File size: 65,326 Bytes
e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 17137e3 e8694e9 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 | """PyTorch CM3P model."""
from contextlib import nullcontext
from dataclasses import dataclass
from typing import Any, Optional, Union
import torch
import torch.utils.checkpoint
from torch import nn
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
from transformers import ModernBertModel, AutoModel, AutoModelForSequenceClassification, AutoModelForMaskedLM
from transformers.activations import ACT2FN
from transformers.modeling_outputs import (
BaseModelOutput,
BaseModelOutputWithPooling, MaskedLMOutput,
)
from transformers.modeling_utils import PreTrainedModel
from transformers.utils import ModelOutput, auto_docstring, can_return_tuple, logging
from .configuration_cm3p import CM3PConfig, CM3PMetadataConfig, CM3PBeatmapConfig, CM3PAudioConfig
logger = logging.get_logger(__name__)
# contrastive loss function, adapted from
# https://sachinruk.github.io/blog/2021-03-07-clip.html
def contrastive_loss(logits: torch.Tensor, target: torch.Tensor = None) -> torch.Tensor:
target = target if target is not None else torch.arange(len(logits), device=logits.device)
return nn.functional.cross_entropy(logits, target)
# CM3P loss function, adapted from CLIP
def cm3p_loss(similarity: torch.Tensor, metadata_variation_classes: torch.LongTensor = None) -> torch.Tensor:
if similarity.dim() == 3: # (metadata_batch_size, variations, beatmap_batch_size)
metadata_batch_size = similarity.size(0)
num_variations = similarity.size(1)
beatmap_batch_size = similarity.size(2)
assert metadata_batch_size == beatmap_batch_size
true_metadata_indices = (metadata_variation_classes == 0).int().argmax(dim=1)
metadata_loss = contrastive_loss(similarity[torch.arange(metadata_batch_size), true_metadata_indices]) # only use original metadata for loss
beatmap_similarity = similarity.permute(2, 0, 1) # (beatmap_batch_size, metadata_batch_size, variations)
beatmap_similarity = beatmap_similarity.reshape(beatmap_batch_size, -1) # (beatmap_batch_size, metadata_batch_size * variations)
target = torch.arange(0, beatmap_similarity.size(1), num_variations, device=similarity.device) # (metadata_batch_size,)
target += true_metadata_indices
beatmap_loss = contrastive_loss(beatmap_similarity, target=target)
else:
metadata_loss = contrastive_loss(similarity)
beatmap_loss = contrastive_loss(similarity.t())
return (metadata_loss + beatmap_loss) / 2.0
def _get_vector_norm(tensor: torch.Tensor) -> torch.Tensor:
"""
This method is equivalent to tensor.norm(p=2, dim=-1, keepdim=True) and used to make
model `executorch` exportable. See issue https://github.com/pytorch/executorch/issues/3566
"""
square_tensor = torch.pow(tensor, 2)
sum_tensor = torch.sum(square_tensor, dim=-1, keepdim=True)
normed_tensor = torch.pow(sum_tensor, 0.5)
return normed_tensor
def _unpad_cm3p_input(
inputs: torch.Tensor,
attention_mask: torch.Tensor,
position_ids: Optional[torch.Tensor] = None,
labels: Optional[torch.Tensor] = None,
) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor, int, Optional[torch.Tensor], Optional[torch.Tensor]]:
"""
Remove padding from input sequences.
Args:
inputs: (batch, seqlen, ...) or (batch, seqlen)
attention_mask: (batch, seqlen), bool / int, 1 means valid and 0 means not valid.
position_ids: (batch, seqlen), int, position ids
labels: (batch, seqlen), int, labels
Returns:
unpadded_inputs: (total_nnz, ...), where total_nnz = number of tokens selected in attention_mask.
indices: (total_nnz)
cu_seqlens: (batch + 1), the cumulative sequence lengths
max_seqlen_in_batch: int
unpadded_position_ids: (total_nnz) or None
unpadded_labels: (total_nnz) or None
"""
seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
max_seqlen_in_batch = int(seqlens_in_batch.max().item())
cu_seqlens = torch.nn.functional.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))
if inputs.dim() == 2:
unpadded_inputs = inputs.flatten()[indices]
else:
batch, seqlen, *rest = inputs.shape
shape = batch * seqlen
unpadded_inputs = inputs.view(shape, *rest)[indices]
unpadded_position_ids = position_ids.flatten()[indices] if position_ids is not None else None
unpadded_labels = labels.flatten()[indices] if labels is not None else None
return unpadded_inputs, indices, cu_seqlens, max_seqlen_in_batch, unpadded_position_ids, unpadded_labels
def _pad_cm3p_output(
inputs: torch.Tensor,
indices: torch.Tensor,
batch: int,
seqlen: int,
) -> torch.Tensor:
"""
Add padding to sequences.
Args:
inputs: (total_nnz, ...) or (total_nnz,), where total_nnz = number of tokens selected in attention_mask.
indices: (total_nnz)
batch: int, batch size
seqlen: int, max sequence length
Returns:
padded_inputs: (batch, seqlen, ...) or (batch, seqlen)
"""
if inputs.dim() == 1:
output = torch.zeros(batch * seqlen, dtype=inputs.dtype, device=inputs.device)
output[indices] = inputs
padded_inputs = output.view(batch, seqlen)
else:
_, *rest = inputs.shape
output = torch.zeros(batch * seqlen, *rest, dtype=inputs.dtype, device=inputs.device)
output[indices] = inputs
padded_inputs = output.view(batch, seqlen, *rest)
return padded_inputs
@dataclass
class BeatmapClassifierOutput(ModelOutput):
"""
Base class for outputs of beatmap classification models.
Args:
loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
Classification (or regression if config.num_labels==1) loss.
logits (`torch.FloatTensor` of shape `(batch_size, config.num_labels)`):
Classification (or regression if config.num_labels==1) scores (before SoftMax).
hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, +
one for the output of each stage) of shape `(batch_size, sequence_length, hidden_size)`. Hidden-states
(also called feature maps) of the model at the output of each stage.
attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, patch_size,
sequence_length)`.
Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
heads.
"""
loss: Optional[torch.FloatTensor] = None
logits: Optional[torch.FloatTensor] = None
hidden_states: Optional[tuple[torch.FloatTensor, ...]] = None
attentions: Optional[tuple[torch.FloatTensor, ...]] = None
@dataclass
@auto_docstring(
custom_intro="""
Base class for audio model's outputs that also contains a pooling of the last hidden states.
"""
)
class CM3PAudioModelOutput(BaseModelOutput):
r"""
audio_embeds (`torch.FloatTensor` of shape `(batch_size * sequence_length, output_dim)` *optional* returned when model is initialized with `with_projection=True`):
The audio embeddings obtained by applying the projection layer to the last hidden state.
"""
audio_embeds: Optional[torch.FloatTensor] = None
@dataclass
@auto_docstring(
custom_intro="""
Base class for beatmap model's outputs that also contains beatmap embeddings of the pooling of the last hidden states.
"""
)
class CM3PBeatmapModelOutput(BaseModelOutputWithPooling):
r"""
audio_model_output (`BaseModelOutput`):
The output of the audio model, which contains the last hidden state, hidden states, and attentions.
beatmap_embeds (`torch.FloatTensor` of shape `(batch_size, output_dim)` *optional* returned when model is initialized with `with_projection=True`):
The beatmap embeddings obtained by applying the projection layer to the pooler_output.
"""
beatmap_embeds: Optional[torch.FloatTensor] = None
audio_model_output: Optional[CM3PAudioModelOutput] = None
@dataclass
@auto_docstring(
custom_intro="""
Base class for metadata model's outputs that also contains a pooling of the last hidden states.
"""
)
class CM3PMetadataModelOutput(BaseModelOutput):
r"""
metadata_embeds (`torch.FloatTensor` of shape `(batch_size, output_dim)` *optional* returned when model is initialized with `with_projection=True`):
The metadata embeddings obtained by applying the projection layer to the pooler_output.
"""
metadata_embeds: Optional[torch.FloatTensor] = None
@dataclass
@auto_docstring
class CM3POutput(ModelOutput):
r"""
loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `return_loss` is `True`):
Contrastive loss for beatmap-metadata similarity.
logits_per_beatmap (`torch.FloatTensor` of shape `(beatmap_batch_size, metadata_batch_size)`):
The scaled dot product scores between `beatmap_embeds` and `metadata_embeds`. This represents the beatmap-metadata
similarity scores.
logits_per_metadata (`torch.FloatTensor` of shape `(metadata_batch_size, beatmap_batch_size)`):
The scaled dot product scores between `metadata_embeds` and `beatmap_embeds`. This represents the metadata-beatmap
similarity scores.
metadata_embeds (`torch.FloatTensor` of shape `(batch_size, output_dim`):
The metadata embeddings obtained by applying the projection layer to the pooled output of [`CM3PMetadataModel`].
beatmap_embeds (`torch.FloatTensor` of shape `(batch_size, output_dim`):
The beatmap embeddings obtained by applying the projection layer to the pooled output of [`CM3PBeatmapModel`].
logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, vocab_size)`, *optional*, returned when `labels` is provided):
Prediction scores of the masked language modeling head. Only computed if `labels` is provided.
metadata_model_output (`BaseModelOutputWithPooling`):
The output of the [`CM3PMetadataModel`].
beatmap_model_output (`BaseModelOutputWithPooling`):
The output of the [`CM3PBeatmapModel`].
"""
loss: Optional[torch.FloatTensor] = None
logits_per_beatmap: Optional[torch.Tensor] = None
logits_per_metadata: Optional[torch.Tensor] = None
metadata_embeds: Optional[torch.FloatTensor] = None
beatmap_embeds: Optional[torch.FloatTensor] = None
logits: Optional[torch.FloatTensor] = None
metadata_model_output: BaseModelOutputWithPooling = None
beatmap_model_output: BaseModelOutputWithPooling = None
def to_tuple(self) -> tuple[Any]:
return tuple(
self[k] if k not in ["metadata_model_output", "beatmap_model_output"] else getattr(self, k).to_tuple()
for k in self.keys()
)
@auto_docstring
class CM3PPreTrainedModel(PreTrainedModel):
config_class = CM3PConfig
base_model_prefix = "cm3p"
supports_gradient_checkpointing = True
_supports_flash_attn_2 = True
_supports_sdpa = True
_supports_flex_attn = False
def _init_weights(self, module):
"""Initialize the weights"""
if isinstance(module, (nn.Linear, nn.Conv1d)):
nn.init.normal_(module.weight, std=self.config.initializer_range)
if module.bias is not None:
module.bias.data.zero_()
elif isinstance(module, nn.LayerNorm):
module.weight.data.fill_(1.0)
if module.bias is not None:
module.bias.data.zero_()
elif isinstance(module, ModernBertModel):
module.initialize_weights()
elif isinstance(module, CM3PModel):
nn.init.normal_(
module.metadata_projection.weight,
std=module.metadata_embed_dim**-0.5 * self.config.initializer_factor,
)
nn.init.normal_(
module.beatmap_projection.weight,
std=module.beatmap_embed_dim**-0.5 * self.config.initializer_factor,
)
elif isinstance(module, CM3PBeatmapModelWithProjection):
nn.init.normal_(
module.beatmap_projection.weight,
std=self.config.hidden_size**-0.5 * self.config.initializer_factor,
)
elif isinstance(module, CM3PMetadataModelWithProjection):
nn.init.normal_(
module.metadata_projection.weight,
std=self.config.hidden_size**-0.5 * self.config.initializer_factor,
)
elif isinstance(module, CM3PForBeatmapClassification):
nn.init.normal_(
module.classifier.weight,
std=self.config.hidden_size**-0.5 * self.config.initializer_factor,
)
class CM3PMetadataTransformer(nn.Module):
def __init__(self, config: CM3PMetadataConfig):
super().__init__()
self.config = config
# noinspection PyTypeChecker
self.encoder = ModernBertModel(config)
def get_input_embeddings(self):
return self.encoder.get_input_embeddings()
def set_input_embeddings(self, value):
self.encoder.set_input_embeddings(value)
@can_return_tuple
@auto_docstring
def forward(
self,
input_ids: Optional[torch.Tensor] = None,
attention_mask: Optional[torch.Tensor] = None,
indices: Optional[torch.Tensor] = None,
cu_seqlens: Optional[torch.Tensor] = None,
max_seqlen: Optional[int] = None,
batch_size: Optional[int] = None,
seq_len: Optional[int] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
output_pooler: bool = True,
) -> BaseModelOutputWithPooling:
r"""
indices (`torch.Tensor` of shape `(total_unpadded_tokens,)`, *optional*):
Indices of the non-padding tokens in the input sequence. Used for unpadding the output.
cu_seqlens (`torch.Tensor` of shape `(batch + 1,)`, *optional*):
Cumulative sequence lengths of the input sequences. Used to index the unpadded tensors.
max_seqlen (`int`, *optional*):
Maximum sequence length in the batch excluding padding tokens. Used to unpad input_ids and pad output tensors.
batch_size (`int`, *optional*):
Batch size of the input sequences. Used to pad the output tensors.
seq_len (`int`, *optional*):
Sequence length of the input sequences including padding tokens. Used to pad the output tensors.
output_pooler (`bool`, *optional*, defaults to `True`):
Whether to return the pooled output of the model. The pooled output is usually the representation of
the first token (CLS) or the mean of the token representations.
"""
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
)
if input_ids is None:
raise ValueError("You have to specify input_ids")
is_3d = input_ids.dim() == 3
batch_size_3d = input_ids.size(0)
if is_3d:
# flatten to 2D batch if multiple metadata variations are provided
input_ids = input_ids.view(-1, input_ids.size(-1))
if attention_mask is not None:
attention_mask = attention_mask.view(-1, attention_mask.size(-1))
encoder_outputs: BaseModelOutput = self.encoder(
input_ids=input_ids,
attention_mask=attention_mask,
indices=indices,
cu_seqlens=cu_seqlens,
max_seqlen=max_seqlen,
batch_size=batch_size,
seq_len=seq_len,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
)
last_hidden_state = encoder_outputs.last_hidden_state
pooled_output = None
if is_3d:
# un-flatten back to 3D batch (batch_size, variations, seq_length, hidden_size)
last_hidden_state = last_hidden_state.view(
batch_size_3d, -1, last_hidden_state.size(-2), last_hidden_state.size(-1)
)
if attention_mask is not None:
attention_mask = attention_mask.view(batch_size_3d, -1, attention_mask.size(-1))
if output_pooler:
if indices is not None:
raise NotImplementedError("Pooling with unpadded input is not implemented yet.")
if self.config.cls_embed:
pooled_output = last_hidden_state[..., 0, :]
elif attention_mask is not None:
# Use the attention mask to exclude padding tokens
expanded_attention_mask = attention_mask.unsqueeze(-1).float()
masked_hidden_states = last_hidden_state * expanded_attention_mask
sum_hidden_states = torch.sum(masked_hidden_states, dim=-2)
sum_attention_mask = torch.sum(expanded_attention_mask, dim=-2)
pooled_output = sum_hidden_states / torch.clamp(sum_attention_mask, min=1e-9)
pooled_output = pooled_output.to(dtype=last_hidden_state.dtype)
else:
pooled_output = torch.mean(last_hidden_state, dim=-2)
return BaseModelOutputWithPooling(
last_hidden_state=last_hidden_state,
pooler_output=pooled_output,
hidden_states=encoder_outputs.hidden_states,
attentions=encoder_outputs.attentions,
)
@auto_docstring(
custom_intro="""
The metadata model from CM3P without any head or projection on top.
"""
)
class CM3PMetadataModel(CM3PPreTrainedModel):
config_class = CM3PMetadataConfig
def __init__(self, config: CM3PMetadataConfig):
super().__init__(config)
self.metadata_model = CM3PMetadataTransformer(config)
# Initialize weights and apply final processing
self.post_init()
def get_input_embeddings(self) -> nn.Module:
return self.metadata_model.encoder.embeddings.tok_embeddings
def set_input_embeddings(self, value):
self.metadata_model.encoder.embeddings.tok_embeddings = value
@can_return_tuple
@auto_docstring
def forward(
self,
input_ids: Optional[torch.Tensor] = None,
attention_mask: Optional[torch.Tensor] = None,
indices: Optional[torch.Tensor] = None,
cu_seqlens: Optional[torch.Tensor] = None,
max_seqlen: Optional[int] = None,
batch_size: Optional[int] = None,
seq_len: Optional[int] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
output_pooler: bool = True,
) -> BaseModelOutputWithPooling:
r"""
indices (`torch.Tensor` of shape `(total_unpadded_tokens,)`, *optional*):
Indices of the non-padding tokens in the input sequence. Used for unpadding the output.
cu_seqlens (`torch.Tensor` of shape `(batch + 1,)`, *optional*):
Cumulative sequence lengths of the input sequences. Used to index the unpadded tensors.
max_seqlen (`int`, *optional*):
Maximum sequence length in the batch excluding padding tokens. Used to unpad input_ids and pad output tensors.
batch_size (`int`, *optional*):
Batch size of the input sequences. Used to pad the output tensors.
seq_len (`int`, *optional*):
Sequence length of the input sequences including padding tokens. Used to pad the output tensors.
output_pooler (`bool`, *optional*, defaults to `True`):
Whether to return the pooled output of the model. The pooled output is usually the representation of
the first token (CLS) or the mean of the token representations.
"""
return self.metadata_model(
input_ids=input_ids,
attention_mask=attention_mask,
indices=indices,
cu_seqlens=cu_seqlens,
max_seqlen=max_seqlen,
batch_size=batch_size,
seq_len=seq_len,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
output_pooler=output_pooler,
)
class CM3PMultiModalProjector(nn.Module):
def __init__(self, config: CM3PAudioConfig):
super().__init__()
self.linear_1 = nn.Linear(config.projector_intermediate_size, config.projector_dim, bias=False)
self.act = ACT2FN[config.projector_hidden_act]
self.linear_2 = nn.Linear(config.projector_dim, config.projector_dim, bias=False)
def forward(self, audio_features):
hidden_states = self.linear_1(audio_features)
hidden_states = self.act(hidden_states)
hidden_states = self.linear_2(hidden_states)
return hidden_states
class CM3PAudioEncoder(nn.Module):
def __init__(self, config: CM3PAudioConfig):
super().__init__()
self.config = config
self.conv1 = nn.Conv1d(config.n_mels, config.hidden_size, kernel_size=3, padding=1)
self.conv2 = nn.Conv1d(config.hidden_size, config.hidden_size, kernel_size=3, stride=2, padding=1)
# noinspection PyTypeChecker
self.encoder = ModernBertModel(config)
self.multi_modal_projector = CM3PMultiModalProjector(config)
def forward(
self,
input_features: torch.FloatTensor,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
) -> CM3PAudioModelOutput:
# Conv layers from Whisper followed by an modern Bert encoder
inputs_embeds = nn.functional.gelu(self.conv1(input_features))
inputs_embeds = nn.functional.gelu(self.conv2(inputs_embeds))
inputs_embeds = inputs_embeds.permute(0, 2, 1).contiguous()
position_ids = torch.arange(inputs_embeds.size(1), device=inputs_embeds.device).unsqueeze(0).repeat(
inputs_embeds.size(0), 1)
encoder_outputs: BaseModelOutput = self.encoder(
inputs_embeds=inputs_embeds,
position_ids=position_ids,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
)
# Reduce the sequence length and project to the beatmap hidden size
audio_hidden_states = encoder_outputs.last_hidden_state
audio_hidden_states = audio_hidden_states.reshape(-1, self.config.projector_intermediate_size)
audio_embeds = self.multi_modal_projector(audio_hidden_states)
audio_outputs = CM3PAudioModelOutput(
audio_embeds=audio_embeds,
last_hidden_state=encoder_outputs.last_hidden_state,
hidden_states=encoder_outputs.hidden_states,
attentions=encoder_outputs.attentions,
)
return audio_outputs
class CM3PBeatmapTransformer(nn.Module):
def __init__(self, config: CM3PBeatmapConfig):
super().__init__()
self.config = config
self.audio_encoder = CM3PAudioEncoder(config.audio_config)
# noinspection PyTypeChecker
self.encoder = ModernBertModel(config)
def get_input_embeddings(self):
return self.encoder.get_input_embeddings()
def set_input_embeddings(self, value):
self.encoder.set_input_embeddings(value)
@can_return_tuple
@auto_docstring
def forward(
self,
input_ids: Optional[torch.LongTensor] = None,
input_features: Optional[torch.FloatTensor] = None,
attention_mask: Optional[torch.FloatTensor] = None,
sliding_window_mask: Optional[torch.FloatTensor] = None,
position_ids: Optional[torch.LongTensor] = None,
inputs_embeds: Optional[torch.FloatTensor] = None,
indices: Optional[torch.Tensor] = None,
cu_seqlens: Optional[torch.Tensor] = None,
max_seqlen: Optional[int] = None,
batch_size: Optional[int] = None,
seq_len: Optional[int] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
output_pooler: bool = True,
) -> CM3PBeatmapModelOutput:
r"""
input_features (`torch.FloatTensor` of shape `(batch_size, num_frames, num_mels)`, *optional*):
The audio frames to be processed by the audio encoder. If provided, the model will use these frames to
compute the beatmap embeddings.
sliding_window_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
Mask to avoid performing attention on padding or far-away tokens. In ModernBert, only every few layers
perform global attention, while the rest perform local attention. This mask is used to avoid attending to
far-away tokens in the local attention layers when not using Flash Attention.
indices (`torch.Tensor` of shape `(total_unpadded_tokens,)`, *optional*):
Indices of the non-padding tokens in the input sequence. Used for unpadding the output.
cu_seqlens (`torch.Tensor` of shape `(batch + 1,)`, *optional*):
Cumulative sequence lengths of the input sequences. Used to index the unpadded tensors.
max_seqlen (`int`, *optional*):
Maximum sequence length in the batch excluding padding tokens. Used to unpad input_ids and pad output tensors.
batch_size (`int`, *optional*):
Batch size of the input sequences. Used to pad the output tensors.
seq_len (`int`, *optional*):
Sequence length of the input sequences including padding tokens. Used to pad the output tensors.
output_pooler (`bool`, *optional*, defaults to `True`):
Whether to return the pooled output of the model. The pooled output is usually the representation of
the first token (CLS) or the mean of the token representations.
"""
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
)
if inputs_embeds is None:
inputs_embeds = self.get_input_embeddings()(input_ids)
audio_model_outputs = None
if input_features is not None:
audio_model_outputs = self.audio_encoder(
input_features=input_features,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
)
# replace text-audio token placeholders with audio embeddings
audio_embeds = audio_model_outputs.audio_embeds.to(dtype=inputs_embeds.dtype)
audio_token_mask = input_ids == self.config.audio_token_id
inputs_embeds[audio_token_mask] = audio_embeds
encoder_outputs: BaseModelOutput = self.encoder(
inputs_embeds=inputs_embeds,
attention_mask=attention_mask,
sliding_window_mask=sliding_window_mask,
position_ids=position_ids,
indices=indices,
cu_seqlens=cu_seqlens,
max_seqlen=max_seqlen,
batch_size=batch_size,
seq_len=seq_len,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
)
last_hidden_state = encoder_outputs.last_hidden_state
pooled_output = None
if output_pooler:
if indices is not None:
if self.config.cls_embed:
pooled_output = last_hidden_state[cu_seqlens[:-1]]
else:
raise NotImplementedError("Pooling with unpadded input is not implemented yet.")
else:
if self.config.cls_embed:
pooled_output = last_hidden_state[:, 0]
elif attention_mask is not None:
# Use the attention mask to exclude padding tokens
expanded_attention_mask = attention_mask.unsqueeze(-1).float()
masked_hidden_states = last_hidden_state * expanded_attention_mask
sum_hidden_states = torch.sum(masked_hidden_states, dim=1)
sum_attention_mask = torch.sum(expanded_attention_mask, dim=1)
pooled_output = sum_hidden_states / torch.clamp(sum_attention_mask, min=1e-9)
pooled_output = pooled_output.to(dtype=last_hidden_state.dtype)
else:
pooled_output = torch.mean(last_hidden_state, dim=1)
return CM3PBeatmapModelOutput(
last_hidden_state=last_hidden_state,
pooler_output=pooled_output,
hidden_states=encoder_outputs.hidden_states,
attentions=encoder_outputs.attentions,
audio_model_output=audio_model_outputs,
)
@auto_docstring(
custom_intro="""
The beatmap model from CM3P without any head or projection on top.
"""
)
class CM3PBeatmapModel(CM3PPreTrainedModel):
config_class = CM3PBeatmapConfig
main_input_name = "input_ids"
def __init__(self, config: CM3PBeatmapConfig):
super().__init__(config)
self.beatmap_model = CM3PBeatmapTransformer(config)
# Initialize weights and apply final processing
self.post_init()
def get_input_embeddings(self) -> nn.Module:
return self.beatmap_model.encoder.embeddings.tok_embeddings
def set_input_embeddings(self, value):
self.beatmap_model.encoder.embeddings.tok_embeddings = value
@can_return_tuple
@auto_docstring
def forward(
self,
input_ids: Optional[torch.LongTensor] = None,
input_features: Optional[torch.FloatTensor] = None,
attention_mask: Optional[torch.FloatTensor] = None,
position_ids: Optional[torch.LongTensor] = None,
inputs_embeds: Optional[torch.FloatTensor] = None,
indices: Optional[torch.Tensor] = None,
cu_seqlens: Optional[torch.Tensor] = None,
max_seqlen: Optional[int] = None,
batch_size: Optional[int] = None,
seq_len: Optional[int] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
output_pooler: bool = True,
) -> CM3PBeatmapModelOutput:
r"""
input_features (`torch.FloatTensor` of shape `(batch_size, num_frames, num_mels)`, *optional*):
The audio frames to be processed by the audio encoder. If provided, the model will use these frames to
compute the beatmap embeddings.
indices (`torch.Tensor` of shape `(total_unpadded_tokens,)`, *optional*):
Indices of the non-padding tokens in the input sequence. Used for unpadding the output.
cu_seqlens (`torch.Tensor` of shape `(batch + 1,)`, *optional*):
Cumulative sequence lengths of the input sequences. Used to index the unpadded tensors.
max_seqlen (`int`, *optional*):
Maximum sequence length in the batch excluding padding tokens. Used to unpad input_ids and pad output tensors.
batch_size (`int`, *optional*):
Batch size of the input sequences. Used to pad the output tensors.
seq_len (`int`, *optional*):
Sequence length of the input sequences including padding tokens. Used to pad the output tensors.
output_pooler (`bool`, *optional*, defaults to `True`):
Whether to return the pooled output of the model. The pooled output is usually the representation of
the first token (CLS) or the mean of the token representations.
"""
return self.beatmap_model(
input_ids=input_ids,
input_features=input_features,
attention_mask=attention_mask,
position_ids=position_ids,
inputs_embeds=inputs_embeds,
indices=indices,
cu_seqlens=cu_seqlens,
max_seqlen=max_seqlen,
batch_size=batch_size,
seq_len=seq_len,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
output_pooler=output_pooler,
)
@auto_docstring
class CM3PModel(CM3PPreTrainedModel):
config_class = CM3PConfig
def __init__(self, config: CM3PConfig):
super().__init__(config)
if not isinstance(config.metadata_config, CM3PMetadataConfig):
raise TypeError(
"config.metadata_config is expected to be of type CM3PMetadataConfig but is of type"
f" {type(config.metadata_config)}."
)
if not isinstance(config.beatmap_config, CM3PBeatmapConfig):
raise TypeError(
"config.beatmap_config is expected to be of type CM3PBeatmapConfig but is of type"
f" {type(config.beatmap_config)}."
)
metadata_config = config.metadata_config
beatmap_config = config.beatmap_config
self.projection_dim: int = config.projection_dim
self.metadata_embed_dim: int = metadata_config.hidden_size
self.beatmap_embed_dim: int = beatmap_config.hidden_size
self.loss_type = config.loss_type
metadata_model = CM3PMetadataModel._from_config(metadata_config)
self.metadata_model = metadata_model.metadata_model
beatmap_model = CM3PBeatmapModel._from_config(beatmap_config)
self.beatmap_model = beatmap_model.beatmap_model
self.beatmap_projection = nn.Linear(self.beatmap_embed_dim, self.projection_dim, bias=False)
self.metadata_projection = nn.Linear(self.metadata_embed_dim, self.projection_dim, bias=False)
self.logit_scale = nn.Parameter(torch.tensor(self.config.logit_scale_init_value))
if config.has_decoder_head:
self.head = CM3PPredictionHead(beatmap_config)
self.decoder = nn.Linear(beatmap_config.hidden_size, beatmap_config.vocab_size, bias=beatmap_config.decoder_bias)
# Initialize weights and apply final processing
self.post_init()
@auto_docstring
def get_metadata_features(
self,
input_ids: Optional[torch.LongTensor] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
) -> torch.FloatTensor:
r"""
input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
The input IDs for the metadata model. The model will use these IDs to compute the metadata embeddings.
Returns:
metadata_features (`torch.FloatTensor` of shape `(batch_size, output_dim`): The metadata embeddings obtained by
applying the projection layer to the pooled output of [`CM3PMetadataModel`].
"""
# Use CM3P model's config for some fields (if specified) instead of those of beatmap & metadata components.
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
)
metadata_outputs: BaseModelOutputWithPooling = self.metadata_model(
input_ids=input_ids,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
)
pooled_output = metadata_outputs.pooler_output
metadata_features = self.metadata_projection(pooled_output)
return metadata_features
@auto_docstring
def get_beatmap_features(
self,
input_ids: Optional[torch.LongTensor] = None,
input_features: Optional[torch.FloatTensor] = None,
attention_mask: Optional[torch.Tensor] = None,
position_ids: Optional[torch.LongTensor] = None,
inputs_embeds: Optional[torch.FloatTensor] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
) -> torch.FloatTensor:
r"""
input_features (`torch.FloatTensor` of shape `(batch_size, num_frames, num_mels)`, *optional*):
The audio frames to be processed by the audio encoder. If provided, the model will use these frames to
compute the beatmap embeddings.
Returns:
beatmap_features (`torch.FloatTensor` of shape `(batch_size, output_dim`): The beatmap embeddings obtained by
applying the projection layer to the pooled output of [`CM3PBeatmapModel`].
"""
# Use CM3P model's config for some fields (if specified) instead of those of beatmap & metadata components.
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
)
beatmap_outputs: BaseModelOutputWithPooling = self.beatmap_model(
input_ids=input_ids,
input_features=input_features,
attention_mask=attention_mask,
position_ids=position_ids,
inputs_embeds=inputs_embeds,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
)
pooled_output = beatmap_outputs.pooler_output
beatmap_features = self.beatmap_projection(pooled_output)
return beatmap_features
@torch.compile(dynamic=True)
def compiled_head(self, output: torch.Tensor) -> torch.Tensor:
return self.decoder(self.head(output))
@can_return_tuple
@auto_docstring
def forward(
self,
input_ids: Optional[torch.LongTensor] = None,
input_features: Optional[torch.FloatTensor] = None,
metadata_ids: Optional[torch.LongTensor] = None,
attention_mask: Optional[torch.Tensor] = None,
metadata_attention_mask: Optional[torch.Tensor] = None,
position_ids: Optional[torch.LongTensor] = None,
inputs_embeds: Optional[torch.FloatTensor] = None,
metadata_variation_classes: Optional[torch.LongTensor] = None,
labels: Optional[torch.Tensor] = None,
indices: Optional[torch.Tensor] = None,
cu_seqlens: Optional[torch.Tensor] = None,
max_seqlen: Optional[int] = None,
batch_size: Optional[int] = None,
seq_len: Optional[int] = None,
return_loss: Optional[bool] = True,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
output_logits: Optional[bool] = None,
**kwargs,
) -> CM3POutput:
r"""
input_features (`torch.FloatTensor` of shape `(batch_size, num_frames, num_mels)`, *optional*):
The audio frames to be processed by the audio encoder. If provided, the model will use these frames to
compute the beatmap embeddings.
metadata_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)` or `(batch_size, variations, sequence_length)`):
The input IDs for the metadata model. The model will use these IDs to compute the metadata embeddings.
metadata_attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)` or `(batch_size, variations, sequence_length)`, *optional*):
The attention mask for the metadata model. If provided, the model will not attend to the padded tokens.
metadata_variation_classes (`torch.LongTensor` of shape `(batch_size, variations)`, *optional*):
Tells the model what kind of variation each metadata sequence is.
0 indicates the original metadata, -1 indicates paddidng, and any positive integer indicates a specific variation class.
indices (`torch.Tensor` of shape `(total_unpadded_tokens,)`, *optional*):
Indices of the non-padding tokens in the input sequence. Used for unpadding the output.
cu_seqlens (`torch.Tensor` of shape `(batch + 1,)`, *optional*):
Cumulative sequence lengths of the input sequences. Used to index the unpadded tensors.
max_seqlen (`int`, *optional*):
Maximum sequence length in the batch excluding padding tokens. Used to unpad input_ids and pad output tensors.
batch_size (`int`, *optional*):
Batch size of the input sequences. Used to pad the output tensors.
seq_len (`int`, *optional*):
Sequence length of the input sequences including padding tokens. Used to pad the output tensors.
return_loss (`bool`, *optional*):
Whether to return the contrastive loss.
output_logits (`bool`, *optional*):
Whether to return the logits from the decoder head.
"""
# Use CM3P model's config for some fields (if specified) instead of those of beatmap & metadata components.
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
)
output_logits = output_logits if output_logits is not None else self.config.has_decoder_head
if metadata_ids is not None and metadata_ids.dim() == 3 and return_loss and metadata_variation_classes is None:
raise ValueError("When providing multiple metadata variations, metadata_variation_classes must be provided in order to compute loss correctly.")
if output_logits and not self.config.has_decoder_head:
raise ValueError("Cannot return logits when the model is not configured with a decoder head.")
# noinspection PyProtectedMember
if self.config._attn_implementation == "flash_attention_2":
if indices is None and cu_seqlens is None and max_seqlen is None:
if batch_size is None and seq_len is None:
if inputs_embeds is not None:
batch_size, seq_len = inputs_embeds.shape[:2]
else:
batch_size, seq_len = input_ids.shape[:2]
device = input_ids.device if input_ids is not None else inputs_embeds.device
if attention_mask is None:
attention_mask = torch.ones((batch_size, seq_len), device=device, dtype=torch.bool)
if inputs_embeds is None:
with torch.no_grad():
input_ids, indices, cu_seqlens, max_seqlen, position_ids, labels = _unpad_cm3p_input(
inputs=input_ids, attention_mask=attention_mask, position_ids=position_ids, labels=labels
)
else:
inputs_embeds, indices, cu_seqlens, max_seqlen, position_ids, labels = _unpad_cm3p_input(
inputs=inputs_embeds, attention_mask=attention_mask, position_ids=position_ids, labels=labels
)
beatmap_embeds = None
beatmap_outputs = None
metadata_embeds = None
metadata_outputs = None
logits_per_beatmap = None
logits_per_metadata = None
loss = 0 if return_loss else None
logits = None
if input_ids is not None:
beatmap_outputs: BaseModelOutputWithPooling = self.beatmap_model(
input_ids=input_ids,
input_features=input_features,
attention_mask=attention_mask,
position_ids=position_ids,
inputs_embeds=inputs_embeds,
indices=indices,
cu_seqlens=cu_seqlens,
max_seqlen=max_seqlen,
batch_size=batch_size,
seq_len=seq_len,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
)
beatmap_embeds = beatmap_outputs.pooler_output
beatmap_embeds = self.beatmap_projection(beatmap_embeds)
beatmap_embeds = beatmap_embeds / _get_vector_norm(beatmap_embeds)
if metadata_ids is not None:
metadata_outputs: BaseModelOutputWithPooling = self.metadata_model(
input_ids=metadata_ids,
attention_mask=metadata_attention_mask,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
)
metadata_embeds = metadata_outputs.pooler_output
metadata_embeds = self.metadata_projection(metadata_embeds)
metadata_embeds = metadata_embeds / _get_vector_norm(metadata_embeds)
if metadata_embeds is not None and beatmap_embeds is not None:
# cosine similarity as logits
logits_per_metadata = torch.matmul(metadata_embeds, beatmap_embeds.t().to(metadata_embeds.device))
logits_per_metadata = logits_per_metadata * self.logit_scale.exp().to(metadata_embeds.device)
if logits_per_metadata.dim() == 3:
logits_per_beatmap = logits_per_metadata.permute(2, 0, 1)
else:
logits_per_beatmap = logits_per_metadata.t()
if return_loss:
loss = cm3p_loss(logits_per_metadata, metadata_variation_classes)
if output_logits:
logits = (
self.compiled_head(beatmap_outputs.last_hidden_state)
if self.config.beatmap_config.reference_compile
else self.decoder(self.head(beatmap_outputs.last_hidden_state))
)
if labels is not None and return_loss:
mlm_loss = self.loss_function(logits, labels, vocab_size=self.config.beatmap_config.vocab_size, **kwargs)
loss += 0.5 * mlm_loss
# noinspection PyProtectedMember
if self.config._attn_implementation == "flash_attention_2":
with nullcontext() if self.config.beatmap_config.repad_logits_with_grad or labels is None else torch.no_grad():
logits = _pad_cm3p_output(inputs=logits, indices=indices, batch=batch_size, seqlen=seq_len)
return CM3POutput(
loss=loss,
logits_per_beatmap=logits_per_beatmap,
logits_per_metadata=logits_per_metadata,
metadata_embeds=metadata_embeds,
beatmap_embeds=beatmap_embeds,
logits=logits,
metadata_model_output=metadata_outputs,
beatmap_model_output=beatmap_outputs,
)
@auto_docstring
class CM3PMetadataModelWithProjection(CM3PPreTrainedModel):
config_class = CM3PMetadataConfig
def __init__(self, config: CM3PMetadataConfig):
super().__init__(config)
metadata_model = CM3PMetadataModel._from_config(config)
self.metadata_model = metadata_model.metadata_model
self.metadata_projection = nn.Linear(config.hidden_size, config.projection_dim, bias=False)
# Initialize weights and apply final processing
self.post_init()
def get_input_embeddings(self) -> nn.Module:
return self.metadata_model.get_input_embeddings()
def set_input_embeddings(self, value):
self.metadata_model.set_input_embeddings(value)
@can_return_tuple
@auto_docstring
def forward(
self,
input_ids: Optional[torch.Tensor] = None,
attention_mask: Optional[torch.Tensor] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
) -> CM3PMetadataModelOutput:
r"""
input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
The input IDs for the metadata model. The model will use these IDs to compute the metadata embeddings.
Returns:
metadata_features (`torch.FloatTensor` of shape `(batch_size, output_dim`): The metadata embeddings obtained by
applying the projection layer to the pooled output of [`CM3PMetadataModel`].
"""
metadata_outputs: BaseModelOutputWithPooling = self.metadata_model(
input_ids=input_ids,
attention_mask=attention_mask,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
)
pooled_output = metadata_outputs.pooler_output
metadata_embeds = self.metadata_projection(pooled_output)
return CM3PMetadataModelOutput(
metadata_embeds=metadata_embeds,
last_hidden_state=metadata_outputs.last_hidden_state,
hidden_states=metadata_outputs.hidden_states,
attentions=metadata_outputs.attentions,
)
@auto_docstring
class CM3PBeatmapModelWithProjection(CM3PPreTrainedModel):
config_class = CM3PBeatmapConfig
def __init__(self, config: CM3PBeatmapConfig):
super().__init__(config)
beatmap_model = CM3PBeatmapModel._from_config(config)
self.beatmap_model = beatmap_model.beatmap_model
self.beatmap_projection = nn.Linear(config.hidden_size, config.projection_dim, bias=False)
# Initialize weights and apply final processing
self.post_init()
def get_input_embeddings(self) -> nn.Module:
return self.beatmap_model.get_input_embeddings()
def set_input_embeddings(self, value):
self.beatmap_model.set_input_embeddings(value)
@can_return_tuple
@auto_docstring
def forward(
self,
input_ids: Optional[torch.LongTensor] = None,
input_features: Optional[torch.FloatTensor] = None,
attention_mask: Optional[torch.Tensor] = None,
position_ids: Optional[torch.LongTensor] = None,
inputs_embeds: Optional[torch.FloatTensor] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
) -> CM3PBeatmapModelOutput:
r"""
input_features (`torch.FloatTensor` of shape `(batch_size, num_frames, num_mels)`, *optional*):
The audio frames to be processed by the audio encoder. If provided, the model will use these frames to
compute the beatmap embeddings.
Returns:
beatmap_features (`torch.FloatTensor` of shape `(batch_size, output_dim`): The beatmap embeddings obtained by
applying the projection layer to the pooled output of [`CM3PBeatmapModel`].
"""
beatmap_outputs: BaseModelOutputWithPooling = self.beatmap_model(
input_ids=input_ids,
input_features=input_features,
attention_mask=attention_mask,
position_ids=position_ids,
inputs_embeds=inputs_embeds,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
)
pooled_output = beatmap_outputs.pooler_output
beatmap_embeds = self.beatmap_projection(pooled_output)
return CM3PBeatmapModelOutput(
beatmap_embeds=beatmap_embeds,
pooler_output=pooled_output,
last_hidden_state=beatmap_outputs.last_hidden_state,
hidden_states=beatmap_outputs.hidden_states,
attentions=beatmap_outputs.attentions,
)
@auto_docstring(
custom_intro="""
CM3P beatmap encoder with an beatmap classification head on top (a linear layer on top of the pooled final hidden states of
the beatmap embeddings) e.g. for BeatmapNet.
"""
)
class CM3PForBeatmapClassification(CM3PPreTrainedModel):
config_class = CM3PBeatmapConfig
base_model_prefix = "beatmap_model"
def __init__(self, config: CM3PBeatmapConfig) -> None:
super().__init__(config)
self.num_labels = config.num_labels
beatmap_model = CM3PBeatmapModel._from_config(config)
self.beatmap_model = beatmap_model.beatmap_model
# Classifier head
self.classifier = (
nn.Linear(config.hidden_size, config.num_labels) if config.num_labels > 0 else nn.Identity()
)
# Initialize weights and apply final processing
self.post_init()
@can_return_tuple
@auto_docstring
def forward(
self,
input_ids: Optional[torch.LongTensor] = None,
input_features: Optional[torch.FloatTensor] = None,
attention_mask: Optional[torch.Tensor] = None,
position_ids: Optional[torch.LongTensor] = None,
inputs_embeds: Optional[torch.FloatTensor] = None,
labels: Optional[torch.Tensor] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
) -> BeatmapClassifierOutput:
r"""
input_features (`torch.FloatTensor` of shape `(batch_size, num_frames, num_mels)`, *optional*):
The audio frames to be processed by the audio encoder. If provided, the model will use these frames to
compute the beatmap embeddings.
labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
Labels for computing the beatmap classification/regression loss. Indices should be in `[0, ...,
config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
`config.num_labels > 1` a classification loss is computed (Cross-Entropy).
"""
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
)
outputs: BaseModelOutputWithPooling = self.beatmap_model(
input_ids=input_ids,
input_features=input_features,
attention_mask=attention_mask,
position_ids=position_ids,
inputs_embeds=inputs_embeds,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
)
pooled_output = outputs.pooler_output
logits = self.classifier(pooled_output)
loss = None
if labels is not None:
# move labels to correct device to enable model parallelism
labels = labels.to(logits.device)
if self.config.problem_type is None:
if self.num_labels == 1:
self.config.problem_type = "regression"
elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
self.config.problem_type = "single_label_classification"
else:
self.config.problem_type = "multi_label_classification"
if self.config.problem_type == "regression":
loss_fct = MSELoss()
if self.num_labels == 1:
loss = loss_fct(logits.squeeze(), labels.squeeze())
else:
loss = loss_fct(logits, labels)
elif self.config.problem_type == "single_label_classification":
loss_fct = CrossEntropyLoss()
loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
elif self.config.problem_type == "multi_label_classification":
loss_fct = BCEWithLogitsLoss()
loss = loss_fct(logits, labels)
return BeatmapClassifierOutput(
loss=loss,
logits=logits,
hidden_states=outputs.hidden_states,
attentions=outputs.attentions,
)
class CM3PPredictionHead(nn.Module):
def __init__(self, config: CM3PBeatmapConfig):
super().__init__()
self.config = config
self.dense = nn.Linear(config.hidden_size, config.hidden_size, config.classifier_bias)
self.act = ACT2FN[config.classifier_activation]
self.norm = nn.LayerNorm(config.hidden_size, eps=config.norm_eps, bias=config.norm_bias)
def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
return self.norm(self.act(self.dense(hidden_states)))
class CM3PForMaskedLM(CM3PPreTrainedModel):
config_class = CM3PBeatmapConfig
base_model_prefix = "beatmap_model"
_tied_weights_keys = ["decoder.weight"]
def __init__(self, config: CM3PBeatmapConfig):
super().__init__(config)
self.config = config
beatmap_model = CM3PBeatmapModel._from_config(config)
self.beatmap_model = beatmap_model.beatmap_model
self.head = CM3PPredictionHead(config)
self.decoder = nn.Linear(config.hidden_size, config.vocab_size, bias=config.decoder_bias)
self.sparse_prediction = self.config.sparse_prediction
self.sparse_pred_ignore_index = self.config.sparse_pred_ignore_index
# Initialize weights and apply final processing
self.post_init()
def get_output_embeddings(self):
return self.decoder
def set_output_embeddings(self, new_embeddings: nn.Linear):
self.decoder = new_embeddings
@torch.compile(dynamic=True)
def compiled_head(self, output: torch.Tensor) -> torch.Tensor:
return self.decoder(self.head(output))
@auto_docstring
def forward(
self,
input_ids: Optional[torch.LongTensor] = None,
input_features: Optional[torch.FloatTensor] = None,
attention_mask: Optional[torch.Tensor] = None,
sliding_window_mask: Optional[torch.Tensor] = None,
position_ids: Optional[torch.Tensor] = None,
inputs_embeds: Optional[torch.Tensor] = None,
labels: Optional[torch.Tensor] = None,
indices: Optional[torch.Tensor] = None,
cu_seqlens: Optional[torch.Tensor] = None,
max_seqlen: Optional[int] = None,
batch_size: Optional[int] = None,
seq_len: Optional[int] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
**kwargs,
) -> Union[tuple[torch.Tensor], MaskedLMOutput]:
r"""
input_features (`torch.FloatTensor` of shape `(batch_size, num_frames, num_mels)`, *optional*):
The audio frames to be processed by the audio encoder. If provided, the model will use these frames to
compute the beatmap embeddings.
sliding_window_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
Mask to avoid performing attention on padding or far-away tokens. In ModernBert, only every few layers
perform global attention, while the rest perform local attention. This mask is used to avoid attending to
far-away tokens in the local attention layers when not using Flash Attention.
indices (`torch.Tensor` of shape `(total_unpadded_tokens,)`, *optional*):
Indices of the non-padding tokens in the input sequence. Used for unpadding the output.
cu_seqlens (`torch.Tensor` of shape `(batch + 1,)`, *optional*):
Cumulative sequence lengths of the input sequences. Used to index the unpadded tensors.
max_seqlen (`int`, *optional*):
Maximum sequence length in the batch excluding padding tokens. Used to unpad input_ids and pad output tensors.
batch_size (`int`, *optional*):
Batch size of the input sequences. Used to pad the output tensors.
seq_len (`int`, *optional*):
Sequence length of the input sequences including padding tokens. Used to pad the output tensors.
"""
# noinspection PyProtectedMember
if self.config._attn_implementation == "flash_attention_2":
if indices is None and cu_seqlens is None and max_seqlen is None:
if batch_size is None and seq_len is None:
if inputs_embeds is not None:
batch_size, seq_len = inputs_embeds.shape[:2]
else:
batch_size, seq_len = input_ids.shape[:2]
device = input_ids.device if input_ids is not None else inputs_embeds.device
if attention_mask is None:
attention_mask = torch.ones((batch_size, seq_len), device=device, dtype=torch.bool)
if inputs_embeds is None:
with torch.no_grad():
input_ids, indices, cu_seqlens, max_seqlen, position_ids, labels = _unpad_cm3p_input(
inputs=input_ids, attention_mask=attention_mask, position_ids=position_ids, labels=labels
)
else:
inputs_embeds, indices, cu_seqlens, max_seqlen, position_ids, labels = _unpad_cm3p_input(
inputs=inputs_embeds, attention_mask=attention_mask, position_ids=position_ids, labels=labels
)
outputs = self.beatmap_model(
input_ids=input_ids,
input_features=input_features,
attention_mask=attention_mask,
sliding_window_mask=sliding_window_mask,
position_ids=position_ids,
inputs_embeds=inputs_embeds,
indices=indices,
cu_seqlens=cu_seqlens,
max_seqlen=max_seqlen,
batch_size=batch_size,
seq_len=seq_len,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
output_pooler=False,
)
last_hidden_state = outputs.last_hidden_state
if self.sparse_prediction and labels is not None:
# flatten labels and output first
labels = labels.view(-1)
last_hidden_state = last_hidden_state.view(labels.shape[0], -1)
# then filter out the non-masked tokens
mask_tokens = labels != self.sparse_pred_ignore_index
last_hidden_state = last_hidden_state[mask_tokens]
labels = labels[mask_tokens]
logits = (
self.compiled_head(last_hidden_state)
if self.config.reference_compile
else self.decoder(self.head(last_hidden_state))
)
loss = None
if labels is not None:
loss = self.loss_function(logits, labels, vocab_size=self.config.vocab_size, **kwargs)
# noinspection PyProtectedMember
if self.config._attn_implementation == "flash_attention_2":
with nullcontext() if self.config.repad_logits_with_grad or labels is None else torch.no_grad():
logits = _pad_cm3p_output(inputs=logits, indices=indices, batch=batch_size, seqlen=seq_len)
return MaskedLMOutput(
loss=loss,
logits=logits,
hidden_states=outputs.hidden_states,
attentions=outputs.attentions,
)
AutoModel.register(CM3PMetadataConfig, CM3PMetadataModel)
AutoModel.register(CM3PBeatmapConfig, CM3PBeatmapModel)
AutoModel.register(CM3PConfig, CM3PModel)
AutoModelForSequenceClassification.register(CM3PBeatmapConfig, CM3PForBeatmapClassification)
AutoModelForMaskedLM.register(CM3PBeatmapConfig, CM3PForMaskedLM)
__all__ = [
"CM3PModel",
"CM3PPreTrainedModel",
"CM3PMetadataModel",
"CM3PMetadataModelWithProjection",
"CM3PBeatmapModel",
"CM3PBeatmapModelWithProjection",
"CM3PForBeatmapClassification",
"CM3PForMaskedLM",
]
|