michael-guenther
commited on
Commit
β’
95b4916
1
Parent(s):
eb21270
add mlm model and adjust naming
Browse files- README.md +5 -0
- config.json +4 -4
- configuration_bert.py β configuration_xlm_roberta.py +1 -1
- convert_roberta_weights_to_flash.py +29 -44
- embedding.py +1 -1
- modeling_bert.py β modeling_xlm_roberta.py +210 -148
- pytorch_model.bin +2 -2
- bert_padding.py β xlm_padding.py +0 -0
README.md
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Converting Weights
|
2 |
+
|
3 |
+
```
|
4 |
+
python3 -m "xlm-roberta-flash-implementation".convert_roberta_weights_to_flash --output pytorch_model_xlmr_flash.bin
|
5 |
+
```
|
config.json
CHANGED
@@ -1,9 +1,9 @@
|
|
1 |
{
|
2 |
"auto_map": {
|
3 |
-
"AutoConfig": "
|
4 |
-
"AutoModel": "
|
5 |
-
"AutoModelForPreTraining": "
|
6 |
-
"AutoModelForMaskedLM": "
|
7 |
},
|
8 |
"attention_probs_dropout_prob": 0.1,
|
9 |
"bos_token_id": 0,
|
|
|
1 |
{
|
2 |
"auto_map": {
|
3 |
+
"AutoConfig": "configuration_xlm_roberta.XLMRobertaFlashConfig",
|
4 |
+
"AutoModel": "modeling_xlm_roberta.XLMRobertaModel",
|
5 |
+
"AutoModelForPreTraining": "modeling_xlm_roberta.XLMRobertaForPreTraining",
|
6 |
+
"AutoModelForMaskedLM": "modeling_xlm_roberta.XLMRobertaForMaskedLM"
|
7 |
},
|
8 |
"attention_probs_dropout_prob": 0.1,
|
9 |
"bos_token_id": 0,
|
configuration_bert.py β configuration_xlm_roberta.py
RENAMED
@@ -1,6 +1,6 @@
|
|
1 |
from transformers import PretrainedConfig
|
2 |
|
3 |
-
class
|
4 |
def __init__(
|
5 |
self,
|
6 |
vocab_size=30522,
|
|
|
1 |
from transformers import PretrainedConfig
|
2 |
|
3 |
+
class XLMRobertaFlashConfig(PretrainedConfig):
|
4 |
def __init__(
|
5 |
self,
|
6 |
vocab_size=30522,
|
convert_roberta_weights_to_flash.py
CHANGED
@@ -1,9 +1,10 @@
|
|
1 |
import re
|
2 |
from collections import OrderedDict
|
3 |
-
from transformers import
|
4 |
from transformers import XLMRobertaForMaskedLM
|
5 |
|
6 |
-
from
|
|
|
7 |
import torch
|
8 |
|
9 |
import click
|
@@ -16,12 +17,6 @@ def remap_state_dict(state_dict, config: PretrainedConfig):
|
|
16 |
Map the state_dict of a Huggingface BERT model to be flash_attn compatible.
|
17 |
"""
|
18 |
|
19 |
-
# Replace Roberta with Bert
|
20 |
-
def key_mapping_roberta(key):
|
21 |
-
return re.sub(r"^roberta.", "bert.", key)
|
22 |
-
|
23 |
-
state_dict = OrderedDict((key_mapping_roberta(k), v) for k, v in state_dict.items())
|
24 |
-
|
25 |
# LayerNorm
|
26 |
def key_mapping_ln_gamma_beta(key):
|
27 |
key = re.sub(r"LayerNorm.gamma$", "LayerNorm.weight", key)
|
@@ -34,21 +29,21 @@ def remap_state_dict(state_dict, config: PretrainedConfig):
|
|
34 |
|
35 |
# Layers
|
36 |
def key_mapping_layers(key):
|
37 |
-
return re.sub(r"^
|
38 |
|
39 |
state_dict = OrderedDict((key_mapping_layers(k), v) for k, v in state_dict.items())
|
40 |
|
41 |
# LayerNorm
|
42 |
def key_mapping_ln(key):
|
43 |
-
key = re.sub(r"^
|
44 |
key = re.sub(
|
45 |
-
r"^
|
46 |
-
r"
|
47 |
key,
|
48 |
)
|
49 |
key = re.sub(
|
50 |
-
r"^
|
51 |
-
r"
|
52 |
key,
|
53 |
)
|
54 |
key = re.sub(
|
@@ -63,13 +58,13 @@ def remap_state_dict(state_dict, config: PretrainedConfig):
|
|
63 |
# MLP
|
64 |
def key_mapping_mlp(key):
|
65 |
key = re.sub(
|
66 |
-
r"^
|
67 |
-
r"
|
68 |
key,
|
69 |
)
|
70 |
key = re.sub(
|
71 |
-
r"^
|
72 |
-
r"
|
73 |
key,
|
74 |
)
|
75 |
return key
|
@@ -79,33 +74,33 @@ def remap_state_dict(state_dict, config: PretrainedConfig):
|
|
79 |
# Attention
|
80 |
last_layer_subset = getattr(config, "last_layer_subset", False)
|
81 |
for d in range(config.num_hidden_layers):
|
82 |
-
Wq = state_dict.pop(f"
|
83 |
-
Wk = state_dict.pop(f"
|
84 |
-
Wv = state_dict.pop(f"
|
85 |
-
bq = state_dict.pop(f"
|
86 |
-
bk = state_dict.pop(f"
|
87 |
-
bv = state_dict.pop(f"
|
88 |
if not (last_layer_subset and d == config.num_hidden_layers - 1):
|
89 |
-
state_dict[f"
|
90 |
[Wq, Wk, Wv], dim=0
|
91 |
)
|
92 |
-
state_dict[f"
|
93 |
[bq, bk, bv], dim=0
|
94 |
)
|
95 |
else:
|
96 |
-
state_dict[f"
|
97 |
-
state_dict[f"
|
98 |
[Wk, Wv], dim=0
|
99 |
)
|
100 |
-
state_dict[f"
|
101 |
-
state_dict[f"
|
102 |
[bk, bv], dim=0
|
103 |
)
|
104 |
|
105 |
def key_mapping_attn(key):
|
106 |
return re.sub(
|
107 |
-
r"^
|
108 |
-
r"
|
109 |
key,
|
110 |
)
|
111 |
|
@@ -121,8 +116,8 @@ def remap_state_dict(state_dict, config: PretrainedConfig):
|
|
121 |
# Word embedding
|
122 |
pad_vocab_size_multiple = getattr(config, "pad_vocab_size_multiple", 1)
|
123 |
if pad_vocab_size_multiple > 1:
|
124 |
-
word_embeddings = state_dict["
|
125 |
-
state_dict["
|
126 |
word_embeddings, (0, 0, 0, config.vocab_size - word_embeddings.shape[0])
|
127 |
)
|
128 |
decoder_weight = state_dict["cls.predictions.decoder.weight"]
|
@@ -137,16 +132,6 @@ def remap_state_dict(state_dict, config: PretrainedConfig):
|
|
137 |
decoder_bias, (0, config.vocab_size - decoder_bias.shape[0]), value=-100.0
|
138 |
)
|
139 |
|
140 |
-
# Embeddings
|
141 |
-
def key_remove_bert(key):
|
142 |
-
return re.sub(r"^bert.", "", key)
|
143 |
-
|
144 |
-
state_dict = OrderedDict(
|
145 |
-
(key_remove_bert(k), v)
|
146 |
-
for k, v in state_dict.items()
|
147 |
-
if not k.startswith('lm_head')
|
148 |
-
)
|
149 |
-
|
150 |
return state_dict
|
151 |
|
152 |
|
|
|
1 |
import re
|
2 |
from collections import OrderedDict
|
3 |
+
from transformers import PretrainedConfig
|
4 |
from transformers import XLMRobertaForMaskedLM
|
5 |
|
6 |
+
from .configuration_xlm_roberta import XLMRobertaFlashConfig as BertConfig
|
7 |
+
from .modeling_xlm_roberta import XLMRobertaForMaskedLM as BertModel
|
8 |
import torch
|
9 |
|
10 |
import click
|
|
|
17 |
Map the state_dict of a Huggingface BERT model to be flash_attn compatible.
|
18 |
"""
|
19 |
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
# LayerNorm
|
21 |
def key_mapping_ln_gamma_beta(key):
|
22 |
key = re.sub(r"LayerNorm.gamma$", "LayerNorm.weight", key)
|
|
|
29 |
|
30 |
# Layers
|
31 |
def key_mapping_layers(key):
|
32 |
+
return re.sub(r"^roberta.encoder.layer.", "roberta.encoder.layers.", key)
|
33 |
|
34 |
state_dict = OrderedDict((key_mapping_layers(k), v) for k, v in state_dict.items())
|
35 |
|
36 |
# LayerNorm
|
37 |
def key_mapping_ln(key):
|
38 |
+
key = re.sub(r"^roberta.embeddings.LayerNorm.", "roberta.emb_ln.", key)
|
39 |
key = re.sub(
|
40 |
+
r"^roberta.encoder.layers.(\d+).attention.output.LayerNorm.(weight|bias)",
|
41 |
+
r"roberta.encoder.layers.\1.norm1.\2",
|
42 |
key,
|
43 |
)
|
44 |
key = re.sub(
|
45 |
+
r"^roberta.encoder.layers.(\d+).output.LayerNorm.(weight|bias)",
|
46 |
+
r"roberta.encoder.layers.\1.norm2.\2",
|
47 |
key,
|
48 |
)
|
49 |
key = re.sub(
|
|
|
58 |
# MLP
|
59 |
def key_mapping_mlp(key):
|
60 |
key = re.sub(
|
61 |
+
r"^roberta.encoder.layers.(\d+).intermediate.dense.(weight|bias)",
|
62 |
+
r"roberta.encoder.layers.\1.mlp.fc1.\2",
|
63 |
key,
|
64 |
)
|
65 |
key = re.sub(
|
66 |
+
r"^roberta.encoder.layers.(\d+).output.dense.(weight|bias)",
|
67 |
+
r"roberta.encoder.layers.\1.mlp.fc2.\2",
|
68 |
key,
|
69 |
)
|
70 |
return key
|
|
|
74 |
# Attention
|
75 |
last_layer_subset = getattr(config, "last_layer_subset", False)
|
76 |
for d in range(config.num_hidden_layers):
|
77 |
+
Wq = state_dict.pop(f"roberta.encoder.layers.{d}.attention.self.query.weight")
|
78 |
+
Wk = state_dict.pop(f"roberta.encoder.layers.{d}.attention.self.key.weight")
|
79 |
+
Wv = state_dict.pop(f"roberta.encoder.layers.{d}.attention.self.value.weight")
|
80 |
+
bq = state_dict.pop(f"roberta.encoder.layers.{d}.attention.self.query.bias")
|
81 |
+
bk = state_dict.pop(f"roberta.encoder.layers.{d}.attention.self.key.bias")
|
82 |
+
bv = state_dict.pop(f"roberta.encoder.layers.{d}.attention.self.value.bias")
|
83 |
if not (last_layer_subset and d == config.num_hidden_layers - 1):
|
84 |
+
state_dict[f"roberta.encoder.layers.{d}.mixer.Wqkv.weight"] = torch.cat(
|
85 |
[Wq, Wk, Wv], dim=0
|
86 |
)
|
87 |
+
state_dict[f"roberta.encoder.layers.{d}.mixer.Wqkv.bias"] = torch.cat(
|
88 |
[bq, bk, bv], dim=0
|
89 |
)
|
90 |
else:
|
91 |
+
state_dict[f"roberta.encoder.layers.{d}.mixer.Wq.weight"] = Wq
|
92 |
+
state_dict[f"roberta.encoder.layers.{d}.mixer.Wkv.weight"] = torch.cat(
|
93 |
[Wk, Wv], dim=0
|
94 |
)
|
95 |
+
state_dict[f"roberta.encoder.layers.{d}.mixer.Wq.bias"] = bq
|
96 |
+
state_dict[f"roberta.encoder.layers.{d}.mixer.Wkv.bias"] = torch.cat(
|
97 |
[bk, bv], dim=0
|
98 |
)
|
99 |
|
100 |
def key_mapping_attn(key):
|
101 |
return re.sub(
|
102 |
+
r"^roberta.encoder.layers.(\d+).attention.output.dense.(weight|bias)",
|
103 |
+
r"roberta.encoder.layers.\1.mixer.out_proj.\2",
|
104 |
key,
|
105 |
)
|
106 |
|
|
|
116 |
# Word embedding
|
117 |
pad_vocab_size_multiple = getattr(config, "pad_vocab_size_multiple", 1)
|
118 |
if pad_vocab_size_multiple > 1:
|
119 |
+
word_embeddings = state_dict["roberta.embeddings.word_embeddings.weight"]
|
120 |
+
state_dict["roberta.embeddings.word_embeddings.weight"] = F.pad(
|
121 |
word_embeddings, (0, 0, 0, config.vocab_size - word_embeddings.shape[0])
|
122 |
)
|
123 |
decoder_weight = state_dict["cls.predictions.decoder.weight"]
|
|
|
132 |
decoder_bias, (0, config.vocab_size - decoder_bias.shape[0]), value=-100.0
|
133 |
)
|
134 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
135 |
return state_dict
|
136 |
|
137 |
|
embedding.py
CHANGED
@@ -11,7 +11,7 @@ from torch import Tensor
|
|
11 |
from transformers.models.xlm_roberta.modeling_xlm_roberta import create_position_ids_from_input_ids
|
12 |
|
13 |
|
14 |
-
class
|
15 |
def __init__(
|
16 |
self,
|
17 |
embed_dim,
|
|
|
11 |
from transformers.models.xlm_roberta.modeling_xlm_roberta import create_position_ids_from_input_ids
|
12 |
|
13 |
|
14 |
+
class XLMRobertaEmbeddings(nn.Module):
|
15 |
def __init__(
|
16 |
self,
|
17 |
embed_dim,
|
modeling_bert.py β modeling_xlm_roberta.py
RENAMED
@@ -13,28 +13,32 @@ import re
|
|
13 |
from collections import OrderedDict
|
14 |
from collections.abc import Sequence
|
15 |
from functools import partial
|
16 |
-
from typing import Any, Mapping
|
17 |
|
18 |
import torch
|
19 |
import torch.nn as nn
|
20 |
import torch.nn.functional as F
|
21 |
from einops import rearrange
|
22 |
-
from transformers import
|
23 |
from transformers.modeling_utils import PreTrainedModel
|
|
|
|
|
|
|
24 |
from transformers.models.bert.modeling_bert import (
|
25 |
BaseModelOutputWithPoolingAndCrossAttentions,
|
26 |
BertForPreTrainingOutput,
|
27 |
)
|
28 |
|
29 |
-
from
|
|
|
|
|
30 |
index_first_axis,
|
31 |
index_first_axis_residual,
|
32 |
pad_input,
|
33 |
unpad_input,
|
34 |
)
|
35 |
-
from .
|
36 |
from .block import Block
|
37 |
-
from .embedding import
|
38 |
from .mha import MHA
|
39 |
from .mlp import FusedMLP, Mlp
|
40 |
|
@@ -155,8 +159,8 @@ def _init_weights(module, initializer_range=0.02):
|
|
155 |
nn.init.zeros_(module.weight[module.padding_idx])
|
156 |
|
157 |
|
158 |
-
class
|
159 |
-
def __init__(self, config:
|
160 |
super().__init__()
|
161 |
self.use_flash_attn = getattr(config, "use_flash_attn", False)
|
162 |
self.layers = nn.ModuleList(
|
@@ -218,7 +222,7 @@ class BertEncoder(nn.Module):
|
|
218 |
return hidden_states
|
219 |
|
220 |
|
221 |
-
class
|
222 |
def __init__(self, config):
|
223 |
super().__init__()
|
224 |
fused_bias_fc = getattr(config, "fused_bias_fc", False)
|
@@ -237,7 +241,7 @@ class BertPooler(nn.Module):
|
|
237 |
return pooled_output
|
238 |
|
239 |
|
240 |
-
class
|
241 |
def __init__(self, config):
|
242 |
super().__init__()
|
243 |
fused_bias_fc = getattr(config, "fused_bias_fc", False)
|
@@ -268,7 +272,7 @@ class BertPredictionHeadTransform(nn.Module):
|
|
268 |
return hidden_states
|
269 |
|
270 |
|
271 |
-
class
|
272 |
def __init__(self, config):
|
273 |
super().__init__()
|
274 |
fused_bias_fc = getattr(config, "fused_bias_fc", False)
|
@@ -276,7 +280,7 @@ class BertLMPredictionHead(nn.Module):
|
|
276 |
raise ImportError("fused_dense is not installed")
|
277 |
linear_cls = nn.Linear if not fused_bias_fc else FusedDense
|
278 |
|
279 |
-
self.transform =
|
280 |
|
281 |
# The output weights are the same as the input embeddings, but there is
|
282 |
# an output-only bias for each token.
|
@@ -288,10 +292,10 @@ class BertLMPredictionHead(nn.Module):
|
|
288 |
return hidden_states
|
289 |
|
290 |
|
291 |
-
class
|
292 |
def __init__(self, config):
|
293 |
super().__init__()
|
294 |
-
self.predictions =
|
295 |
self.seq_relationship = nn.Linear(config.hidden_size, 2)
|
296 |
|
297 |
def forward(self, sequence_output, pooled_output):
|
@@ -300,64 +304,22 @@ class BertPreTrainingHeads(nn.Module):
|
|
300 |
return prediction_scores, seq_relationship_score
|
301 |
|
302 |
|
303 |
-
|
304 |
-
# """An abstract class to handle weights initialization and
|
305 |
-
# a simple interface for dowloading and loading pretrained models.
|
306 |
-
# """
|
307 |
-
#
|
308 |
-
# def __init__(self, config, *inputs, **kwargs):
|
309 |
-
# super().__init__()
|
310 |
-
# if not isinstance(config, BertConfig):
|
311 |
-
# raise ValueError(
|
312 |
-
# "Parameter config in `{}(config)` should be an instance of class `BertConfig`. "
|
313 |
-
# "To create a model from a Google pretrained model use "
|
314 |
-
# "`model = {}.from_pretrained(PRETRAINED_MODEL_NAME)`".format(
|
315 |
-
# self.__class__.__name__, self.__class__.__name__
|
316 |
-
# )
|
317 |
-
# )
|
318 |
-
# self.config = config
|
319 |
-
#
|
320 |
-
# @classmethod
|
321 |
-
# def from_pretrained(cls, model_name, config, *inputs, **kwargs):
|
322 |
-
# """
|
323 |
-
# Instantiate a BertPreTrainedModel from a pre-trained model file or a pytorch state dict.
|
324 |
-
# Download and cache the pre-trained model file if needed.
|
325 |
-
#
|
326 |
-
# Params:
|
327 |
-
# pretrained_model_name_or_path: either:
|
328 |
-
# - a path or url to a pretrained model archive containing:
|
329 |
-
# . `bert_config.json` a configuration file for the model
|
330 |
-
# . `pytorch_model.bin` a PyTorch dump of a BertForPretraining instance
|
331 |
-
# - a path or url to a pretrained model archive containing:
|
332 |
-
# . `bert_config.json` a configuration file for the model
|
333 |
-
# . `model.chkpt` a TensorFlow checkpoint
|
334 |
-
# *inputs, **kwargs: additional input for the specific Bert class
|
335 |
-
# (ex: num_labels for BertForSequenceClassification)
|
336 |
-
# """
|
337 |
-
# # Instantiate model.
|
338 |
-
# model = cls(config, *inputs, **kwargs)
|
339 |
-
# load_return = model.load_state_dict(
|
340 |
-
# remap_state_dict(state_dict_from_pretrained(model_name), config), strict=False
|
341 |
-
# )
|
342 |
-
# logger.info(load_return)
|
343 |
-
# return model
|
344 |
-
|
345 |
-
class BertPreTrainedModel(PreTrainedModel):
|
346 |
"""An abstract class to handle weights initialization and
|
347 |
a simple interface for dowloading and loading pretrained models.
|
348 |
"""
|
349 |
-
config_class =
|
350 |
-
base_model_prefix = "
|
351 |
supports_gradient_checkpointing = True
|
352 |
|
353 |
def _set_gradient_checkpointing(self, module, value=False):
|
354 |
-
if isinstance(module,
|
355 |
module.gradient_checkpointing = value
|
356 |
|
357 |
|
358 |
|
359 |
-
class
|
360 |
-
def __init__(self, config:
|
361 |
super().__init__(config)
|
362 |
self.pad_vocab_size_multiple = getattr(config, "pad_vocab_size_multiple", 1)
|
363 |
if config.vocab_size % self.pad_vocab_size_multiple != 0:
|
@@ -369,7 +331,7 @@ class BertModel(BertPreTrainedModel):
|
|
369 |
raise ImportError("Triton is not installed")
|
370 |
assert config.hidden_act in ["gelu", "gelu_new", "gelu_fast", "gelu_pytorch_tanh"]
|
371 |
|
372 |
-
self.embeddings =
|
373 |
config.hidden_size,
|
374 |
config.vocab_size,
|
375 |
config.max_position_embeddings,
|
@@ -378,11 +340,12 @@ class BertModel(BertPreTrainedModel):
|
|
378 |
)
|
379 |
self.emb_drop = nn.Dropout(config.hidden_dropout_prob)
|
380 |
self.emb_ln = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
|
381 |
-
self.encoder =
|
382 |
-
self.pooler =
|
383 |
|
384 |
self.apply(partial(_init_weights, initializer_range=config.initializer_range))
|
385 |
|
|
|
386 |
def forward(
|
387 |
self,
|
388 |
input_ids,
|
@@ -390,12 +353,22 @@ class BertModel(BertPreTrainedModel):
|
|
390 |
token_type_ids=None,
|
391 |
attention_mask=None,
|
392 |
masked_tokens_mask=None,
|
|
|
|
|
393 |
):
|
394 |
-
"""If masked_tokens_mask is not None (i.e. last_layer_subset == True in
|
395 |
we only want the output for the masked tokens. This means that we only compute the last
|
396 |
layer output for these tokens.
|
397 |
masked_tokens_mask: (batch, seqlen), dtype=torch.bool
|
398 |
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
399 |
hidden_states = self.embeddings(
|
400 |
input_ids, position_ids=position_ids, token_type_ids=token_type_ids
|
401 |
)
|
@@ -437,111 +410,200 @@ class BertModel(BertPreTrainedModel):
|
|
437 |
sequence_output = sequence_output[masked_tokens_mask[subset_mask]]
|
438 |
pooled_output = self.pooler(pool_input, pool=False) if self.pooler is not None else None
|
439 |
|
|
|
|
|
|
|
440 |
return BaseModelOutputWithPoolingAndCrossAttentions(
|
441 |
last_hidden_state=sequence_output,
|
442 |
pooler_output=pooled_output,
|
443 |
)
|
444 |
|
445 |
|
446 |
-
class
|
447 |
-
|
448 |
-
|
449 |
-
|
450 |
super().__init__(config)
|
451 |
-
# If dense_seq_output, we only need to pass the hidden states for the masked out tokens
|
452 |
-
# (around 15%) to the classifier heads.
|
453 |
-
self.dense_seq_output = getattr(config, "dense_seq_output", False)
|
454 |
-
# If last_layer_subset, we only need the compute the last layer for a subset of tokens
|
455 |
-
# (e.g., the tokens we need to compute the masked LM loss and the next-sentence prediction).
|
456 |
-
self.last_layer_subset = getattr(config, "last_layer_subset", False)
|
457 |
-
if self.last_layer_subset:
|
458 |
-
assert self.dense_seq_output, "last_layer_subset requires dense_seq_output"
|
459 |
-
use_xentropy = getattr(config, "use_xentropy", False)
|
460 |
-
if use_xentropy and CrossEntropyLoss is None:
|
461 |
-
raise ImportError("xentropy_cuda is not installed")
|
462 |
-
loss_cls = (
|
463 |
-
nn.CrossEntropyLoss
|
464 |
-
if not use_xentropy
|
465 |
-
else partial(CrossEntropyLoss, inplace_backward=True)
|
466 |
-
)
|
467 |
|
468 |
-
|
469 |
-
|
470 |
-
|
471 |
-
|
|
|
|
|
|
|
|
|
472 |
|
473 |
# Initialize weights and apply final processing
|
474 |
-
self.
|
475 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
476 |
|
477 |
-
def tie_weights(self):
|
478 |
-
self.cls.predictions.decoder.weight = self.bert.embeddings.word_embeddings.weight
|
479 |
|
480 |
def forward(
|
481 |
self,
|
482 |
-
input_ids,
|
483 |
-
|
484 |
-
token_type_ids=None,
|
485 |
-
|
486 |
-
|
487 |
-
|
488 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
489 |
"""
|
490 |
-
|
491 |
-
mask).
|
492 |
-
Outputs:
|
493 |
-
if `labels` and `next_sentence_label` are not `None`:
|
494 |
-
Outputs the total_loss which is the sum of the masked language modeling loss and the next
|
495 |
-
sentence classification loss.
|
496 |
-
if `labels` or `next_sentence_label` is `None`:
|
497 |
-
Outputs a tuple comprising
|
498 |
-
- the masked language modeling logits of shape [batch_size, sequence_length, vocab_size], and
|
499 |
-
- the next sentence classification logits of shape [batch_size, 2].
|
500 |
|
501 |
-
|
502 |
-
masked_tokens_mask = labels > 0 if (self.last_layer_subset and labels is not None) else None
|
503 |
-
outputs = self.bert(
|
504 |
input_ids,
|
505 |
-
|
506 |
token_type_ids=token_type_ids,
|
507 |
-
|
508 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
509 |
)
|
510 |
-
sequence_output
|
511 |
-
|
512 |
-
|
513 |
-
|
514 |
-
|
515 |
-
|
516 |
-
|
517 |
-
|
518 |
-
|
519 |
-
|
520 |
-
if
|
521 |
-
|
522 |
-
|
523 |
-
|
524 |
-
|
525 |
-
|
526 |
-
|
527 |
-
|
528 |
-
|
529 |
-
rearrange(prediction_scores, "... v -> (...) v"),
|
530 |
-
rearrange(labels, "... -> (...)"),
|
531 |
-
)
|
532 |
-
next_sentence_loss = self.nsp_loss(
|
533 |
-
rearrange(seq_relationship_score, "... t -> (...) t"),
|
534 |
-
rearrange(next_sentence_label, "... -> (...)"),
|
535 |
-
)
|
536 |
-
total_loss = masked_lm_loss.float() + next_sentence_loss.float()
|
537 |
-
|
538 |
-
return BertForPreTrainingOutput(
|
539 |
-
loss=total_loss,
|
540 |
-
prediction_logits=prediction_scores,
|
541 |
-
seq_relationship_logits=seq_relationship_score,
|
542 |
)
|
543 |
|
544 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
545 |
def remap_state_dict(state_dict, config: PretrainedConfig):
|
546 |
"""
|
547 |
Map the state_dict of a Huggingface BERT model to be flash_attn compatible.
|
|
|
13 |
from collections import OrderedDict
|
14 |
from collections.abc import Sequence
|
15 |
from functools import partial
|
|
|
16 |
|
17 |
import torch
|
18 |
import torch.nn as nn
|
19 |
import torch.nn.functional as F
|
20 |
from einops import rearrange
|
21 |
+
from transformers import PretrainedConfig
|
22 |
from transformers.modeling_utils import PreTrainedModel
|
23 |
+
from transformers.modeling_outputs import MaskedLMOutput
|
24 |
+
from transformers.models.xlm_roberta.modeling_xlm_roberta import XLMRobertaLMHead
|
25 |
+
|
26 |
from transformers.models.bert.modeling_bert import (
|
27 |
BaseModelOutputWithPoolingAndCrossAttentions,
|
28 |
BertForPreTrainingOutput,
|
29 |
)
|
30 |
|
31 |
+
from typing import Optional, Tuple, Union
|
32 |
+
|
33 |
+
from .xlm_padding import (
|
34 |
index_first_axis,
|
35 |
index_first_axis_residual,
|
36 |
pad_input,
|
37 |
unpad_input,
|
38 |
)
|
39 |
+
from .configuration_xlm_roberta import XLMRobertaFlashConfig
|
40 |
from .block import Block
|
41 |
+
from .embedding import XLMRobertaEmbeddings
|
42 |
from .mha import MHA
|
43 |
from .mlp import FusedMLP, Mlp
|
44 |
|
|
|
159 |
nn.init.zeros_(module.weight[module.padding_idx])
|
160 |
|
161 |
|
162 |
+
class XLMRobertaEncoder(nn.Module):
|
163 |
+
def __init__(self, config: XLMRobertaFlashConfig):
|
164 |
super().__init__()
|
165 |
self.use_flash_attn = getattr(config, "use_flash_attn", False)
|
166 |
self.layers = nn.ModuleList(
|
|
|
222 |
return hidden_states
|
223 |
|
224 |
|
225 |
+
class XLMRobertaPooler(nn.Module):
|
226 |
def __init__(self, config):
|
227 |
super().__init__()
|
228 |
fused_bias_fc = getattr(config, "fused_bias_fc", False)
|
|
|
241 |
return pooled_output
|
242 |
|
243 |
|
244 |
+
class XLMRobertaPredictionHeadTransform(nn.Module):
|
245 |
def __init__(self, config):
|
246 |
super().__init__()
|
247 |
fused_bias_fc = getattr(config, "fused_bias_fc", False)
|
|
|
272 |
return hidden_states
|
273 |
|
274 |
|
275 |
+
class XLMRobertaLMPredictionHead(nn.Module):
|
276 |
def __init__(self, config):
|
277 |
super().__init__()
|
278 |
fused_bias_fc = getattr(config, "fused_bias_fc", False)
|
|
|
280 |
raise ImportError("fused_dense is not installed")
|
281 |
linear_cls = nn.Linear if not fused_bias_fc else FusedDense
|
282 |
|
283 |
+
self.transform = XLMRobertaPredictionHeadTransform(config)
|
284 |
|
285 |
# The output weights are the same as the input embeddings, but there is
|
286 |
# an output-only bias for each token.
|
|
|
292 |
return hidden_states
|
293 |
|
294 |
|
295 |
+
class XLMRobertaPreTrainingHeads(nn.Module):
|
296 |
def __init__(self, config):
|
297 |
super().__init__()
|
298 |
+
self.predictions = XLMRobertaLMPredictionHead(config)
|
299 |
self.seq_relationship = nn.Linear(config.hidden_size, 2)
|
300 |
|
301 |
def forward(self, sequence_output, pooled_output):
|
|
|
304 |
return prediction_scores, seq_relationship_score
|
305 |
|
306 |
|
307 |
+
class XLMRobertaPreTrainedModel(PreTrainedModel):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
308 |
"""An abstract class to handle weights initialization and
|
309 |
a simple interface for dowloading and loading pretrained models.
|
310 |
"""
|
311 |
+
config_class = XLMRobertaFlashConfig
|
312 |
+
base_model_prefix = "roberta"
|
313 |
supports_gradient_checkpointing = True
|
314 |
|
315 |
def _set_gradient_checkpointing(self, module, value=False):
|
316 |
+
if isinstance(module, XLMRobertaEncoder):
|
317 |
module.gradient_checkpointing = value
|
318 |
|
319 |
|
320 |
|
321 |
+
class XLMRobertaModel(XLMRobertaPreTrainedModel):
|
322 |
+
def __init__(self, config: XLMRobertaFlashConfig, add_pooling_layer=True):
|
323 |
super().__init__(config)
|
324 |
self.pad_vocab_size_multiple = getattr(config, "pad_vocab_size_multiple", 1)
|
325 |
if config.vocab_size % self.pad_vocab_size_multiple != 0:
|
|
|
331 |
raise ImportError("Triton is not installed")
|
332 |
assert config.hidden_act in ["gelu", "gelu_new", "gelu_fast", "gelu_pytorch_tanh"]
|
333 |
|
334 |
+
self.embeddings = XLMRobertaEmbeddings(
|
335 |
config.hidden_size,
|
336 |
config.vocab_size,
|
337 |
config.max_position_embeddings,
|
|
|
340 |
)
|
341 |
self.emb_drop = nn.Dropout(config.hidden_dropout_prob)
|
342 |
self.emb_ln = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
|
343 |
+
self.encoder = XLMRobertaEncoder(config)
|
344 |
+
self.pooler = XLMRobertaPooler(config) if add_pooling_layer else None
|
345 |
|
346 |
self.apply(partial(_init_weights, initializer_range=config.initializer_range))
|
347 |
|
348 |
+
|
349 |
def forward(
|
350 |
self,
|
351 |
input_ids,
|
|
|
353 |
token_type_ids=None,
|
354 |
attention_mask=None,
|
355 |
masked_tokens_mask=None,
|
356 |
+
return_dict=None,
|
357 |
+
**kwargs,
|
358 |
):
|
359 |
+
"""If masked_tokens_mask is not None (i.e. last_layer_subset == True in XLMForPreTraining),
|
360 |
we only want the output for the masked tokens. This means that we only compute the last
|
361 |
layer output for these tokens.
|
362 |
masked_tokens_mask: (batch, seqlen), dtype=torch.bool
|
363 |
"""
|
364 |
+
|
365 |
+
if kwargs:
|
366 |
+
for key, value in kwargs.items():
|
367 |
+
if value is not None:
|
368 |
+
logger.warning('Flash attention implementation does not support kwargs: %s', key)
|
369 |
+
|
370 |
+
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
371 |
+
|
372 |
hidden_states = self.embeddings(
|
373 |
input_ids, position_ids=position_ids, token_type_ids=token_type_ids
|
374 |
)
|
|
|
410 |
sequence_output = sequence_output[masked_tokens_mask[subset_mask]]
|
411 |
pooled_output = self.pooler(pool_input, pool=False) if self.pooler is not None else None
|
412 |
|
413 |
+
if not return_dict:
|
414 |
+
return sequence_output, pooled_output
|
415 |
+
|
416 |
return BaseModelOutputWithPoolingAndCrossAttentions(
|
417 |
last_hidden_state=sequence_output,
|
418 |
pooler_output=pooled_output,
|
419 |
)
|
420 |
|
421 |
|
422 |
+
class XLMRobertaForMaskedLM(XLMRobertaPreTrainedModel):
|
423 |
+
_tied_weights_keys = ["lm_head.decoder.weight", "lm_head.decoder.bias"]
|
424 |
+
|
425 |
+
def __init__(self, config):
|
426 |
super().__init__(config)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
427 |
|
428 |
+
if config.is_decoder:
|
429 |
+
logger.warning(
|
430 |
+
"If you want to use `XLMRobertaForMaskedLM` make sure `config.is_decoder=False` for "
|
431 |
+
"bi-directional self-attention."
|
432 |
+
)
|
433 |
+
|
434 |
+
self.roberta = XLMRobertaModel(config, add_pooling_layer=False)
|
435 |
+
self.lm_head = XLMRobertaLMHead(config)
|
436 |
|
437 |
# Initialize weights and apply final processing
|
438 |
+
self.post_init()
|
439 |
+
|
440 |
+
def get_input_embeddings(self):
|
441 |
+
return self.roberta.embeddings.word_embeddings
|
442 |
+
|
443 |
+
def get_output_embeddings(self):
|
444 |
+
return self.lm_head.decoder
|
445 |
+
|
446 |
+
def set_output_embeddings(self, new_embeddings):
|
447 |
+
self.lm_head.decoder = new_embeddings
|
448 |
|
|
|
|
|
449 |
|
450 |
def forward(
|
451 |
self,
|
452 |
+
input_ids: Optional[torch.LongTensor] = None,
|
453 |
+
attention_mask: Optional[torch.FloatTensor] = None,
|
454 |
+
token_type_ids: Optional[torch.LongTensor] = None,
|
455 |
+
position_ids: Optional[torch.LongTensor] = None,
|
456 |
+
head_mask: Optional[torch.FloatTensor] = None,
|
457 |
+
inputs_embeds: Optional[torch.FloatTensor] = None,
|
458 |
+
encoder_hidden_states: Optional[torch.FloatTensor] = None,
|
459 |
+
encoder_attention_mask: Optional[torch.FloatTensor] = None,
|
460 |
+
labels: Optional[torch.LongTensor] = None,
|
461 |
+
output_attentions: Optional[bool] = None,
|
462 |
+
output_hidden_states: Optional[bool] = None,
|
463 |
+
return_dict: Optional[bool] = None,
|
464 |
+
) -> Union[Tuple[torch.Tensor], MaskedLMOutput]:
|
465 |
+
r"""
|
466 |
+
labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
|
467 |
+
Labels for computing the masked language modeling loss. Indices should be in `[-100, 0, ...,
|
468 |
+
config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored (masked), the
|
469 |
+
loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`
|
470 |
+
kwargs (`Dict[str, any]`, optional, defaults to *{}*):
|
471 |
+
Used to hide legacy arguments that have been deprecated.
|
472 |
"""
|
473 |
+
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
474 |
|
475 |
+
outputs = self.roberta(
|
|
|
|
|
476 |
input_ids,
|
477 |
+
attention_mask=attention_mask,
|
478 |
token_type_ids=token_type_ids,
|
479 |
+
position_ids=position_ids,
|
480 |
+
head_mask=head_mask,
|
481 |
+
inputs_embeds=inputs_embeds,
|
482 |
+
encoder_hidden_states=encoder_hidden_states,
|
483 |
+
encoder_attention_mask=encoder_attention_mask,
|
484 |
+
output_attentions=output_attentions,
|
485 |
+
output_hidden_states=output_hidden_states,
|
486 |
+
return_dict=return_dict,
|
487 |
)
|
488 |
+
sequence_output = outputs[0]
|
489 |
+
prediction_scores = self.lm_head(sequence_output)
|
490 |
+
|
491 |
+
masked_lm_loss = None
|
492 |
+
if labels is not None:
|
493 |
+
# move labels to correct device to enable model parallelism
|
494 |
+
labels = labels.to(prediction_scores.device)
|
495 |
+
loss_fct = CrossEntropyLoss()
|
496 |
+
masked_lm_loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), labels.view(-1))
|
497 |
+
|
498 |
+
if not return_dict:
|
499 |
+
output = (prediction_scores,) + outputs[2:]
|
500 |
+
return ((masked_lm_loss,) + output) if masked_lm_loss is not None else output
|
501 |
+
|
502 |
+
return MaskedLMOutput(
|
503 |
+
loss=masked_lm_loss,
|
504 |
+
logits=prediction_scores,
|
505 |
+
hidden_states=outputs.hidden_states,
|
506 |
+
attentions=outputs.attentions,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
507 |
)
|
508 |
|
509 |
|
510 |
+
# class XLMRobertaForPreTraining(XLMRobertaPreTrainedModel):
|
511 |
+
# def __init__(self, config: XLMRobertaFlashConfig):
|
512 |
+
# super().__init__(config)
|
513 |
+
# # If dense_seq_output, we only need to pass the hidden states for the masked out tokens
|
514 |
+
# # (around 15%) to the classifier heads.
|
515 |
+
# self.dense_seq_output = getattr(config, "dense_seq_output", False)
|
516 |
+
# # If last_layer_subset, we only need the compute the last layer for a subset of tokens
|
517 |
+
# # (e.g., the tokens we need to compute the masked LM loss and the next-sentence prediction).
|
518 |
+
# self.last_layer_subset = getattr(config, "last_layer_subset", False)
|
519 |
+
# if self.last_layer_subset:
|
520 |
+
# assert self.dense_seq_output, "last_layer_subset requires dense_seq_output"
|
521 |
+
# use_xentropy = getattr(config, "use_xentropy", False)
|
522 |
+
# if use_xentropy and CrossEntropyLoss is None:
|
523 |
+
# raise ImportError("xentropy_cuda is not installed")
|
524 |
+
# loss_cls = (
|
525 |
+
# nn.CrossEntropyLoss
|
526 |
+
# if not use_xentropy
|
527 |
+
# else partial(CrossEntropyLoss, inplace_backward=True)
|
528 |
+
# )
|
529 |
+
#
|
530 |
+
# self.xlm = XLMRobertaModel(config)
|
531 |
+
# self.cls = XLMRobertaPreTrainingHeads(config)
|
532 |
+
# self.mlm_loss = loss_cls(ignore_index=0)
|
533 |
+
# self.nsp_loss = loss_cls(ignore_index=-1)
|
534 |
+
#
|
535 |
+
# # Initialize weights and apply final processing
|
536 |
+
# self.apply(partial(_init_weights, initializer_range=config.initializer_range))
|
537 |
+
# self.tie_weights()
|
538 |
+
#
|
539 |
+
# def tie_weights(self):
|
540 |
+
# self.cls.predictions.decoder.weight = self.xlm.embeddings.word_embeddings.weight
|
541 |
+
#
|
542 |
+
# def forward(
|
543 |
+
# self,
|
544 |
+
# input_ids,
|
545 |
+
# position_ids=None,
|
546 |
+
# token_type_ids=None,
|
547 |
+
# attention_mask=None,
|
548 |
+
# labels=None,
|
549 |
+
# next_sentence_label=None,
|
550 |
+
# ):
|
551 |
+
# """
|
552 |
+
# If labels are provided, they must be 0 for masked out tokens (as specified in the attention
|
553 |
+
# mask).
|
554 |
+
# Outputs:
|
555 |
+
# if `labels` and `next_sentence_label` are not `None`:
|
556 |
+
# Outputs the total_loss which is the sum of the masked language modeling loss and the next
|
557 |
+
# sentence classification loss.
|
558 |
+
# if `labels` or `next_sentence_label` is `None`:
|
559 |
+
# Outputs a tuple comprising
|
560 |
+
# - the masked language modeling logits of shape [batch_size, sequence_length, vocab_size], and
|
561 |
+
# - the next sentence classification logits of shape [batch_size, 2].
|
562 |
+
#
|
563 |
+
# """
|
564 |
+
# masked_tokens_mask = labels > 0 if (self.last_layer_subset and labels is not None) else None
|
565 |
+
# outputs = self.xlm(
|
566 |
+
# input_ids,
|
567 |
+
# position_ids=position_ids,
|
568 |
+
# token_type_ids=token_type_ids,
|
569 |
+
# attention_mask=attention_mask.bool() if attention_mask is not None else None,
|
570 |
+
# masked_tokens_mask=masked_tokens_mask,
|
571 |
+
# )
|
572 |
+
# sequence_output, pooled_output = outputs.last_hidden_state, outputs.pooler_output
|
573 |
+
# if self.dense_seq_output and labels is not None:
|
574 |
+
# masked_token_idx = torch.nonzero(labels.flatten() > 0, as_tuple=False).flatten()
|
575 |
+
# if not self.last_layer_subset:
|
576 |
+
# sequence_output = index_first_axis(
|
577 |
+
# rearrange(sequence_output, "b s d -> (b s) d"), masked_token_idx
|
578 |
+
# )
|
579 |
+
# prediction_scores, seq_relationship_score = self.cls(sequence_output, pooled_output)
|
580 |
+
#
|
581 |
+
# total_loss = None
|
582 |
+
# if labels is not None and next_sentence_label is not None:
|
583 |
+
# if (
|
584 |
+
# self.dense_seq_output and labels is not None
|
585 |
+
# ): # prediction_scores are already flattened
|
586 |
+
# masked_lm_loss = self.mlm_loss(
|
587 |
+
# prediction_scores, labels.flatten()[masked_token_idx]
|
588 |
+
# )
|
589 |
+
# else:
|
590 |
+
# masked_lm_loss = self.mlm_loss(
|
591 |
+
# rearrange(prediction_scores, "... v -> (...) v"),
|
592 |
+
# rearrange(labels, "... -> (...)"),
|
593 |
+
# )
|
594 |
+
# next_sentence_loss = self.nsp_loss(
|
595 |
+
# rearrange(seq_relationship_score, "... t -> (...) t"),
|
596 |
+
# rearrange(next_sentence_label, "... -> (...)"),
|
597 |
+
# )
|
598 |
+
# total_loss = masked_lm_loss.float() + next_sentence_loss.float()
|
599 |
+
#
|
600 |
+
# return BertForPreTrainingOutput(
|
601 |
+
# loss=total_loss,
|
602 |
+
# prediction_logits=prediction_scores,
|
603 |
+
# seq_relationship_logits=seq_relationship_score,
|
604 |
+
# )
|
605 |
+
|
606 |
+
|
607 |
def remap_state_dict(state_dict, config: PretrainedConfig):
|
608 |
"""
|
609 |
Map the state_dict of a Huggingface BERT model to be flash_attn compatible.
|
pytorch_model.bin
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:cfa8fa7c7e120199548fe7149512c0adfe58f6bc13ce19f09b895aa25e8af910
|
3 |
+
size 1113232188
|
bert_padding.py β xlm_padding.py
RENAMED
File without changes
|