File size: 14,930 Bytes
f4460da | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 | """Transformers-compatible implementation of the pinyin-code causal LM."""
from __future__ import annotations
import torch
from torch import nn
from torch.nn import functional as F
from transformers import PreTrainedModel
from transformers.generation import GenerationMixin
from transformers.modeling_outputs import (
BaseModelOutput,
CausalLMOutput,
SequenceClassifierOutput,
)
from .configuration_pinyin_code import PinyinCodeConfig
class CausalSelfAttention(nn.Module):
"""Multi-head masked self-attention matching the original training module."""
def __init__(self, config: PinyinCodeConfig) -> None:
super().__init__()
if config.n_embd % config.n_head != 0:
raise ValueError("n_embd must be divisible by n_head")
self.n_head = config.n_head
self.head_dim = config.n_embd // config.n_head
self.dropout_p = config.dropout
self.qkv = nn.Linear(config.n_embd, 3 * config.n_embd)
self.proj = nn.Linear(config.n_embd, config.n_embd)
self.resid_dropout = nn.Dropout(config.dropout)
def forward(self, x: torch.Tensor, attention_mask: torch.Tensor | None = None) -> torch.Tensor:
batch_size, seq_len, embd = x.shape
q, k, v = self.qkv(x).split(embd, dim=2)
q = q.view(batch_size, seq_len, self.n_head, self.head_dim).transpose(1, 2)
k = k.view(batch_size, seq_len, self.n_head, self.head_dim).transpose(1, 2)
v = v.view(batch_size, seq_len, self.n_head, self.head_dim).transpose(1, 2)
dropout_p = self.dropout_p if self.training else 0.0
if attention_mask is not None:
causal_mask = torch.ones(
seq_len,
seq_len,
device=x.device,
dtype=torch.bool,
).tril()
key_mask = attention_mask[:, None, None, :seq_len].to(dtype=torch.bool)
attn_mask = causal_mask.view(1, 1, seq_len, seq_len) & key_mask
y = F.scaled_dot_product_attention(
q,
k,
v,
attn_mask=attn_mask,
dropout_p=dropout_p,
is_causal=False,
)
else:
y = F.scaled_dot_product_attention(
q,
k,
v,
dropout_p=dropout_p,
is_causal=True,
)
y = y.transpose(1, 2).contiguous().view(batch_size, seq_len, embd)
return self.resid_dropout(self.proj(y))
class FeedForward(nn.Module):
"""Transformer MLP block."""
def __init__(self, config: PinyinCodeConfig) -> None:
super().__init__()
self.net = nn.Sequential(
nn.Linear(config.n_embd, 4 * config.n_embd),
nn.GELU(),
nn.Linear(4 * config.n_embd, config.n_embd),
nn.Dropout(config.dropout),
)
def forward(self, x: torch.Tensor) -> torch.Tensor:
return self.net(x)
class TransformerBlock(nn.Module):
"""Pre-norm Transformer block."""
def __init__(self, config: PinyinCodeConfig) -> None:
super().__init__()
self.ln_1 = nn.LayerNorm(config.n_embd)
self.attn = CausalSelfAttention(config)
self.ln_2 = nn.LayerNorm(config.n_embd)
self.mlp = FeedForward(config)
def forward(self, x: torch.Tensor, attention_mask: torch.Tensor | None = None) -> torch.Tensor:
x = x + self.attn(self.ln_1(x), attention_mask=attention_mask)
x = x + self.mlp(self.ln_2(x))
return x
class PinyinCodePreTrainedModel(PreTrainedModel):
"""Base class for pinyin-code Transformers models."""
config_class = PinyinCodeConfig
base_model_prefix = "pinyin_code"
supports_gradient_checkpointing = False
def _init_weights(self, module: nn.Module) -> None:
if isinstance(module, nn.Linear):
nn.init.normal_(module.weight, mean=0.0, std=0.02)
if module.bias is not None:
nn.init.zeros_(module.bias)
elif isinstance(module, nn.Embedding):
nn.init.normal_(module.weight, mean=0.0, std=0.02)
class PinyinCodeModel(PinyinCodePreTrainedModel):
"""Base decoder model returned by ``AutoModel``."""
def __init__(self, config: PinyinCodeConfig, init_weights: bool = True) -> None:
super().__init__(config)
self.token_embedding = nn.Embedding(config.vocab_size, config.n_embd)
self.position_embedding = nn.Embedding(config.block_size, config.n_embd)
self.dropout = nn.Dropout(config.dropout)
self.blocks = nn.ModuleList(TransformerBlock(config) for _ in range(config.n_layer))
self.ln_f = nn.LayerNorm(config.n_embd)
if init_weights:
self.post_init()
def get_input_embeddings(self) -> nn.Embedding:
return self.token_embedding
def set_input_embeddings(self, value: nn.Embedding) -> None:
self.token_embedding = value
def forward(
self,
input_ids: torch.Tensor | None = None,
attention_mask: torch.Tensor | None = None,
inputs_embeds: torch.Tensor | None = None,
position_ids: torch.Tensor | None = None,
output_hidden_states: bool | None = None,
return_dict: bool | None = None,
**kwargs,
) -> BaseModelOutput | tuple:
return_dict = True if return_dict is None else return_dict
output_hidden_states = (
self.config.output_hidden_states
if output_hidden_states is None
else output_hidden_states
)
if input_ids is None and inputs_embeds is None:
raise ValueError("You must provide either input_ids or inputs_embeds")
if input_ids is not None and inputs_embeds is not None:
raise ValueError("You cannot provide both input_ids and inputs_embeds")
if inputs_embeds is None:
_, seq_len = input_ids.shape
if seq_len > self.config.block_size:
raise ValueError(
f"Sequence length {seq_len} exceeds block size {self.config.block_size}"
)
inputs_embeds = self.token_embedding(input_ids)
else:
seq_len = inputs_embeds.shape[1]
if seq_len > self.config.block_size:
raise ValueError(
f"Sequence length {seq_len} exceeds block size {self.config.block_size}"
)
if position_ids is None:
if attention_mask is not None:
position_ids = attention_mask.long().cumsum(dim=-1) - 1
position_ids = position_ids.clamp_min(0)
else:
position_ids = torch.arange(seq_len, device=inputs_embeds.device)
position_ids = position_ids[:, -seq_len:] if position_ids.ndim == 2 else position_ids
x = inputs_embeds + self.position_embedding(position_ids)
x = self.dropout(x)
all_hidden_states = (x,) if output_hidden_states else None
for block in self.blocks:
x = block(x, attention_mask=attention_mask)
if output_hidden_states:
all_hidden_states = all_hidden_states + (x,)
hidden_states = self.ln_f(x)
if output_hidden_states:
all_hidden_states = all_hidden_states + (hidden_states,)
if not return_dict:
output = (hidden_states,)
if output_hidden_states:
output = output + (all_hidden_states,)
return output
return BaseModelOutput(
last_hidden_state=hidden_states,
hidden_states=all_hidden_states,
)
class PinyinCodeForCausalLM(PinyinCodeModel, GenerationMixin):
"""Compact GPT-style causal language model using the original architecture."""
_tied_weights_keys = {"lm_head.weight": "token_embedding.weight"}
_keys_to_ignore_on_load_missing = [r"lm_head\.weight"]
def __init__(self, config: PinyinCodeConfig) -> None:
super().__init__(config, init_weights=False)
self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
self.post_init()
self.tie_weights()
def get_output_embeddings(self) -> nn.Linear:
return self.lm_head
def set_output_embeddings(self, new_embeddings: nn.Linear) -> None:
self.lm_head = new_embeddings
def tie_weights(self, *args, **kwargs) -> None:
self.lm_head.weight = self.token_embedding.weight
def prepare_inputs_for_generation(
self,
input_ids: torch.Tensor,
past_key_values=None,
attention_mask: torch.Tensor | None = None,
**kwargs,
) -> dict:
if input_ids.shape[1] > self.config.block_size:
input_ids = input_ids[:, -self.config.block_size :]
if attention_mask is not None:
attention_mask = attention_mask[:, -self.config.block_size :]
position_ids = None
if attention_mask is not None:
position_ids = attention_mask.long().cumsum(dim=-1) - 1
position_ids = position_ids.clamp_min(0)
return {
"input_ids": input_ids,
"attention_mask": attention_mask,
"position_ids": position_ids,
}
def forward(
self,
input_ids: torch.Tensor | None = None,
attention_mask: torch.Tensor | None = None,
labels: torch.Tensor | None = None,
inputs_embeds: torch.Tensor | None = None,
position_ids: torch.Tensor | None = None,
output_hidden_states: bool | None = None,
return_dict: bool | None = None,
**kwargs,
) -> CausalLMOutput | tuple:
return_dict = True if return_dict is None else return_dict
decoder_outputs = PinyinCodeModel.forward(
self,
input_ids=input_ids,
attention_mask=attention_mask,
inputs_embeds=inputs_embeds,
position_ids=position_ids,
output_hidden_states=output_hidden_states,
return_dict=True,
)
logits = self.lm_head(decoder_outputs.last_hidden_state)
loss = None
if labels is not None:
loss = F.cross_entropy(
logits[:, :-1, :].contiguous().view(-1, logits.size(-1)),
labels[:, 1:].contiguous().view(-1),
ignore_index=-100,
)
if not return_dict:
output = (logits,)
if decoder_outputs.hidden_states is not None:
output = output + (decoder_outputs.hidden_states,)
return ((loss,) + output) if loss is not None else output
return CausalLMOutput(
loss=loss,
logits=logits,
hidden_states=decoder_outputs.hidden_states,
)
class PinyinCodeForSequenceClassification(PinyinCodeModel):
"""Sequence classifier using the pinyin-code decoder backbone."""
_keys_to_ignore_on_load_unexpected = [r"lm_head\.weight"]
def __init__(self, config: PinyinCodeConfig) -> None:
super().__init__(config, init_weights=False)
self.num_labels = config.num_labels
classifier_dropout = getattr(config, "classifier_dropout", None)
if classifier_dropout is None:
classifier_dropout = getattr(config, "hidden_dropout_prob", config.dropout)
self.score_dropout = nn.Dropout(classifier_dropout)
self.classifier = nn.Linear(config.hidden_size, config.num_labels)
self.post_init()
def forward(
self,
input_ids: torch.Tensor | None = None,
attention_mask: torch.Tensor | None = None,
labels: torch.Tensor | None = None,
token_type_ids: torch.Tensor | None = None,
position_ids: torch.Tensor | None = None,
inputs_embeds: torch.Tensor | None = None,
output_attentions: bool | None = None,
output_hidden_states: bool | None = None,
return_dict: bool | None = None,
**kwargs,
) -> SequenceClassifierOutput | tuple:
return_dict = True if return_dict is None else return_dict
outputs = PinyinCodeModel.forward(
self,
input_ids=input_ids,
attention_mask=attention_mask,
inputs_embeds=inputs_embeds,
position_ids=position_ids,
output_hidden_states=output_hidden_states,
return_dict=True,
**kwargs,
)
hidden_states = outputs.last_hidden_state
batch_size, sequence_length, _ = hidden_states.shape
if attention_mask is not None:
sequence_indices = attention_mask.to(hidden_states.device).long().sum(dim=-1) - 1
else:
sequence_indices = torch.full(
(batch_size,),
sequence_length - 1,
dtype=torch.long,
device=hidden_states.device,
)
sequence_indices = sequence_indices.clamp(min=0, max=sequence_length - 1)
batch_indices = torch.arange(batch_size, device=hidden_states.device)
pooled_hidden_states = hidden_states[batch_indices, sequence_indices]
logits = self.classifier(self.score_dropout(pooled_hidden_states))
loss = None
if labels is not None:
labels = labels.to(logits.device)
if getattr(self.config, "problem_type", None) is None:
if self.num_labels == 1:
self.config.problem_type = "regression"
elif labels.dtype in (torch.long, torch.int):
self.config.problem_type = "single_label_classification"
else:
self.config.problem_type = "multi_label_classification"
if self.config.problem_type == "regression":
if self.num_labels == 1:
loss = F.mse_loss(logits.squeeze(), labels.squeeze())
else:
loss = F.mse_loss(logits, labels)
elif self.config.problem_type == "single_label_classification":
loss = F.cross_entropy(
logits.view(-1, self.num_labels),
labels.view(-1),
)
elif self.config.problem_type == "multi_label_classification":
loss = F.binary_cross_entropy_with_logits(logits, labels)
if not return_dict:
output = (logits,)
if outputs.hidden_states is not None:
output = output + (outputs.hidden_states,)
attentions = getattr(outputs, "attentions", None)
if attentions is not None:
output = output + (attentions,)
return ((loss,) + output) if loss is not None else output
return SequenceClassifierOutput(
loss=loss,
logits=logits,
hidden_states=outputs.hidden_states,
attentions=getattr(outputs, "attentions", None),
)
|