Commit
·
e0ea168
1
Parent(s):
f9b3adb
refactor: lora
Browse filesSigned-off-by: jupyterjazz <[email protected]>
- configuration_xlm_roberta.py +8 -2
- modeling_lora.py +79 -67
- modeling_xlm_roberta.py +25 -2
configuration_xlm_roberta.py
CHANGED
|
@@ -22,7 +22,10 @@ class XLMRobertaFlashConfig(PretrainedConfig):
|
|
| 22 |
position_embedding_type="absolute",
|
| 23 |
use_cache=True,
|
| 24 |
classifier_dropout=None,
|
| 25 |
-
|
|
|
|
|
|
|
|
|
|
| 26 |
load_trained_adapters=False,
|
| 27 |
use_flash_attn=True,
|
| 28 |
torch_dtype=None,
|
|
@@ -47,8 +50,11 @@ class XLMRobertaFlashConfig(PretrainedConfig):
|
|
| 47 |
self.position_embedding_type = position_embedding_type
|
| 48 |
self.use_cache = use_cache
|
| 49 |
self.classifier_dropout = classifier_dropout
|
| 50 |
-
self.num_loras = num_loras
|
| 51 |
self.load_trained_adapters = load_trained_adapters
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
self.use_flash_attn = use_flash_attn
|
| 53 |
self.emb_pooler = emb_pooler
|
| 54 |
if torch_dtype and hasattr(torch, torch_dtype) and type(getattr(torch, torch_dtype)) is torch.dtype:
|
|
|
|
| 22 |
position_embedding_type="absolute",
|
| 23 |
use_cache=True,
|
| 24 |
classifier_dropout=None,
|
| 25 |
+
lora_adaptations=None,
|
| 26 |
+
lora_rank=4,
|
| 27 |
+
lora_dropout_p=0.0,
|
| 28 |
+
lora_alpha=1,
|
| 29 |
load_trained_adapters=False,
|
| 30 |
use_flash_attn=True,
|
| 31 |
torch_dtype=None,
|
|
|
|
| 50 |
self.position_embedding_type = position_embedding_type
|
| 51 |
self.use_cache = use_cache
|
| 52 |
self.classifier_dropout = classifier_dropout
|
|
|
|
| 53 |
self.load_trained_adapters = load_trained_adapters
|
| 54 |
+
self.lora_adaptations = lora_adaptations
|
| 55 |
+
self.lora_rank = lora_rank
|
| 56 |
+
self.lora_dropout_p = lora_dropout_p
|
| 57 |
+
self.lora_alpha = lora_alpha
|
| 58 |
self.use_flash_attn = use_flash_attn
|
| 59 |
self.emb_pooler = emb_pooler
|
| 60 |
if torch_dtype and hasattr(torch, torch_dtype) and type(getattr(torch, torch_dtype)) is torch.dtype:
|
modeling_lora.py
CHANGED
|
@@ -9,14 +9,18 @@ from torch import nn
|
|
| 9 |
from torch.nn import Parameter
|
| 10 |
from transformers import PretrainedConfig
|
| 11 |
|
| 12 |
-
from .modeling_xlm_roberta import
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
|
| 15 |
def initialized_weights(
|
| 16 |
-
shape: Tuple[int],
|
| 17 |
) -> torch.Tensor:
|
| 18 |
weight_data = []
|
| 19 |
-
for _ in range(
|
| 20 |
new_adaption = torch.zeros(shape)
|
| 21 |
if init == "kaiming":
|
| 22 |
nn.init.kaiming_uniform_(new_adaption, a=math.sqrt(5))
|
|
@@ -45,15 +49,16 @@ class LoRAParametrization(nn.Module):
|
|
| 45 |
WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
|
| 46 |
SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
|
| 47 |
"""
|
|
|
|
| 48 |
def __init__(
|
| 49 |
self,
|
| 50 |
fan_in: int,
|
| 51 |
fan_out: int,
|
| 52 |
layer_type: str = "linear",
|
| 53 |
-
|
| 54 |
rank: int = 4,
|
| 55 |
-
|
| 56 |
-
|
| 57 |
):
|
| 58 |
super().__init__()
|
| 59 |
# if weight is stored as (fan_out, fan_in), the memory layout of A & B follows (W + BA)x
|
|
@@ -63,25 +68,23 @@ class LoRAParametrization(nn.Module):
|
|
| 63 |
|
| 64 |
if layer_type == "linear":
|
| 65 |
self.lora_A = nn.Parameter(
|
| 66 |
-
initialized_weights((rank, fan_in),
|
| 67 |
)
|
| 68 |
-
self.lora_B = nn.Parameter(torch.zeros((
|
| 69 |
elif layer_type == "embedding":
|
| 70 |
-
self.lora_A = nn.Parameter(torch.zeros((
|
| 71 |
self.lora_B = nn.Parameter(
|
| 72 |
initialized_weights(
|
| 73 |
-
(rank, fan_out),
|
| 74 |
)
|
| 75 |
)
|
| 76 |
else:
|
| 77 |
raise NotImplementedError
|
| 78 |
|
| 79 |
-
self.lora_alpha, self.rank =
|
| 80 |
-
self.scaling =
|
| 81 |
-
self.lora_dropout = (
|
| 82 |
-
|
| 83 |
-
)
|
| 84 |
-
self.dropout_fn = self._dropout if lora_dropout_p > 0 else lambda x: x
|
| 85 |
self.register_buffer(
|
| 86 |
"lora_dropout_mask",
|
| 87 |
torch.ones(self.swap((1, fan_in)), dtype=self.lora_A.dtype),
|
|
@@ -128,42 +131,52 @@ class LoRAParametrization(nn.Module):
|
|
| 128 |
def from_linear(
|
| 129 |
cls,
|
| 130 |
layer: nn.Module,
|
| 131 |
-
|
| 132 |
-
rank: int
|
| 133 |
-
|
| 134 |
-
|
| 135 |
):
|
| 136 |
assert isinstance(layer, nn.Linear)
|
| 137 |
fan_out, fan_in = layer.weight.shape
|
| 138 |
return cls(
|
| 139 |
fan_in,
|
| 140 |
fan_out,
|
| 141 |
-
|
| 142 |
layer_type="linear",
|
| 143 |
rank=rank,
|
| 144 |
-
|
| 145 |
-
|
| 146 |
)
|
| 147 |
|
| 148 |
@classmethod
|
| 149 |
def from_embedding(
|
| 150 |
-
cls,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 151 |
):
|
| 152 |
assert isinstance(layer, nn.Embedding)
|
| 153 |
fan_in, fan_out = layer.weight.shape
|
| 154 |
return cls(
|
| 155 |
fan_in,
|
| 156 |
fan_out,
|
| 157 |
-
|
| 158 |
layer_type="embedding",
|
| 159 |
rank=rank,
|
| 160 |
-
|
| 161 |
-
|
| 162 |
)
|
| 163 |
|
| 164 |
@classmethod
|
| 165 |
def add_to_layer(
|
| 166 |
-
cls,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 167 |
):
|
| 168 |
if isinstance(layer, nn.Linear):
|
| 169 |
parametrize.register_parametrization(
|
|
@@ -171,10 +184,10 @@ class LoRAParametrization(nn.Module):
|
|
| 171 |
"weight",
|
| 172 |
cls.from_linear(
|
| 173 |
layer,
|
| 174 |
-
|
| 175 |
rank=rank,
|
| 176 |
-
|
| 177 |
-
|
| 178 |
),
|
| 179 |
)
|
| 180 |
elif isinstance(layer, nn.Embedding):
|
|
@@ -183,10 +196,10 @@ class LoRAParametrization(nn.Module):
|
|
| 183 |
"weight",
|
| 184 |
cls.from_embedding(
|
| 185 |
layer,
|
| 186 |
-
|
| 187 |
rank=rank,
|
| 188 |
-
|
| 189 |
-
|
| 190 |
),
|
| 191 |
)
|
| 192 |
|
|
@@ -195,15 +208,14 @@ class LoRAParametrization(nn.Module):
|
|
| 195 |
if isinstance(layer, LoRAParametrization):
|
| 196 |
layer.current_task = task_idx
|
| 197 |
|
| 198 |
-
@staticmethod
|
| 199 |
-
def merge_lora_into_layer(layer: nn.Module):
|
| 200 |
-
if hasattr(layer, "parametrizations"):
|
| 201 |
-
for attr_name in layer.parametrizations.keys():
|
| 202 |
-
parametrize.remove_parametrizations(layer, attr_name, leave_parametrized=True)
|
| 203 |
-
|
| 204 |
|
| 205 |
class XLMRobertaLoRA(XLMRobertaPreTrainedModel):
|
| 206 |
-
def __init__(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 207 |
super().__init__(config)
|
| 208 |
|
| 209 |
if roberta is None:
|
|
@@ -211,10 +223,17 @@ class XLMRobertaLoRA(XLMRobertaPreTrainedModel):
|
|
| 211 |
else:
|
| 212 |
self.roberta = roberta
|
| 213 |
|
| 214 |
-
self.
|
| 215 |
-
self.
|
| 216 |
-
self.
|
|
|
|
| 217 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 218 |
self.main_params_trainable = False
|
| 219 |
self._task_idx = None
|
| 220 |
# By default, we select the first LoRA
|
|
@@ -237,13 +256,6 @@ class XLMRobertaLoRA(XLMRobertaPreTrainedModel):
|
|
| 237 |
if "lora" not in name:
|
| 238 |
param.requires_grad_(val)
|
| 239 |
|
| 240 |
-
def merge_lora(self):
|
| 241 |
-
"""Merges currently selected LoRA into main weights."""
|
| 242 |
-
if self._is_merged:
|
| 243 |
-
raise Exception('LoRA has already been merged, cannot merge again')
|
| 244 |
-
self._is_merged = True
|
| 245 |
-
self.apply(LoRAParametrization.merge_lora_into_layer)
|
| 246 |
-
|
| 247 |
@classmethod
|
| 248 |
def from_pretrained(
|
| 249 |
cls,
|
|
@@ -259,31 +271,33 @@ class XLMRobertaLoRA(XLMRobertaPreTrainedModel):
|
|
| 259 |
use_safetensors: bool = None,
|
| 260 |
**kwargs,
|
| 261 |
):
|
| 262 |
-
config = XLMRobertaFlashConfig.from_pretrained(
|
|
|
|
|
|
|
| 263 |
if config.load_trained_adapters:
|
| 264 |
return super().from_pretrained(
|
| 265 |
-
pretrained_model_name_or_path,
|
| 266 |
-
*model_args,
|
| 267 |
-
**kwargs
|
| 268 |
)
|
| 269 |
else:
|
| 270 |
-
roberta = XLMRobertaModel.from_pretrained(
|
|
|
|
|
|
|
| 271 |
return cls(config, roberta=roberta)
|
| 272 |
|
| 273 |
-
def _register_lora(self,
|
| 274 |
self.apply(
|
| 275 |
partial(
|
| 276 |
LoRAParametrization.add_to_layer,
|
| 277 |
-
|
| 278 |
rank=rank,
|
| 279 |
-
|
| 280 |
-
|
| 281 |
)
|
| 282 |
)
|
| 283 |
|
| 284 |
@property
|
| 285 |
def current_task(self):
|
| 286 |
-
"""
|
| 287 |
:return: Integer or None (when LoRA is disabled)
|
| 288 |
"""
|
| 289 |
return self._task_idx
|
|
@@ -296,9 +310,7 @@ class XLMRobertaLoRA(XLMRobertaPreTrainedModel):
|
|
| 296 |
:param task_idx: Which LoRA to use
|
| 297 |
:return:
|
| 298 |
"""
|
| 299 |
-
|
| 300 |
-
raise Exception('LoRA has been merged, cannot select new task')
|
| 301 |
-
assert task_idx is None or 0 <= task_idx < self._num_adaptions
|
| 302 |
if self._task_idx != task_idx:
|
| 303 |
# In this case, we need to update the LoRAs everywhere
|
| 304 |
self._task_idx = task_idx
|
|
@@ -306,9 +318,9 @@ class XLMRobertaLoRA(XLMRobertaPreTrainedModel):
|
|
| 306 |
partial(LoRAParametrization.select_task_for_layer, task_idx=task_idx)
|
| 307 |
)
|
| 308 |
|
| 309 |
-
def forward(self, *args,
|
| 310 |
-
if
|
| 311 |
-
self.current_task =
|
| 312 |
return self.roberta(*args, **kwargs)
|
| 313 |
|
| 314 |
def parameters(self, recurse: bool = True) -> Iterator[Parameter]:
|
|
|
|
| 9 |
from torch.nn import Parameter
|
| 10 |
from transformers import PretrainedConfig
|
| 11 |
|
| 12 |
+
from .modeling_xlm_roberta import (
|
| 13 |
+
XLMRobertaFlashConfig,
|
| 14 |
+
XLMRobertaModel,
|
| 15 |
+
XLMRobertaPreTrainedModel,
|
| 16 |
+
)
|
| 17 |
|
| 18 |
|
| 19 |
def initialized_weights(
|
| 20 |
+
shape: Tuple[int], num_adaptations: int, init: str = "kaiming"
|
| 21 |
) -> torch.Tensor:
|
| 22 |
weight_data = []
|
| 23 |
+
for _ in range(num_adaptations):
|
| 24 |
new_adaption = torch.zeros(shape)
|
| 25 |
if init == "kaiming":
|
| 26 |
nn.init.kaiming_uniform_(new_adaption, a=math.sqrt(5))
|
|
|
|
| 49 |
WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
|
| 50 |
SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
|
| 51 |
"""
|
| 52 |
+
|
| 53 |
def __init__(
|
| 54 |
self,
|
| 55 |
fan_in: int,
|
| 56 |
fan_out: int,
|
| 57 |
layer_type: str = "linear",
|
| 58 |
+
num_adaptations: int = 1,
|
| 59 |
rank: int = 4,
|
| 60 |
+
dropout_p: float = 0.0,
|
| 61 |
+
alpha: float = 1,
|
| 62 |
):
|
| 63 |
super().__init__()
|
| 64 |
# if weight is stored as (fan_out, fan_in), the memory layout of A & B follows (W + BA)x
|
|
|
|
| 68 |
|
| 69 |
if layer_type == "linear":
|
| 70 |
self.lora_A = nn.Parameter(
|
| 71 |
+
initialized_weights((rank, fan_in), num_adaptations, init="kaiming")
|
| 72 |
)
|
| 73 |
+
self.lora_B = nn.Parameter(torch.zeros((num_adaptations, fan_out, rank)))
|
| 74 |
elif layer_type == "embedding":
|
| 75 |
+
self.lora_A = nn.Parameter(torch.zeros((num_adaptations, fan_in, rank)))
|
| 76 |
self.lora_B = nn.Parameter(
|
| 77 |
initialized_weights(
|
| 78 |
+
(rank, fan_out), num_adaptations=num_adaptations, init="normal"
|
| 79 |
)
|
| 80 |
)
|
| 81 |
else:
|
| 82 |
raise NotImplementedError
|
| 83 |
|
| 84 |
+
self.lora_alpha, self.rank = alpha, rank
|
| 85 |
+
self.scaling = alpha / rank
|
| 86 |
+
self.lora_dropout = nn.Dropout(p=dropout_p) if dropout_p > 0 else lambda x: x
|
| 87 |
+
self.dropout_fn = self._dropout if dropout_p > 0 else lambda x: x
|
|
|
|
|
|
|
| 88 |
self.register_buffer(
|
| 89 |
"lora_dropout_mask",
|
| 90 |
torch.ones(self.swap((1, fan_in)), dtype=self.lora_A.dtype),
|
|
|
|
| 131 |
def from_linear(
|
| 132 |
cls,
|
| 133 |
layer: nn.Module,
|
| 134 |
+
num_adaptations: int,
|
| 135 |
+
rank: int,
|
| 136 |
+
dropout_p: float,
|
| 137 |
+
alpha: float,
|
| 138 |
):
|
| 139 |
assert isinstance(layer, nn.Linear)
|
| 140 |
fan_out, fan_in = layer.weight.shape
|
| 141 |
return cls(
|
| 142 |
fan_in,
|
| 143 |
fan_out,
|
| 144 |
+
num_adaptations=num_adaptations,
|
| 145 |
layer_type="linear",
|
| 146 |
rank=rank,
|
| 147 |
+
dropout_p=dropout_p,
|
| 148 |
+
alpha=alpha,
|
| 149 |
)
|
| 150 |
|
| 151 |
@classmethod
|
| 152 |
def from_embedding(
|
| 153 |
+
cls,
|
| 154 |
+
layer: nn.Module,
|
| 155 |
+
num_adaptations: int,
|
| 156 |
+
rank: int,
|
| 157 |
+
dropout_p: float,
|
| 158 |
+
alpha: float,
|
| 159 |
):
|
| 160 |
assert isinstance(layer, nn.Embedding)
|
| 161 |
fan_in, fan_out = layer.weight.shape
|
| 162 |
return cls(
|
| 163 |
fan_in,
|
| 164 |
fan_out,
|
| 165 |
+
num_adaptations=num_adaptations,
|
| 166 |
layer_type="embedding",
|
| 167 |
rank=rank,
|
| 168 |
+
dropout_p=dropout_p,
|
| 169 |
+
alpha=alpha,
|
| 170 |
)
|
| 171 |
|
| 172 |
@classmethod
|
| 173 |
def add_to_layer(
|
| 174 |
+
cls,
|
| 175 |
+
layer: nn.Module,
|
| 176 |
+
num_adaptations: int,
|
| 177 |
+
rank: int,
|
| 178 |
+
dropout_p: float,
|
| 179 |
+
alpha: float,
|
| 180 |
):
|
| 181 |
if isinstance(layer, nn.Linear):
|
| 182 |
parametrize.register_parametrization(
|
|
|
|
| 184 |
"weight",
|
| 185 |
cls.from_linear(
|
| 186 |
layer,
|
| 187 |
+
num_adaptations=num_adaptations,
|
| 188 |
rank=rank,
|
| 189 |
+
dropout_p=dropout_p,
|
| 190 |
+
alpha=alpha,
|
| 191 |
),
|
| 192 |
)
|
| 193 |
elif isinstance(layer, nn.Embedding):
|
|
|
|
| 196 |
"weight",
|
| 197 |
cls.from_embedding(
|
| 198 |
layer,
|
| 199 |
+
num_adaptations=num_adaptations,
|
| 200 |
rank=rank,
|
| 201 |
+
dropout_p=dropout_p,
|
| 202 |
+
alpha=alpha,
|
| 203 |
),
|
| 204 |
)
|
| 205 |
|
|
|
|
| 208 |
if isinstance(layer, LoRAParametrization):
|
| 209 |
layer.current_task = task_idx
|
| 210 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 211 |
|
| 212 |
class XLMRobertaLoRA(XLMRobertaPreTrainedModel):
|
| 213 |
+
def __init__(
|
| 214 |
+
self,
|
| 215 |
+
config: XLMRobertaFlashConfig,
|
| 216 |
+
roberta: Optional[XLMRobertaModel] = None,
|
| 217 |
+
add_pooling_layer=True,
|
| 218 |
+
):
|
| 219 |
super().__init__(config)
|
| 220 |
|
| 221 |
if roberta is None:
|
|
|
|
| 223 |
else:
|
| 224 |
self.roberta = roberta
|
| 225 |
|
| 226 |
+
self._num_adaptations = len(config.lora_adaptations)
|
| 227 |
+
self._rank = config.lora_rank
|
| 228 |
+
self._dropout_p = config.lora_dropout_p
|
| 229 |
+
self._alpha = config.lora_alpha
|
| 230 |
|
| 231 |
+
self._register_lora(
|
| 232 |
+
num_adaptations=self._num_adaptations,
|
| 233 |
+
rank=self._rank,
|
| 234 |
+
dropout_p=self._dropout_p,
|
| 235 |
+
alpha=self._alpha,
|
| 236 |
+
)
|
| 237 |
self.main_params_trainable = False
|
| 238 |
self._task_idx = None
|
| 239 |
# By default, we select the first LoRA
|
|
|
|
| 256 |
if "lora" not in name:
|
| 257 |
param.requires_grad_(val)
|
| 258 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 259 |
@classmethod
|
| 260 |
def from_pretrained(
|
| 261 |
cls,
|
|
|
|
| 271 |
use_safetensors: bool = None,
|
| 272 |
**kwargs,
|
| 273 |
):
|
| 274 |
+
config = XLMRobertaFlashConfig.from_pretrained(
|
| 275 |
+
pretrained_model_name_or_path, *model_args, **kwargs
|
| 276 |
+
)
|
| 277 |
if config.load_trained_adapters:
|
| 278 |
return super().from_pretrained(
|
| 279 |
+
pretrained_model_name_or_path, *model_args, **kwargs
|
|
|
|
|
|
|
| 280 |
)
|
| 281 |
else:
|
| 282 |
+
roberta = XLMRobertaModel.from_pretrained(
|
| 283 |
+
pretrained_model_name_or_path, *model_args, **kwargs
|
| 284 |
+
)
|
| 285 |
return cls(config, roberta=roberta)
|
| 286 |
|
| 287 |
+
def _register_lora(self, num_adaptations, rank, dropout_p, alpha):
|
| 288 |
self.apply(
|
| 289 |
partial(
|
| 290 |
LoRAParametrization.add_to_layer,
|
| 291 |
+
num_adaptations=num_adaptations,
|
| 292 |
rank=rank,
|
| 293 |
+
dropout_p=dropout_p,
|
| 294 |
+
alpha=alpha,
|
| 295 |
)
|
| 296 |
)
|
| 297 |
|
| 298 |
@property
|
| 299 |
def current_task(self):
|
| 300 |
+
"""Which LoRA is currently selected
|
| 301 |
:return: Integer or None (when LoRA is disabled)
|
| 302 |
"""
|
| 303 |
return self._task_idx
|
|
|
|
| 310 |
:param task_idx: Which LoRA to use
|
| 311 |
:return:
|
| 312 |
"""
|
| 313 |
+
assert task_idx is None or 0 <= task_idx < self._num_adaptations
|
|
|
|
|
|
|
| 314 |
if self._task_idx != task_idx:
|
| 315 |
# In this case, we need to update the LoRAs everywhere
|
| 316 |
self._task_idx = task_idx
|
|
|
|
| 318 |
partial(LoRAParametrization.select_task_for_layer, task_idx=task_idx)
|
| 319 |
)
|
| 320 |
|
| 321 |
+
def forward(self, *args, lora_adaptation: Union[None, int] = -1, **kwargs):
|
| 322 |
+
if lora_adaptation is None or lora_adaptation >= 0:
|
| 323 |
+
self.current_task = lora_adaptation
|
| 324 |
return self.roberta(*args, **kwargs)
|
| 325 |
|
| 326 |
def parameters(self, recurse: bool = True) -> Iterator[Parameter]:
|
modeling_xlm_roberta.py
CHANGED
|
@@ -452,6 +452,7 @@ class XLMRobertaModel(XLMRobertaPreTrainedModel):
|
|
| 452 |
convert_to_tensor: bool = False,
|
| 453 |
device: Optional[torch.device] = None,
|
| 454 |
normalize_embeddings: bool = False,
|
|
|
|
| 455 |
**tokenizer_kwargs,
|
| 456 |
) -> Union[List[torch.Tensor], np.ndarray, torch.Tensor]:
|
| 457 |
"""
|
|
@@ -481,6 +482,12 @@ class XLMRobertaModel(XLMRobertaPreTrainedModel):
|
|
| 481 |
If set to true, returned vectors will have length 1. In that case, the
|
| 482 |
faster dot-product (util.dot_score) instead of cosine similarity can
|
| 483 |
be used.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 484 |
tokenizer_kwargs(`Dict[str, Any]`, *optional*, defaults to {}):
|
| 485 |
Keyword arguments for the tokenizer
|
| 486 |
Returns:
|
|
@@ -518,6 +525,22 @@ class XLMRobertaModel(XLMRobertaPreTrainedModel):
|
|
| 518 |
if device is not None:
|
| 519 |
self.to(device)
|
| 520 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 521 |
permutation = np.argsort([-len(i) for i in sentences])
|
| 522 |
inverse_permutation = np.argsort(permutation)
|
| 523 |
sentences = [sentences[idx] for idx in permutation]
|
|
@@ -547,7 +570,7 @@ class XLMRobertaModel(XLMRobertaPreTrainedModel):
|
|
| 547 |
return_tensors='pt',
|
| 548 |
**tokenizer_kwargs,
|
| 549 |
).to(self.device)
|
| 550 |
-
token_embs = self.forward(**encoded_input)[0]
|
| 551 |
|
| 552 |
# Accumulate in fp32 to avoid overflow
|
| 553 |
token_embs = token_embs.float()
|
|
@@ -1253,4 +1276,4 @@ class XLMRobertaForSequenceClassification(XLMRobertaPreTrainedModel):
|
|
| 1253 |
logits=logits,
|
| 1254 |
hidden_states=outputs.hidden_states,
|
| 1255 |
attentions=outputs.attentions,
|
| 1256 |
-
)
|
|
|
|
| 452 |
convert_to_tensor: bool = False,
|
| 453 |
device: Optional[torch.device] = None,
|
| 454 |
normalize_embeddings: bool = False,
|
| 455 |
+
task: Optional[str] = None,
|
| 456 |
**tokenizer_kwargs,
|
| 457 |
) -> Union[List[torch.Tensor], np.ndarray, torch.Tensor]:
|
| 458 |
"""
|
|
|
|
| 482 |
If set to true, returned vectors will have length 1. In that case, the
|
| 483 |
faster dot-product (util.dot_score) instead of cosine similarity can
|
| 484 |
be used.
|
| 485 |
+
task(`str`, *optional*, defaults to None):
|
| 486 |
+
Specifies the task for which the encoding is intended. This
|
| 487 |
+
controls the use of specialized LoRA adapters that are tuned for specific tasks.
|
| 488 |
+
If provided, the corresponding LoRA adapter is enabled, enhancing the model's
|
| 489 |
+
performance for that task. If `None` or not provided, LoRA is disabled, and the
|
| 490 |
+
model uses its original, general-purpose weights.
|
| 491 |
tokenizer_kwargs(`Dict[str, Any]`, *optional*, defaults to {}):
|
| 492 |
Keyword arguments for the tokenizer
|
| 493 |
Returns:
|
|
|
|
| 525 |
if device is not None:
|
| 526 |
self.to(device)
|
| 527 |
|
| 528 |
+
lora_adapter_num = None
|
| 529 |
+
if self.config.lora_adaptations:
|
| 530 |
+
if task:
|
| 531 |
+
if task in self.config.lora_adaptations:
|
| 532 |
+
lora_adapter_num = self.config.lora_adaptations.index(task)
|
| 533 |
+
else:
|
| 534 |
+
raise ValueError(
|
| 535 |
+
f"Unsupported task '{task}'. "
|
| 536 |
+
f"Supported tasks are: {', '.join(self.config.lora_adaptations)}.")
|
| 537 |
+
else:
|
| 538 |
+
logger.warning(
|
| 539 |
+
f"Task-specific embeddings are disabled. To enable, specify the `task` "
|
| 540 |
+
f"argument with one of the supported tasks: {', '.join(self.config.lora_adaptations)}"
|
| 541 |
+
)
|
| 542 |
+
|
| 543 |
+
|
| 544 |
permutation = np.argsort([-len(i) for i in sentences])
|
| 545 |
inverse_permutation = np.argsort(permutation)
|
| 546 |
sentences = [sentences[idx] for idx in permutation]
|
|
|
|
| 570 |
return_tensors='pt',
|
| 571 |
**tokenizer_kwargs,
|
| 572 |
).to(self.device)
|
| 573 |
+
token_embs = self.forward(**encoded_input, lora_adaptation=lora_adapter_num)[0]
|
| 574 |
|
| 575 |
# Accumulate in fp32 to avoid overflow
|
| 576 |
token_embs = token_embs.float()
|
|
|
|
| 1276 |
logits=logits,
|
| 1277 |
hidden_states=outputs.hidden_states,
|
| 1278 |
attentions=outputs.attentions,
|
| 1279 |
+
)
|