File size: 12,277 Bytes
bb2b5cc | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 | # Copyright 2025 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import math
import sys
from collections import defaultdict
from contextlib import contextmanager
import torch
# Record all the torch primitives in advance, so that we can use them without them being modified when we patch torch
# in context managers
TORCH_INIT_FUNCTIONS = {
"uniform_": torch.nn.init.uniform_,
"normal_": torch.nn.init.normal_,
"constant_": torch.nn.init.constant_,
"ones_": torch.nn.init.ones_,
"zeros_": torch.nn.init.zeros_,
"eye_": torch.nn.init.eye_,
"dirac_": torch.nn.init.dirac_,
"xavier_uniform_": torch.nn.init.xavier_uniform_,
"xavier_normal_": torch.nn.init.xavier_normal_,
"kaiming_uniform_": torch.nn.init.kaiming_uniform_,
"kaiming_normal_": torch.nn.init.kaiming_normal_,
"trunc_normal_": torch.nn.init.trunc_normal_,
"orthogonal_": torch.nn.init.orthogonal_,
"sparse_": torch.nn.init.sparse_,
}
def uniform_(
tensor: torch.Tensor, a: float = 0.0, b: float = 1.0, generator: torch.Generator | None = None
) -> torch.Tensor:
if not getattr(tensor, "_is_hf_initialized", False):
return TORCH_INIT_FUNCTIONS["uniform_"](tensor, a=a, b=b, generator=generator)
return tensor
def normal_(
tensor: torch.Tensor, mean: float = 0.0, std: float = 1.0, generator: torch.Generator | None = None
) -> torch.Tensor:
if not getattr(tensor, "_is_hf_initialized", False):
return TORCH_INIT_FUNCTIONS["normal_"](tensor, mean=mean, std=std, generator=generator)
return tensor
def constant_(tensor: torch.Tensor, val: float) -> torch.Tensor:
if not getattr(tensor, "_is_hf_initialized", False):
return TORCH_INIT_FUNCTIONS["constant_"](tensor, val=val)
return tensor
def ones_(tensor: torch.Tensor) -> torch.Tensor:
if not getattr(tensor, "_is_hf_initialized", False):
return TORCH_INIT_FUNCTIONS["ones_"](tensor)
return tensor
def zeros_(tensor: torch.Tensor) -> torch.Tensor:
if not getattr(tensor, "_is_hf_initialized", False):
return TORCH_INIT_FUNCTIONS["zeros_"](tensor)
return tensor
def eye_(tensor: torch.Tensor) -> torch.Tensor:
if not getattr(tensor, "_is_hf_initialized", False):
return TORCH_INIT_FUNCTIONS["eye_"](tensor)
return tensor
def dirac_(tensor: torch.Tensor, groups: int = 1) -> torch.Tensor:
if not getattr(tensor, "_is_hf_initialized", False):
return TORCH_INIT_FUNCTIONS["dirac_"](tensor, groups=groups)
return tensor
def xavier_uniform_(tensor: torch.Tensor, gain: float = 1.0, generator: torch.Generator | None = None) -> torch.Tensor:
if not getattr(tensor, "_is_hf_initialized", False):
return TORCH_INIT_FUNCTIONS["xavier_uniform_"](tensor, gain=gain, generator=generator)
return tensor
def xavier_normal_(tensor: torch.Tensor, gain: float = 1.0, generator: torch.Generator | None = None) -> torch.Tensor:
if not getattr(tensor, "_is_hf_initialized", False):
return TORCH_INIT_FUNCTIONS["xavier_normal_"](tensor, gain=gain, generator=generator)
return tensor
def kaiming_uniform_(
tensor: torch.Tensor,
a: float = 0,
mode: str = "fan_in",
nonlinearity: str = "leaky_relu",
generator: torch.Generator | None = None,
) -> torch.Tensor:
if not getattr(tensor, "_is_hf_initialized", False):
return TORCH_INIT_FUNCTIONS["kaiming_uniform_"](
tensor, a=a, mode=mode, nonlinearity=nonlinearity, generator=generator
)
return tensor
def kaiming_normal_(
tensor: torch.Tensor,
a: float = 0,
mode: str = "fan_in",
nonlinearity: str = "leaky_relu",
generator: torch.Generator | None = None,
) -> torch.Tensor:
if not getattr(tensor, "_is_hf_initialized", False):
return TORCH_INIT_FUNCTIONS["kaiming_normal_"](
tensor, a=a, mode=mode, nonlinearity=nonlinearity, generator=generator
)
return tensor
def trunc_normal_(
tensor: torch.Tensor,
mean: float = 0.0,
std: float = 1.0,
a: float = -2.0,
b: float = 2.0,
generator: torch.Generator | None = None,
) -> torch.Tensor:
if not getattr(tensor, "_is_hf_initialized", False):
return TORCH_INIT_FUNCTIONS["trunc_normal_"](tensor, mean=mean, std=std, a=a, b=b, generator=generator)
return tensor
def orthogonal_(
tensor: torch.Tensor,
gain: float = 1,
generator: torch.Generator | None = None,
) -> torch.Tensor:
if not getattr(tensor, "_is_hf_initialized", False):
return TORCH_INIT_FUNCTIONS["orthogonal_"](tensor, gain=gain, generator=generator)
return tensor
def sparse_(
tensor: torch.Tensor, sparsity: float, std: float = 0.01, generator: torch.Generator | None = None
) -> torch.Tensor:
if not getattr(tensor, "_is_hf_initialized", False):
return TORCH_INIT_FUNCTIONS["sparse_"](tensor, sparsity=sparsity, std=std, generator=generator)
return tensor
def copy_(tensor: torch.Tensor, other: torch.Tensor) -> torch.Tensor:
if not getattr(tensor, "_is_hf_initialized", False):
with torch.no_grad():
return tensor.copy_(other)
return tensor
def _variance_scaling(tensor, mode="fan_in", distribution="normal"):
fan_in, fan_out = torch.nn.init._calculate_fan_in_and_fan_out(tensor)
if mode == "fan_in":
denom = fan_in
elif mode == "fan_out":
denom = fan_out
elif mode == "fan_avg":
denom = (fan_in + fan_out) / 2
variance = 1.0 / denom
if distribution == "truncated_normal":
trunc_normal_(tensor, std=math.sqrt(variance) / 0.87962566103423978)
elif distribution == "normal":
normal_(tensor, std=math.sqrt(variance))
elif distribution == "uniform":
bound = math.sqrt(3 * variance)
uniform_(tensor, -bound, bound)
else:
raise ValueError(f"invalid distribution {distribution}")
def lecun_normal_(tensor):
if not getattr(tensor, "_is_hf_initialized", False):
_variance_scaling(tensor, mode="fan_in", distribution="truncated_normal")
return tensor
def default_flax_embed_init_(tensor):
if not getattr(tensor, "_is_hf_initialized", False):
_variance_scaling(tensor, mode="fan_in", distribution="normal")
return tensor
# Here, we need to check several modules imported, and hot patch all of them, as sometimes torch does
# something like `from torch.nn.init import xavier_uniform_` in their internals (e.g in torch.nn.modules.activations,
# where MultiHeadAttention lives), so the function name is binded at import time and just doing
# `setattr(torch.nn.init, name, globals()[name])` is thus not enough
# The following list should be enough for all torch versions we work with
TORCH_MODULES_TO_PATCH = (
"torch.nn.init",
"torch.nn.modules.activation",
"torch.nn.modules.transformer",
"torch.nn.modules.linear",
"torch.nn.modules.loss",
"torch.nn.modules.batchnorm",
"torch.nn.modules.conv",
"torch.nn.modules.normalization",
"torch.nn.modules.rnn",
"torch.nn.modules.sparse",
)
@contextmanager
def guard_torch_init_functions():
"""
Guard the `torch.nn.init` primitive functions to behave exactly like the functions in this file, i.e. be
protected against the `_is_hf_initialized` flag to avoid re-init if the param was already loaded.
Usually, all models are using the init from `transformers` which are already guarded, but just to make extra sure
and for remote code, we also use this context manager.
"""
originals = defaultdict(dict)
try:
# Replace all torch funcs by the ones in this file
for module_name in TORCH_MODULES_TO_PATCH:
if module_name in sys.modules:
module = sys.modules[module_name]
for func_name in TORCH_INIT_FUNCTIONS.keys():
if hasattr(module, func_name):
originals[module][func_name] = getattr(module, func_name)
setattr(module, func_name, globals()[func_name])
yield
finally:
# Set back the original functions on all modules
for module, functions in originals.items():
for func_name, func in functions.items():
setattr(module, func_name, func)
@contextmanager
def no_init_weights():
"""
Disable weight initialization both at the torch-level, and at the transformers-level (`init_weights`).
This is used to speed-up initializing an empty model with deepspeed, as we do not initialize the model on meta device
with deepspeed, but we still don't need to run expensive weight initializations as we are loading params afterwards.
"""
from .modeling_utils import PreTrainedModel
def empty_func(*args, **kwargs):
pass
originals = defaultdict(dict)
try:
# Replace all torch funcs by empty ones
for module_name in TORCH_MODULES_TO_PATCH:
if module_name in sys.modules:
module = sys.modules[module_name]
for func_name in TORCH_INIT_FUNCTIONS.keys():
if hasattr(module, func_name):
originals[module][func_name] = getattr(module, func_name)
setattr(module, func_name, empty_func)
# Also patch our own `init_weights`
original_init_weights = PreTrainedModel.init_weights
PreTrainedModel.init_weights = empty_func
yield
finally:
# Set back the original torch functions on all modules
for module, functions in originals.items():
for func_name, func in functions.items():
setattr(module, func_name, func)
# Set back `init_weights`
PreTrainedModel.init_weights = original_init_weights
@contextmanager
def no_tie_weights():
"""
Disable weight tying during loading with `from_pretrained`. This is needed as we want to have access to ALL
weights in the state_dict during `from_pretrained`, and otherwise tying them would remove them from it, as it's
called in `post_init` when instantiating.
"""
from .modeling_utils import PreTrainedModel
def empty_func(*args, **kwargs):
pass
try:
original_tie_weights = PreTrainedModel.tie_weights
PreTrainedModel.tie_weights = empty_func
yield
finally:
# Set back the original
PreTrainedModel.tie_weights = original_tie_weights
@contextmanager
def meta_device_safe_creation_ops():
"""
During meta-device model initialisation, ``torch.linspace`` produces meta
tensors that have no data. Custom models loaded from the Hub (remote code)
often call ``.item()`` on these tensors to compute scalar hyperparameters
(e.g. stochastic-depth / drop-path schedules). Native transformers models
already pass ``device="cpu"`` explicitly for such calls (see e.g.
``modeling_swin.py``, ``modeling_pvt_v2.py``), but remote-code models
written before v5 do not.
This context manager patches ``torch.linspace`` to default to
``device="cpu"`` when no explicit device is requested, matching the best
practice already used throughout transformers. Calls that supply an
explicit ``device`` argument (e.g. ``device=self.logits.device``) are left
untouched. ``torch.arange`` is intentionally NOT patched because it is
used in RoPE computations where the device must match model parameters.
"""
original_linspace = torch.linspace
def _safe_linspace(*args, **kwargs):
kwargs.setdefault("device", "cpu")
return original_linspace(*args, **kwargs)
torch.linspace = _safe_linspace
try:
yield
finally:
torch.linspace = original_linspace
|