File size: 12,610 Bytes
eafbe80 | 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 |
from __future__ import annotations
import warnings
from typing import TYPE_CHECKING, Optional
import torch
import torch.nn as nn
from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
from transformers.modeling_utils import PreTrainedModel
from transformers.utils import logging
from transformers.utils.deprecation import deprecate_kwarg
from fla.layers.deltaformer import DeltaFormerAttention
from fla.models.deltaformer.configuration_deltaformer import DeltaFormerConfig
from fla.models.utils import Cache, FLAGenerationMixin
from fla.modules import FusedCrossEntropyLoss, FusedLinearCrossEntropyLoss, RMSNorm
from fla.modules import GatedMLP as DeltaFormerMLP
from fla.modules.l2warp import l2_warp
try:
from transformers.modeling_layers import GradientCheckpointingLayer
except ImportError:
from fla.models.modeling_layers import GradientCheckpointingLayer
logger = logging.get_logger(__name__)
if TYPE_CHECKING:
from transformers.processing_utils import Unpack
class DeltaFormerBlock(GradientCheckpointingLayer):
def __init__(self, config: DeltaFormerConfig, layer_idx: int):
super().__init__()
self.config = config
self.layer_idx = layer_idx
self.attn_norm = (RMSNorm if config.fuse_norm else nn.RMSNorm)(config.hidden_size, eps=config.norm_eps)
self.attn = DeltaFormerAttention(
hidden_size=config.hidden_size,
num_heads=config.num_heads,
num_kv_heads=config.num_kv_heads,
qkv_bias=config.qkv_bias,
qk_norm=config.qk_norm,
rope_theta=config.rope_theta,
max_position_embeddings=config.rope_max_position_embeddings,
layer_idx=layer_idx,
)
self.mlp_norm = (RMSNorm if config.fuse_norm else nn.RMSNorm)(config.hidden_size, eps=config.norm_eps)
self.mlp = DeltaFormerMLP(
hidden_size=config.hidden_size,
hidden_ratio=config.hidden_ratio,
intermediate_size=config.intermediate_size,
hidden_act=config.hidden_act,
fuse_swiglu=config.fuse_swiglu,
)
def forward(
self,
hidden_states: torch.Tensor,
attention_mask: torch.Tensor | None = None,
past_key_values: Cache | list[torch.FloatTensor] | None = None,
use_cache: bool | None = False,
output_attentions: bool | None = False,
**kwargs: Unpack[dict],
) -> tuple[torch.FloatTensor, tuple[torch.FloatTensor, torch.FloatTensor] | None]:
residual = hidden_states
hidden_states = self.attn_norm(hidden_states)
hidden_states, _, past_key_values = self.attn(
hidden_states=hidden_states,
attention_mask=attention_mask,
past_key_values=past_key_values,
use_cache=use_cache,
output_attentions=output_attentions,
**kwargs,
)
if self.config.fuse_norm:
hidden_states, residual = self.mlp_norm(hidden_states, residual, True)
else:
hidden_states = residual + hidden_states
residual = hidden_states
hidden_states = self.mlp_norm(hidden_states)
hidden_states = self.mlp(hidden_states, **kwargs)
hidden_states = residual + hidden_states
outputs = (hidden_states, None, past_key_values)
return outputs
class DeltaFormerPreTrainedModel(PreTrainedModel):
config_class = DeltaFormerConfig
base_model_prefix = 'model'
supports_gradient_checkpointing = True
_no_split_modules = ['DeltaFormerBlock']
_supports_cache_class = True
def __init__(self, *inputs, **kwargs):
super().__init__(*inputs, **kwargs)
def _init_weights(
self,
module: nn.Module,
prenorm_residual_strategy: str | None = None,
num_residuals_per_layer: int = 2,
):
if isinstance(module, (nn.Linear, nn.Conv1d)):
nn.init.normal_(module.weight, mean=0.0, std=self.config.initializer_range)
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=self.config.initializer_range)
elif hasattr(module, 'reset_parameters'):
module.reset_parameters()
class DeltaFormerModel(DeltaFormerPreTrainedModel):
def __init__(self, config: DeltaFormerConfig):
super().__init__(config)
self.padding_idx = config.pad_token_id
self.vocab_size = config.vocab_size
self.embeddings = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
self.layers = nn.ModuleList([DeltaFormerBlock(config, layer_idx) for layer_idx in range(config.num_hidden_layers)])
self.norm = (RMSNorm if config.fuse_norm else nn.RMSNorm)(config.hidden_size, eps=config.norm_eps)
self.gradient_checkpointing = False
self.post_init()
def get_input_embeddings(self):
return self.embeddings
def set_input_embeddings(self, value):
self.embeddings = value
def forward(
self,
input_ids: torch.LongTensor | None = None,
attention_mask: Optional[torch.Tensor] = None, # noqa
inputs_embeds: torch.FloatTensor | None = None,
past_key_values: Cache | list[torch.FloatTensor] | None = None,
use_cache: bool | None = None,
output_attentions: bool | None = None,
output_hidden_states: bool | None = None,
return_dict: bool | None = None,
**kwargs: Unpack[dict],
) -> tuple | BaseModelOutputWithPast:
if output_attentions:
warnings.warn(
"`DeltaFormerModel` does not support output attention weights now, "
"so `output_attentions` is set to `False`.",
)
output_attentions = False
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
output_hidden_states = output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
use_cache = use_cache if use_cache is not None else (self.config.use_cache if not self.training else False)
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
if input_ids is not None and inputs_embeds is not None:
raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
if input_ids is None and inputs_embeds is None:
raise ValueError("You have to specify either input_ids or inputs_embeds")
if inputs_embeds is None:
inputs_embeds = self.embeddings(input_ids)
hidden_states = inputs_embeds
if use_cache and not isinstance(past_key_values, Cache):
past_key_values = Cache.from_legacy_cache(past_key_values)
all_hidden_states = () if output_hidden_states else None
all_attns = () if output_attentions else None
for layer in self.layers:
if output_hidden_states:
all_hidden_states += (hidden_states,)
hidden_states, attentions, past_key_values = layer(
hidden_states,
attention_mask=attention_mask,
past_key_values=past_key_values,
use_cache=use_cache,
output_attentions=output_attentions,
**kwargs,
)
if output_attentions:
all_attns += (attentions,)
hidden_states = self.norm(hidden_states)
if output_hidden_states:
all_hidden_states += (hidden_states,)
if not return_dict:
return tuple(i for i in [hidden_states, past_key_values, all_hidden_states, all_attns] if i is not None)
return BaseModelOutputWithPast(
last_hidden_state=hidden_states,
past_key_values=past_key_values,
hidden_states=all_hidden_states,
attentions=all_attns,
)
class DeltaFormerForCausalLM(DeltaFormerPreTrainedModel, FLAGenerationMixin):
_tied_weights_keys = ["lm_head.weight"]
def __init__(self, config: DeltaFormerConfig):
super().__init__(config)
self.model = DeltaFormerModel(config)
self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
self.criterion = None
self.post_init()
def get_input_embeddings(self):
return self.model.get_input_embeddings()
def set_input_embeddings(self, value):
self.model.set_input_embeddings(value)
def tie_weights(self):
self._tie_or_clone_weights(self.lm_head, self.get_input_embeddings())
def get_output_embeddings(self):
return self.lm_head
def set_output_embeddings(self, new_embeddings):
self.lm_head = new_embeddings
def set_decoder(self, decoder):
self.model = decoder
def get_decoder(self):
return self.model
def generate(self, *args, **kwargs):
try:
return super().generate(*args, **kwargs)
except AttributeError as exception:
if 'past_key_values' in str(exception):
raise AttributeError(
f"You tried to call `generate` with a decoding strategy that manipulates `past_key_values`, "
f"which is not supported for {self.__class__.__name__}. "
f"Try another generation strategy instead. "
f"For the available generation strategies, check this doc: "
f"https://huggingface.co/docs/transformers/en/generation_strategies#decoding-strategies",
)
else:
raise exception
@deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
def forward(
self,
input_ids: torch.LongTensor | None = None,
attention_mask: torch.Tensor | None = None,
inputs_embeds: torch.FloatTensor | None = None,
past_key_values: Cache | list[torch.FloatTensor] | None = None,
use_cache: bool | None = None,
output_attentions: bool | None = None,
output_hidden_states: bool | None = None,
return_dict: bool | None = None,
labels: torch.LongTensor | None = None,
logits_to_keep: int | None = 0,
**kwargs: Unpack[dict],
) -> tuple | CausalLMOutputWithPast:
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
outputs = self.model(
input_ids=input_ids,
attention_mask=attention_mask,
inputs_embeds=inputs_embeds,
past_key_values=past_key_values,
use_cache=use_cache,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_dict=return_dict,
**kwargs,
)
hidden_states = outputs[0]
# For fused linear cross-entropy we do not materialize logits for the full sequence
logits = None if self.config.fuse_linear_cross_entropy else self.lm_head(hidden_states[:, -logits_to_keep:])
loss = None
if labels is not None:
if getattr(self, 'criterion', None) is None:
if self.config.fuse_linear_cross_entropy:
criterion = FusedLinearCrossEntropyLoss(use_l2warp=self.config.use_l2warp)
elif self.config.fuse_cross_entropy:
criterion = FusedCrossEntropyLoss(inplace_backward=True)
else:
criterion = nn.CrossEntropyLoss()
else:
criterion = self.criterion
labels = labels.to(hidden_states.device)
labels = torch.cat((labels[..., 1:], torch.full_like(labels[:, :1], criterion.ignore_index)), 1)
if self.config.fuse_linear_cross_entropy:
loss = criterion(hidden_states, labels, self.lm_head.weight, self.lm_head.bias)
else:
loss = criterion(logits.view(labels.numel(), -1), labels.view(-1))
loss = l2_warp(loss, logits) if self.config.use_l2warp else loss
if not return_dict:
output = (logits,) + outputs[1:]
return ((loss,) + output) if loss is not None else output
return CausalLMOutputWithPast(
loss=loss,
logits=logits,
past_key_values=outputs.past_key_values,
hidden_states=outputs.hidden_states,
attentions=outputs.attentions,
)
|