lazy init test models

* [tests] refactor lazy init test utils

* [lazyinit] fix ops don't support meta

* [tests] lazy init test timm models

* [lazyinit] fix set data

* [lazyinit] handle apex layers

* [tests] lazy init test transformers models

* [tests] lazy init test torchaudio models

* [lazyinit] fix import path

* [tests] lazy init test torchrec models

* [tests] update torch version in CI

* [tests] revert torch version in CI

* [tests] skip lazy init test

colossalai/utils/model/experimental.py

@@ -1,17 +1,16 @@
-from typing import Callable, Optional, Union
+from typing import Callable, List, Optional, Union
 
 import torch
 import torch.nn as nn
 from torch import Tensor
 from torch.utils._pytree import tree_map
 
-from colossalai.fx.profiler import MetaTensor
+from colossalai.fx.profiler.tensor import MetaTensor
 
 # reference: https://pytorch.org/cppdocs/notes/tensor_creation.html
-_TorchFactoryMethod = [
+_NORMAL_FACTORY = [
     "arange",
     "empty",
-    "eye",
     "full",
     "linspace",
     "logspace",
@@ -24,17 +23,39 @@ _TorchFactoryMethod = [
     "tensor",
 ]
 
+# factory function that does not support meta tensor backend
+_NO_META_FACTORY = [
+    "eye",
+]
+
 _EARLY_MATERIALIZED_OPS = ['__getitem__', 'split']
 
+_LEGACY_TENSOR_CONSTRUCTOR = {
+    'FloatTensor': torch.float,
+    'DoubleTensor': torch.double,
+    'HalfTensor': torch.half,
+    'BFloat16Tensor': torch.bfloat16,
+    'ByteTensor': torch.uint8,
+    'CharTensor': torch.int8,
+    'ShortTensor': torch.short,
+    'IntTensor': torch.int,
+    'LongTensor': torch.long,
+    'BoolTensor': torch.bool,
+}
+
 
 class _MyTensor(Tensor):
     """This class is only for correctness verification.
     """
     _pre_op_fn: Callable[['LazyTensor'], None] = lambda *args: None
 
-    def __new__(cls, func, *args, dtype=None, device=None, **kwargs) -> '_MyTensor':
+    def __new__(cls, func, *args, concrete_data=None, **kwargs) -> '_MyTensor':
         cls._pre_op_fn()
-        data = func(*args, dtype=dtype, device=device, **kwargs)
+        if concrete_data is not None:
+            # uniform api as LazyTensor
+            data = concrete_data
+        else:
+            data = func(*args, **kwargs)
         return Tensor._make_subclass(cls, data, require_grad=data.requires_grad)
 
     @classmethod
@@ -66,11 +87,13 @@ class LazyTensor(torch.Tensor):
         >>> x.add_(1) # modifying origin tensor after cloning leads to wrong materialization
         >>> z = x.tolist()
         >>> x.zeros_() # modifying origin tensor after cloning tolist is not allowed
-        >>> x.data = torch.rand(2, 3) # directly set data of a lazy tensor is not allowed
+        >>> nn.utils.weight_norm(self.conv, name="weight", dim=2) # applying weight norm on a lazy tensor is not allowed
+
 
         2. Cases that ``LazyTensor`` becomes eager (early materialization).
         >>> b = a[:, 2:]  # get a slice of a lazy tensor triggers early materialization
         >>> chunks = a.split(3)  # this also triggers early materialization
+        >>> x.data = torch.rand(2, 3) # directly setting data of a lazy tensor triggers early materialization
 
     """
 
@@ -79,12 +102,16 @@ class LazyTensor(torch.Tensor):
     _pre_op_fn: Callable[['LazyTensor'], None] = lambda *args: None
 
     @staticmethod
-    def __new__(cls, func, *args, meta_data=None, **kwargs):
+    def __new__(cls, func, *args, meta_data=None, concrete_data=None, **kwargs):
-        if meta_data is None:
+        if concrete_data is not None:
-            device = kwargs.get('device', 'cpu')
+            # some ops don't support meta backend and should have concrete data
-            elem = func(*args, **{**kwargs, 'device': 'meta'})
+            elem = concrete_data
-            meta_data = MetaTensor(elem, fake_device=device)
+        else:
-        elem = meta_data._tensor
+            if meta_data is None:
+                device = kwargs.get('device', 'cpu')
+                elem = func(*args, **{**kwargs, 'device': 'meta'})
+                meta_data = MetaTensor(elem, fake_device=device)
+            elem = meta_data._tensor
         r = torch.Tensor._make_wrapper_subclass(cls,
                                                 elem.size(),
                                                 strides=elem.stride(),
@@ -96,10 +123,10 @@ class LazyTensor(torch.Tensor):
         r._meta_data = meta_data
         return r
 
-    def __init__(self, func, *args, meta_data=None, **kwargs):
+    def __init__(self, func, *args, meta_data=None, concrete_data=None, **kwargs):
         self._factory_method = (func, args, kwargs)    # (func, args, kwargs)
         self._op_buffer = []    # (func, args, kwargs, replace)
-        self._materialized_data: Optional[torch.Tensor] = None    # materialized data
+        self._materialized_data: Optional[torch.Tensor] = concrete_data    # materialized data
 
     def materialize(self) -> torch.Tensor:
         """Materialize the ``LazyTensor`` to ``torch.Tensor``.
@@ -212,7 +239,7 @@ class LazyTensor(torch.Tensor):
                 if isinstance(x, LazyTensor):
                     if x._materialized_data is not None:
                         # for early materialized tensor, use its materialized data directly
-                        return x._materialized_data
+                        return x._materialized_data.data
                     t = x if is_inplace else x.clone()
                     t._op_buffer.append((func, args, kwargs))
                     meta = x._meta_data.data
@@ -232,13 +259,10 @@ class LazyTensor(torch.Tensor):
                         return lazy_y
                 elif type(y) is Tensor:
                     # for early materialized tensor
-                    with torch._C.DisableTorchFunction():
+                    return LazyTensor(lambda: None, concrete_data=y)
-                        meta = MetaTensor(y.new_empty(y.shape, dtype=y.dtype, device='meta'), fake_device=y.device)
-                    lazy_y = LazyTensor(lambda: None, meta_data=meta)
-                    lazy_y._materialized_data = y
-                    return lazy_y
                 return y
 
+            cls._pre_op_fn()
             o = func(*tree_map(unwrap, args), **tree_map(unwrap, kwargs))
             if isinstance(o, (tuple, list)):
                 return type(o)(wrap(y, i=i) for i, y in enumerate(o))
@@ -266,7 +290,10 @@ class LazyTensor(torch.Tensor):
 
     @data.setter
     def data(self, other: 'LazyTensor'):
-        raise NotImplementedError
+        if other is self:
+            return
+        # TODO(ver217): to avoid infinity recursion, do early materialization
+        self._materialized_data = other._materialize_data()
 
     def tolist(self) -> list:
         t = self.materialize()
@@ -330,18 +357,61 @@ class LazyInitContext:
 
             return wrapper, target
 
+        def wrap_legacy_constructor(target, dtype):
+            # legacy constructor (e.g. torch.LongTensor())
+            def wrapper(*args, **kwargs):
+                if len(args) == 1 and isinstance(args[0], torch.Tensor):
+                    # (Tensor other)
+                    return args[0]
+                elif len(args) == 1:
+                    # (object data, *, torch.device device)
+                    kwargs = {**kwargs, 'dtype': dtype}
+                    replaced, orig = self.overrides['tensor']
+                    return replaced(*args, **kwargs)
+                elif _is_int_tuple(args):
+                    # (tuple of ints size, *, torch.device device)
+                    kwargs = {**kwargs, 'dtype': dtype}
+                    replaced, orig = self.overrides['empty']
+                    return replaced(*args, **kwargs)
+                else:
+                    raise TypeError(
+                        f'new() received an invalid combination of arguments - got {tuple(type(x) for x in args)}, but expected one of:\n * (Tensor other)\n * (tuple of ints size, *, torch.device device)\n * (object data, *, torch.device device)'
+                    )
+
+            return wrapper, target
+
+        def wrap_no_meta_factory(target):
+            # factory functions which don't support meta tensor backend
+            def wrapper(*args, **kwargs):
+                tensor = target(*args, **kwargs)
+                return self.tensor_cls(lambda: None, concrete_data=tensor)
+
+            return wrapper, target
+
         self.overrides = {
             target: wrap_factory_method(getattr(torch, target))
-            for target in _TorchFactoryMethod
+            for target in _NORMAL_FACTORY
             if callable(getattr(torch, target, None))
         }
 
         self.overrides.update({
             target + '_like': wrap_factory_like_method(getattr(torch, target), getattr(torch, target + '_like'))
-            for target in _TorchFactoryMethod
+            for target in _NORMAL_FACTORY
             if callable(getattr(torch, target + '_like', None))
         })
 
+        self.overrides.update({
+            target: wrap_legacy_constructor(getattr(torch, target), dtype)
+            for target, dtype in _LEGACY_TENSOR_CONSTRUCTOR.items()
+            if callable(getattr(torch, target, None))
+        })
+
+        self.overrides.update({
+            target: wrap_no_meta_factory(getattr(torch, target))
+            for target in _NO_META_FACTORY
+            if callable(getattr(torch, target, None))
+        })
+
         for name, (wrapper, orig) in self.overrides.items():
             setattr(torch, name, wrapper)
 
@@ -363,34 +433,65 @@ class LazyInitContext:
             param_lazy_cnt = 0
             buf_cnt = 0
             buf_lazy_cnt = 0
+            non_lazy_numel = 0
+
+        # do post cleaning to handle shared parameter
+        visited_lazy_tensors: List[LazyTensor] = []
+        # handle shared module
+        visited_modules = set()
 
         @torch.no_grad()
         def init_recursively(module: nn.Module):
-            nonlocal param_cnt, param_lazy_cnt, buf_cnt, buf_lazy_cnt
+            nonlocal param_cnt, param_lazy_cnt, buf_cnt, buf_lazy_cnt, non_lazy_numel
             # recursively initialize the module
             for mod in module.children():
-                init_recursively(mod)
+                if id(mod) not in visited_modules:
+                    visited_modules.add(id(mod))
+                    init_recursively(mod)
 
             # initialize tensors directly attached to the current module
             for name, param in module.named_parameters(recurse=False):
                 if verbose:
                     param_cnt += 1
-                    if param._materialized_data is None:
+                    if getattr(param, '_materialized_data', False) is None:
+                        # if no _materialized_data attr, the tensor is not lazy
                         param_lazy_cnt += 1
-                setattr(module, name, param.materialize())
+                    else:
-                param.clean()
+                        non_lazy_numel += param.numel()
+                if hasattr(param, 'materialize'):
+                    # TODO(ver217): apex layers cannot be captured
+                    visited_lazy_tensors.append(param)
+                    setattr(module, name, param.materialize())
 
             for name, buf in module.named_buffers(recurse=False):
                 if verbose:
                     buf_cnt += 1
-                    if buf._materialized_data is None:
+                    if getattr(buf, "_materialized_data", False) is None:
+                        # if no _materialized_data attr, the tensor is not lazy
                         buf_lazy_cnt += 1
-                setattr(module, name, buf.materialize())
+                    else:
-                buf.clean()
+                        non_lazy_numel += buf.numel()
+                if hasattr(buf, 'materialize'):
+                    # TODO(ver217): apex layers cannot be captured
+                    visited_lazy_tensors.append(buf)
+                    setattr(module, name, buf.materialize())
 
         init_recursively(module)
 
+        for t in visited_lazy_tensors:
+            t.clean()
+
         if verbose:
             print(f'Param lazy rate: {param_lazy_cnt}/{param_cnt}')
             print(f'Buffer lazy rate: {buf_lazy_cnt}/{buf_cnt}')
+            print(f'Non-lazy numel: {non_lazy_numel} ({non_lazy_numel/1024**2:.3f} M)')
         return module
+
+
+def _is_int_tuple(args) -> bool:
+    if not isinstance(args, tuple):
+        return False
+    for x in args:
+        if not isinstance(x, int):
+            return False
+    return True

tests/kit/model_zoo/torchrec/__init__.py

@@ -0,0 +1 @@
+from .torchrec import *

tests/test_utils/test_lazy_init/test_models.py

@@ -0,0 +1,23 @@
+import pytest
+
+from tests.kit.model_zoo import model_zoo
+
+# FIXME(ver217): uncomment this line
+# from utils import check_lazy_init
+
+
+# FIXME(ver217): temporarily skip this test since torch 1.11 does not fully support meta tensor
+@pytest.mark.skip
+@pytest.mark.parametrize('subset', ['torchvision', 'diffusers', 'timm', 'transformers', 'torchaudio', 'deepfm', 'dlrm'])
+def test_torchvision_models_lazy_init(subset):
+    sub_model_zoo = model_zoo.get_sub_registry(subset)
+    for name, entry in sub_model_zoo.items():
+        # TODO(ver217): lazy init does not support weight norm, skip these models
+        if name in ('torchaudio_wav2vec2_base', 'torchaudio_hubert_base'):
+            continue
+        # FIXME(ver217): uncomment this line
+        # check_lazy_init(entry, verbose=True)
+
+
+if __name__ == '__main__':
+    test_torchvision_models_lazy_init('torchvision')

tests/test_utils/test_lazy_init/utils.py

@@ -0,0 +1,69 @@
+import random
+from typing import Any, Callable, Optional, Tuple
+
+import numpy as np
+import torch
+
+from colossalai.utils.model.experimental import LazyInitContext, LazyTensor, _MyTensor
+from tests.kit.model_zoo.registry import ModelAttribute
+
+# model_fn, data_gen_fn, output_transform_fn, model_attr
+TestingEntry = Tuple[Callable[[], torch.nn.Module], Callable[[], dict], Callable[[], dict], Optional[ModelAttribute]]
+
+
+def set_seed(seed: int) -> None:
+    random.seed(seed)
+    np.random.seed(seed)
+    torch.manual_seed(seed)
+
+
+def assert_model_eqaual(m1: torch.nn.Module, m2: torch.nn.Module) -> None:
+    s1 = m1.state_dict()
+    s2 = m2.state_dict()
+
+    assert len(s1) == len(s2), f'len {len(s1)} vs {len(s2)}'
+
+    for (n1, t1), (n2, t2) in zip(s1.items(), s2.items()):
+        assert n1 == n2
+        assert torch.equal(t1, t2), f'{n1} {t1} vs {t2}'
+
+
+def assert_forward_equal(m1: torch.nn.Module, m2: torch.nn.Module, data_gen_fn: Callable[[], dict],
+                         output_transform_fn: Callable[[Any], dict]) -> None:
+    data = data_gen_fn()
+
+    m1.eval()
+    m2.eval()
+    # run forward
+    with torch.no_grad():
+        outputs1 = m1(**data)
+        outputs2 = m2(**data)
+
+    # compare output
+    transformed_out1 = output_transform_fn(outputs1)
+    transformed_out2 = output_transform_fn(outputs2)
+
+    assert len(transformed_out1) == len(transformed_out2)
+
+    for key, out1 in transformed_out1.items():
+        out2 = transformed_out2[key]
+        assert torch.allclose(out1, out2, atol=1e-5), \
+            f'{m1.__class__.__name__} has inconsistent outputs, {out1} vs {out2}'
+
+
+def check_lazy_init(entry: TestingEntry, seed: int = 42, verbose: bool = False, check_forward: bool = False) -> None:
+    model_fn, data_gen_fn, output_transform_fn, model_attr = entry
+    _MyTensor._pre_op_fn = lambda *args: set_seed(seed)
+    LazyTensor._pre_op_fn = lambda *args: set_seed(seed)
+    ctx = LazyInitContext(tensor_cls=_MyTensor)
+    with ctx:
+        model = model_fn()
+    ctx = LazyInitContext()
+    with ctx:
+        deferred_model = model_fn()
+    deferred_model = ctx.materialize(deferred_model, verbose=verbose)
+    assert_model_eqaual(model, deferred_model)
+    if check_forward:
+        assert_forward_equal(model, deferred_model, data_gen_fn, output_transform_fn)
+    if verbose:
+        print(f'{model.__class__.__name__} pass')