Skip to content

vllm.model_executor.models.bert

BertAttention

Bases: Module

Source code in vllm/model_executor/models/bert.py 
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
class BertAttention(nn.Module):

    def __init__(
        self,
        hidden_size: int,
        num_attention_heads: int,
        layer_norm_eps: float,
        cache_config: Optional[CacheConfig] = None,
        quant_config: Optional[QuantizationConfig] = None,
        prefix: str = "",
    ):
        super().__init__()

        self.self = BertSelfAttention(hidden_size=hidden_size,
                                      num_attention_heads=num_attention_heads,
                                      cache_config=cache_config,
                                      quant_config=quant_config,
                                      prefix=f"{prefix}.output")

        self.output = BertSelfOutput(hidden_size=hidden_size,
                                     layer_norm_eps=layer_norm_eps,
                                     quant_config=quant_config,
                                     prefix=f"{prefix}.output")

    def forward(
        self,
        hidden_states: torch.Tensor,
    ) -> torch.Tensor:
        self_output = self.self(hidden_states)
        return self.output(self_output, hidden_states)

output instance-attribute 

output = BertSelfOutput(
    hidden_size=hidden_size,
    layer_norm_eps=layer_norm_eps,
    quant_config=quant_config,
    prefix=f"{prefix}.output",
)

self instance-attribute 

self = BertSelfAttention(
    hidden_size=hidden_size,
    num_attention_heads=num_attention_heads,
    cache_config=cache_config,
    quant_config=quant_config,
    prefix=f"{prefix}.output",
)

__init__ 

__init__(
    hidden_size: int,
    num_attention_heads: int,
    layer_norm_eps: float,
    cache_config: Optional[CacheConfig] = None,
    quant_config: Optional[QuantizationConfig] = None,
    prefix: str = "",
)
Source code in vllm/model_executor/models/bert.py 
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
def __init__(
    self,
    hidden_size: int,
    num_attention_heads: int,
    layer_norm_eps: float,
    cache_config: Optional[CacheConfig] = None,
    quant_config: Optional[QuantizationConfig] = None,
    prefix: str = "",
):
    super().__init__()

    self.self = BertSelfAttention(hidden_size=hidden_size,
                                  num_attention_heads=num_attention_heads,
                                  cache_config=cache_config,
                                  quant_config=quant_config,
                                  prefix=f"{prefix}.output")

    self.output = BertSelfOutput(hidden_size=hidden_size,
                                 layer_norm_eps=layer_norm_eps,
                                 quant_config=quant_config,
                                 prefix=f"{prefix}.output")

forward 

forward(hidden_states: Tensor) -> Tensor
Source code in vllm/model_executor/models/bert.py 
186
187
188
189
190
191
def forward(
    self,
    hidden_states: torch.Tensor,
) -> torch.Tensor:
    self_output = self.self(hidden_states)
    return self.output(self_output, hidden_states)

BertEmbedding

Bases: Module

Source code in vllm/model_executor/models/bert.py 
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
class BertEmbedding(nn.Module):

    def __init__(self, config: BertConfig):

        super().__init__()
        self.size = config.hidden_size
        self.word_embeddings = VocabParallelEmbedding(config.vocab_size,
                                                      config.hidden_size)
        self.position_embeddings = VocabParallelEmbedding(
            config.max_position_embeddings, config.hidden_size)
        self.token_type_embeddings = VocabParallelEmbedding(
            config.type_vocab_size, config.hidden_size)
        self.LayerNorm = nn.LayerNorm(config.hidden_size,
                                      eps=config.layer_norm_eps)
        self.position_ids = nn.Parameter(
            torch.empty((1, config.max_position_embeddings)), )

        self.position_embedding_type = config.position_embedding_type
        if self.position_embedding_type != "absolute":
            raise ValueError("Only 'absolute' position_embedding_type" +
                             " is supported")

    def forward(
        self,
        input_ids: torch.Tensor,
        seq_lens: torch.Tensor,
        position_ids: torch.Tensor,
        token_type_ids: Optional[torch.Tensor] = None,
    ) -> torch.Tensor:
        input_shape = input_ids.size()

        # Input embeddings.
        inputs_embeds = self.word_embeddings(input_ids)

        # Position embeddings.
        position_embeddings = self.position_embeddings(position_ids)

        if token_type_ids is None:
            token_type_ids = torch.zeros(input_shape,
                                         dtype=torch.long,
                                         device=inputs_embeds.device)

        token_type_embeddings = self.token_type_embeddings(token_type_ids)

        embeddings = inputs_embeds + token_type_embeddings + position_embeddings
        embeddings = self.LayerNorm(embeddings)
        return embeddings

LayerNorm instance-attribute 

LayerNorm = LayerNorm(hidden_size, eps=layer_norm_eps)

position_embedding_type instance-attribute 

position_embedding_type = position_embedding_type

position_embeddings instance-attribute 

position_embeddings = VocabParallelEmbedding(
    max_position_embeddings, hidden_size
)

position_ids instance-attribute 

position_ids = Parameter(
    empty((1, max_position_embeddings))
)

size instance-attribute 

size = hidden_size

token_type_embeddings instance-attribute 

token_type_embeddings = VocabParallelEmbedding(
    type_vocab_size, hidden_size
)

word_embeddings instance-attribute 

word_embeddings = VocabParallelEmbedding(
    vocab_size, hidden_size
)

__init__ 

__init__(config: BertConfig)
Source code in vllm/model_executor/models/bert.py 
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
def __init__(self, config: BertConfig):

    super().__init__()
    self.size = config.hidden_size
    self.word_embeddings = VocabParallelEmbedding(config.vocab_size,
                                                  config.hidden_size)
    self.position_embeddings = VocabParallelEmbedding(
        config.max_position_embeddings, config.hidden_size)
    self.token_type_embeddings = VocabParallelEmbedding(
        config.type_vocab_size, config.hidden_size)
    self.LayerNorm = nn.LayerNorm(config.hidden_size,
                                  eps=config.layer_norm_eps)
    self.position_ids = nn.Parameter(
        torch.empty((1, config.max_position_embeddings)), )

    self.position_embedding_type = config.position_embedding_type
    if self.position_embedding_type != "absolute":
        raise ValueError("Only 'absolute' position_embedding_type" +
                         " is supported")

forward 

forward(
    input_ids: Tensor,
    seq_lens: Tensor,
    position_ids: Tensor,
    token_type_ids: Optional[Tensor] = None,
) -> Tensor
Source code in vllm/model_executor/models/bert.py 
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
def forward(
    self,
    input_ids: torch.Tensor,
    seq_lens: torch.Tensor,
    position_ids: torch.Tensor,
    token_type_ids: Optional[torch.Tensor] = None,
) -> torch.Tensor:
    input_shape = input_ids.size()

    # Input embeddings.
    inputs_embeds = self.word_embeddings(input_ids)

    # Position embeddings.
    position_embeddings = self.position_embeddings(position_ids)

    if token_type_ids is None:
        token_type_ids = torch.zeros(input_shape,
                                     dtype=torch.long,
                                     device=inputs_embeds.device)

    token_type_embeddings = self.token_type_embeddings(token_type_ids)

    embeddings = inputs_embeds + token_type_embeddings + position_embeddings
    embeddings = self.LayerNorm(embeddings)
    return embeddings

BertEmbeddingModel

Bases: Module, SupportsV0Only, SupportsQuant

A model that uses Bert to provide embedding functionalities.

This class encapsulates the BertModel and provides an interface for embedding operations and customized pooling functions.

Attributes:

Name Type Description
model

An instance of BertModel used for forward operations.
_pooler

An instance of Pooler used for pooling operations.
Source code in vllm/model_executor/models/bert.py 
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
class BertEmbeddingModel(nn.Module, SupportsV0Only, SupportsQuant):
    """A model that uses Bert to provide embedding functionalities.

   This class encapsulates the BertModel and provides an interface for
   embedding operations and customized pooling functions.

   Attributes:
       model: An instance of BertModel used for forward operations.
       _pooler: An instance of Pooler used for pooling operations.
   """
    hf_to_vllm_mapper = WeightsMapper(orig_to_new_prefix={"model.": ""})

    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
        super().__init__()
        pooler_config = vllm_config.model_config.pooler_config
        self.model = self._build_model(vllm_config=vllm_config,
                                       prefix=maybe_prefix(prefix, "model"))
        self._pooler = self._build_pooler(pooler_config)

    def forward(
        self,
        input_ids: Optional[torch.Tensor],
        positions: torch.Tensor,
        intermediate_tensors: Optional[IntermediateTensors] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
    ) -> torch.Tensor:
        return self.model(input_ids=input_ids,
                          position_ids=positions,
                          inputs_embeds=inputs_embeds,
                          intermediate_tensors=intermediate_tensors)

    def pooler(
        self,
        hidden_states: torch.Tensor,
        pooling_metadata: PoolingMetadata,
    ) -> Optional[PoolerOutput]:
        return self._pooler(hidden_states, pooling_metadata)

    def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]):
        weights = self.hf_to_vllm_mapper.apply(weights)
        weights = ((name, data) for name, data in weights
                   if not name.startswith("lm_head."))
        self.model.load_weights(weights)

    def _build_model(self,
                     vllm_config: VllmConfig,
                     prefix: str = "") -> BertModel:
        return BertModel(vllm_config=vllm_config,
                         prefix=prefix,
                         embedding_class=BertEmbedding)

    def _build_pooler(self, pooler_config: PoolerConfig) -> Pooler:
        return Pooler.from_config_with_defaults(pooler_config,
                                                pooling_type=PoolingType.CLS,
                                                normalize=True,
                                                softmax=False)

_pooler instance-attribute 

_pooler = _build_pooler(pooler_config)

hf_to_vllm_mapper class-attribute instance-attribute 

hf_to_vllm_mapper = WeightsMapper(
    orig_to_new_prefix={"model.": ""}
)

model instance-attribute 

model = _build_model(
    vllm_config=vllm_config,
    prefix=maybe_prefix(prefix, "model"),
)

__init__ 

__init__(*, vllm_config: VllmConfig, prefix: str = '')
Source code in vllm/model_executor/models/bert.py 
403
404
405
406
407
408
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
    super().__init__()
    pooler_config = vllm_config.model_config.pooler_config
    self.model = self._build_model(vllm_config=vllm_config,
                                   prefix=maybe_prefix(prefix, "model"))
    self._pooler = self._build_pooler(pooler_config)

_build_model 

_build_model(
    vllm_config: VllmConfig, prefix: str = ""
) -> BertModel
Source code in vllm/model_executor/models/bert.py 
435
436
437
438
439
440
def _build_model(self,
                 vllm_config: VllmConfig,
                 prefix: str = "") -> BertModel:
    return BertModel(vllm_config=vllm_config,
                     prefix=prefix,
                     embedding_class=BertEmbedding)

_build_pooler 

_build_pooler(pooler_config: PoolerConfig) -> Pooler
Source code in vllm/model_executor/models/bert.py 
442
443
444
445
446
def _build_pooler(self, pooler_config: PoolerConfig) -> Pooler:
    return Pooler.from_config_with_defaults(pooler_config,
                                            pooling_type=PoolingType.CLS,
                                            normalize=True,
                                            softmax=False)

forward 

forward(
    input_ids: Optional[Tensor],
    positions: Tensor,
    intermediate_tensors: Optional[
        IntermediateTensors
    ] = None,
    inputs_embeds: Optional[Tensor] = None,
) -> Tensor
Source code in vllm/model_executor/models/bert.py 
410
411
412
413
414
415
416
417
418
419
420
def forward(
    self,
    input_ids: Optional[torch.Tensor],
    positions: torch.Tensor,
    intermediate_tensors: Optional[IntermediateTensors] = None,
    inputs_embeds: Optional[torch.Tensor] = None,
) -> torch.Tensor:
    return self.model(input_ids=input_ids,
                      position_ids=positions,
                      inputs_embeds=inputs_embeds,
                      intermediate_tensors=intermediate_tensors)

load_weights 

load_weights(weights: Iterable[tuple[str, Tensor]])
Source code in vllm/model_executor/models/bert.py 
429
430
431
432
433
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]):
    weights = self.hf_to_vllm_mapper.apply(weights)
    weights = ((name, data) for name, data in weights
               if not name.startswith("lm_head."))
    self.model.load_weights(weights)

pooler 

pooler(
    hidden_states: Tensor, pooling_metadata: PoolingMetadata
) -> Optional[PoolerOutput]
Source code in vllm/model_executor/models/bert.py 
422
423
424
425
426
427
def pooler(
    self,
    hidden_states: torch.Tensor,
    pooling_metadata: PoolingMetadata,
) -> Optional[PoolerOutput]:
    return self._pooler(hidden_states, pooling_metadata)

BertEncoder

Bases: Module

Source code in vllm/model_executor/models/bert.py 
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
@support_torch_compile
class BertEncoder(nn.Module):

    def __init__(self, vllm_config: VllmConfig, prefix: str = ""):
        super().__init__()
        config = vllm_config.model_config.hf_config
        cache_config = vllm_config.cache_config
        quant_config = vllm_config.quant_config
        self.layer = nn.ModuleList([
            BertLayer(config=config,
                      cache_config=cache_config,
                      quant_config=quant_config,
                      prefix=f"{prefix}.layer.{layer_idx}")
            for layer_idx in range(config.num_hidden_layers)
        ])

    def forward(
        self,
        hidden_states: torch.Tensor,
    ) -> torch.Tensor:
        for layer in self.layer:
            hidden_states = layer(hidden_states)
        return hidden_states

layer instance-attribute 

layer = ModuleList(
    [
        BertLayer(
            config=config,
            cache_config=cache_config,
            quant_config=quant_config,
            prefix=f"{prefix}.layer.{layer_idx}",
        )
        for layer_idx in range(num_hidden_layers)
    ]
)

__init__ 

__init__(vllm_config: VllmConfig, prefix: str = '')
Source code in vllm/model_executor/models/bert.py 
103
104
105
106
107
108
109
110
111
112
113
114
def __init__(self, vllm_config: VllmConfig, prefix: str = ""):
    super().__init__()
    config = vllm_config.model_config.hf_config
    cache_config = vllm_config.cache_config
    quant_config = vllm_config.quant_config
    self.layer = nn.ModuleList([
        BertLayer(config=config,
                  cache_config=cache_config,
                  quant_config=quant_config,
                  prefix=f"{prefix}.layer.{layer_idx}")
        for layer_idx in range(config.num_hidden_layers)
    ])

forward 

forward(hidden_states: Tensor) -> Tensor
Source code in vllm/model_executor/models/bert.py 
116
117
118
119
120
121
122
def forward(
    self,
    hidden_states: torch.Tensor,
) -> torch.Tensor:
    for layer in self.layer:
        hidden_states = layer(hidden_states)
    return hidden_states

BertForSequenceClassification

Bases: Module, SupportsV0Only, SupportsCrossEncoding, SupportsQuant

A model that uses Bert to provide embedding functionalities.

This class encapsulates the BertModel and provides an interface for embedding operations and customized pooling functions.

Attributes:

Name Type Description
model

An instance of BertModel used for forward operations.
_pooler

An instance of Pooler used for pooling operations.
Source code in vllm/model_executor/models/bert.py 
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
class BertForSequenceClassification(nn.Module, SupportsV0Only,
                                    SupportsCrossEncoding, SupportsQuant):
    """A model that uses Bert to provide embedding functionalities.

   This class encapsulates the BertModel and provides an interface for
   embedding operations and customized pooling functions.

   Attributes:
       model: An instance of BertModel used for forward operations.
       _pooler: An instance of Pooler used for pooling operations.
   """

    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
        super().__init__()
        config = vllm_config.model_config.hf_config

        self.default_activation_function = \
            get_cross_encoder_activation_function(config)

        self.num_labels = config.num_labels
        self.bert = BertModel(vllm_config=vllm_config,
                              prefix=maybe_prefix(prefix, "bert"),
                              embedding_class=BertEmbedding,
                              add_pooling_layer=True)
        self.classifier = nn.Linear(config.hidden_size, config.num_labels)
        self._pooler = ClassifierPooler(vllm_config.model_config,
                                        self.classifier, self.bert.pooler)

    def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]):

        self_weights = []

        def weight_filter():
            for name, weight in weights:
                if name.startswith("bert."):
                    yield (name[len("bert."):], weight)
                else:
                    self_weights.append((name, weight))

        self.bert.load_weights(weight_filter())

        params_dict = dict(self.named_parameters())

        for name, loaded_weight in self_weights:
            if name.startswith("classifier"):
                param = params_dict[name]
                weight_loader = getattr(param, "weight_loader",
                                        default_weight_loader)
                weight_loader(param, loaded_weight)

    def pooler(
        self,
        hidden_states: torch.Tensor,
        pooling_metadata: PoolingMetadata,
    ) -> Optional[PoolerOutput]:
        return self._pooler(hidden_states, pooling_metadata)

    def forward(
        self,
        input_ids: Optional[torch.Tensor],
        positions: torch.Tensor,
        intermediate_tensors: Optional[IntermediateTensors] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        token_type_ids: Optional[torch.Tensor] = None,
    ) -> torch.Tensor:
        return self.bert(input_ids=input_ids,
                         position_ids=positions,
                         inputs_embeds=inputs_embeds,
                         intermediate_tensors=intermediate_tensors,
                         token_type_ids=token_type_ids)

_pooler instance-attribute 

_pooler = ClassifierPooler(model_config, classifier, pooler)

bert instance-attribute 

bert = BertModel(
    vllm_config=vllm_config,
    prefix=maybe_prefix(prefix, "bert"),
    embedding_class=BertEmbedding,
    add_pooling_layer=True,
)

classifier instance-attribute 

classifier = Linear(hidden_size, num_labels)

default_activation_function instance-attribute 

default_activation_function = (
    get_cross_encoder_activation_function(config)
)

num_labels instance-attribute 

num_labels = num_labels

__init__ 

__init__(*, vllm_config: VllmConfig, prefix: str = '')
Source code in vllm/model_executor/models/bert.py 
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
    super().__init__()
    config = vllm_config.model_config.hf_config

    self.default_activation_function = \
        get_cross_encoder_activation_function(config)

    self.num_labels = config.num_labels
    self.bert = BertModel(vllm_config=vllm_config,
                          prefix=maybe_prefix(prefix, "bert"),
                          embedding_class=BertEmbedding,
                          add_pooling_layer=True)
    self.classifier = nn.Linear(config.hidden_size, config.num_labels)
    self._pooler = ClassifierPooler(vllm_config.model_config,
                                    self.classifier, self.bert.pooler)

forward 

forward(
    input_ids: Optional[Tensor],
    positions: Tensor,
    intermediate_tensors: Optional[
        IntermediateTensors
    ] = None,
    inputs_embeds: Optional[Tensor] = None,
    token_type_ids: Optional[Tensor] = None,
) -> Tensor
Source code in vllm/model_executor/models/bert.py 
506
507
508
509
510
511
512
513
514
515
516
517
518
def forward(
    self,
    input_ids: Optional[torch.Tensor],
    positions: torch.Tensor,
    intermediate_tensors: Optional[IntermediateTensors] = None,
    inputs_embeds: Optional[torch.Tensor] = None,
    token_type_ids: Optional[torch.Tensor] = None,
) -> torch.Tensor:
    return self.bert(input_ids=input_ids,
                     position_ids=positions,
                     inputs_embeds=inputs_embeds,
                     intermediate_tensors=intermediate_tensors,
                     token_type_ids=token_type_ids)

load_weights 

load_weights(weights: Iterable[tuple[str, Tensor]])
Source code in vllm/model_executor/models/bert.py 
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]):

    self_weights = []

    def weight_filter():
        for name, weight in weights:
            if name.startswith("bert."):
                yield (name[len("bert."):], weight)
            else:
                self_weights.append((name, weight))

    self.bert.load_weights(weight_filter())

    params_dict = dict(self.named_parameters())

    for name, loaded_weight in self_weights:
        if name.startswith("classifier"):
            param = params_dict[name]
            weight_loader = getattr(param, "weight_loader",
                                    default_weight_loader)
            weight_loader(param, loaded_weight)

pooler 

pooler(
    hidden_states: Tensor, pooling_metadata: PoolingMetadata
) -> Optional[PoolerOutput]
Source code in vllm/model_executor/models/bert.py 
499
500
501
502
503
504
def pooler(
    self,
    hidden_states: torch.Tensor,
    pooling_metadata: PoolingMetadata,
) -> Optional[PoolerOutput]:
    return self._pooler(hidden_states, pooling_metadata)

BertIntermediate

Bases: Module

Source code in vllm/model_executor/models/bert.py 
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
class BertIntermediate(nn.Module):

    def __init__(self,
                 hidden_size: int,
                 intermediate_size: int,
                 hidden_act: str,
                 quant_config: Optional[QuantizationConfig] = None,
                 prefix: str = ""):
        super().__init__()
        self.dense = ColumnParallelLinear(input_size=hidden_size,
                                          output_size=intermediate_size,
                                          bias=True,
                                          quant_config=quant_config,
                                          prefix=f"{prefix}.dense")
        self.intermediate_act_fn = get_act_fn(hidden_act)

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        hidden_states, _ = self.dense(hidden_states)
        hidden_states = self.intermediate_act_fn(hidden_states)
        return hidden_states

dense instance-attribute 

dense = ColumnParallelLinear(
    input_size=hidden_size,
    output_size=intermediate_size,
    bias=True,
    quant_config=quant_config,
    prefix=f"{prefix}.dense",
)

intermediate_act_fn instance-attribute 

intermediate_act_fn = get_act_fn(hidden_act)

__init__ 

__init__(
    hidden_size: int,
    intermediate_size: int,
    hidden_act: str,
    quant_config: Optional[QuantizationConfig] = None,
    prefix: str = "",
)
Source code in vllm/model_executor/models/bert.py 
273
274
275
276
277
278
279
280
281
282
283
284
285
def __init__(self,
             hidden_size: int,
             intermediate_size: int,
             hidden_act: str,
             quant_config: Optional[QuantizationConfig] = None,
             prefix: str = ""):
    super().__init__()
    self.dense = ColumnParallelLinear(input_size=hidden_size,
                                      output_size=intermediate_size,
                                      bias=True,
                                      quant_config=quant_config,
                                      prefix=f"{prefix}.dense")
    self.intermediate_act_fn = get_act_fn(hidden_act)

forward 

forward(hidden_states: Tensor) -> Tensor
Source code in vllm/model_executor/models/bert.py 
287
288
289
290
def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
    hidden_states, _ = self.dense(hidden_states)
    hidden_states = self.intermediate_act_fn(hidden_states)
    return hidden_states

BertLayer

Bases: Module

Source code in vllm/model_executor/models/bert.py 
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
class BertLayer(nn.Module):

    def __init__(self,
                 config: BertConfig,
                 cache_config: Optional[CacheConfig] = None,
                 quant_config: Optional[QuantizationConfig] = None,
                 prefix: str = ""):
        super().__init__()

        self.attention = BertAttention(
            hidden_size=config.hidden_size,
            num_attention_heads=config.num_attention_heads,
            layer_norm_eps=config.layer_norm_eps,
            cache_config=cache_config,
            quant_config=quant_config,
            prefix=f"{prefix}.attention")

        self.intermediate = BertIntermediate(
            hidden_size=config.hidden_size,
            intermediate_size=config.intermediate_size,
            hidden_act=config.hidden_act,
            quant_config=quant_config,
            prefix=f"{prefix}.intermediate")

        self.output = BertOutput(hidden_size=config.hidden_size,
                                 intermediate_size=config.intermediate_size,
                                 layer_norm_eps=config.layer_norm_eps,
                                 quant_config=quant_config,
                                 prefix=f"{prefix}.output")

    def forward(self, hidden_states: torch.Tensor):
        attn_output = self.attention(hidden_states)
        intermediate_output = self.intermediate(attn_output)
        output = self.output(intermediate_output, attn_output)
        return output

attention instance-attribute 

attention = BertAttention(
    hidden_size=hidden_size,
    num_attention_heads=num_attention_heads,
    layer_norm_eps=layer_norm_eps,
    cache_config=cache_config,
    quant_config=quant_config,
    prefix=f"{prefix}.attention",
)

intermediate instance-attribute 

intermediate = BertIntermediate(
    hidden_size=hidden_size,
    intermediate_size=intermediate_size,
    hidden_act=hidden_act,
    quant_config=quant_config,
    prefix=f"{prefix}.intermediate",
)

output instance-attribute 

output = BertOutput(
    hidden_size=hidden_size,
    intermediate_size=intermediate_size,
    layer_norm_eps=layer_norm_eps,
    quant_config=quant_config,
    prefix=f"{prefix}.output",
)

__init__ 

__init__(
    config: BertConfig,
    cache_config: Optional[CacheConfig] = None,
    quant_config: Optional[QuantizationConfig] = None,
    prefix: str = "",
)
Source code in vllm/model_executor/models/bert.py 
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
def __init__(self,
             config: BertConfig,
             cache_config: Optional[CacheConfig] = None,
             quant_config: Optional[QuantizationConfig] = None,
             prefix: str = ""):
    super().__init__()

    self.attention = BertAttention(
        hidden_size=config.hidden_size,
        num_attention_heads=config.num_attention_heads,
        layer_norm_eps=config.layer_norm_eps,
        cache_config=cache_config,
        quant_config=quant_config,
        prefix=f"{prefix}.attention")

    self.intermediate = BertIntermediate(
        hidden_size=config.hidden_size,
        intermediate_size=config.intermediate_size,
        hidden_act=config.hidden_act,
        quant_config=quant_config,
        prefix=f"{prefix}.intermediate")

    self.output = BertOutput(hidden_size=config.hidden_size,
                             intermediate_size=config.intermediate_size,
                             layer_norm_eps=config.layer_norm_eps,
                             quant_config=quant_config,
                             prefix=f"{prefix}.output")

forward 

forward(hidden_states: Tensor)
Source code in vllm/model_executor/models/bert.py 
155
156
157
158
159
def forward(self, hidden_states: torch.Tensor):
    attn_output = self.attention(hidden_states)
    intermediate_output = self.intermediate(attn_output)
    output = self.output(intermediate_output, attn_output)
    return output

BertModel

Bases: Module, SupportsQuant

Source code in vllm/model_executor/models/bert.py 
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
class BertModel(nn.Module, SupportsQuant):
    packed_modules_mapping = {"qkv_proj": ["query", "key", "value"]}

    def __init__(self,
                 *,
                 vllm_config: VllmConfig,
                 prefix: str = "",
                 embedding_class: type = BertEmbedding,
                 add_pooling_layer: bool = False):
        super().__init__()
        config = vllm_config.model_config.hf_config
        self.embeddings = embedding_class(config)
        self.encoder = BertEncoder(vllm_config=vllm_config,
                                   prefix=f"{prefix}.encoder")
        self.pooler = BertPooler(config) if add_pooling_layer else None

    def forward(
        self,
        input_ids: torch.Tensor,
        position_ids: torch.Tensor,
        intermediate_tensors: Optional[IntermediateTensors] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        token_type_ids: Optional[torch.Tensor] = None,
    ) -> torch.Tensor:
        if inputs_embeds is not None:
            hidden_states = inputs_embeds
        else:
            attn_metadata = get_forward_context().attn_metadata
            assert hasattr(attn_metadata, "seq_lens_tensor")
            hidden_states = self.embeddings(
                input_ids=input_ids,
                seq_lens=attn_metadata.seq_lens_tensor,
                position_ids=position_ids,
                token_type_ids=token_type_ids)
        return self.encoder(hidden_states)

    def load_weights(self, weights: Iterable[tuple[str,
                                                   torch.Tensor]]) -> set[str]:
        stacked_params_mapping = [
            # (param_name, shard_name, shard_id)
            ("qkv_proj", "query", "q"),
            ("qkv_proj", "key", "k"),
            ("qkv_proj", "value", "v"),
        ]

        params_dict = dict(self.named_parameters())
        loaded_params: set[str] = set()
        for name, loaded_weight in weights:
            if self.pooler is None and "pooler" in name:
                continue
            for (param_name, weight_name, shard_id) in stacked_params_mapping:
                if weight_name not in name:
                    continue
                name = name.replace(weight_name, param_name)
                # Skip loading extra bias for GPTQ models.
                if name.endswith(".bias") and name not in params_dict:
                    continue
                param = params_dict[name]
                weight_loader = param.weight_loader
                weight_loader(param, loaded_weight, shard_id)
                break
            else:
                # Skip loading extra bias for GPTQ models.
                if name.endswith(".bias") and name not in params_dict:
                    continue
                param = params_dict[name]
                weight_loader = getattr(param, "weight_loader",
                                        default_weight_loader)
                weight_loader(param, loaded_weight)
            loaded_params.add(name)
        return loaded_params

embeddings instance-attribute 

embeddings = embedding_class(config)

encoder instance-attribute 

encoder = BertEncoder(
    vllm_config=vllm_config, prefix=f"{prefix}.encoder"
)

packed_modules_mapping class-attribute instance-attribute 

packed_modules_mapping = {
    "qkv_proj": ["query", "key", "value"]
}

pooler instance-attribute 

pooler = BertPooler(config) if add_pooling_layer else None

__init__ 

__init__(
    *,
    vllm_config: VllmConfig,
    prefix: str = "",
    embedding_class: type = BertEmbedding,
    add_pooling_layer: bool = False,
)
Source code in vllm/model_executor/models/bert.py 
321
322
323
324
325
326
327
328
329
330
331
332
def __init__(self,
             *,
             vllm_config: VllmConfig,
             prefix: str = "",
             embedding_class: type = BertEmbedding,
             add_pooling_layer: bool = False):
    super().__init__()
    config = vllm_config.model_config.hf_config
    self.embeddings = embedding_class(config)
    self.encoder = BertEncoder(vllm_config=vllm_config,
                               prefix=f"{prefix}.encoder")
    self.pooler = BertPooler(config) if add_pooling_layer else None

forward 

forward(
    input_ids: Tensor,
    position_ids: Tensor,
    intermediate_tensors: Optional[
        IntermediateTensors
    ] = None,
    inputs_embeds: Optional[Tensor] = None,
    token_type_ids: Optional[Tensor] = None,
) -> Tensor
Source code in vllm/model_executor/models/bert.py 
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
def forward(
    self,
    input_ids: torch.Tensor,
    position_ids: torch.Tensor,
    intermediate_tensors: Optional[IntermediateTensors] = None,
    inputs_embeds: Optional[torch.Tensor] = None,
    token_type_ids: Optional[torch.Tensor] = None,
) -> torch.Tensor:
    if inputs_embeds is not None:
        hidden_states = inputs_embeds
    else:
        attn_metadata = get_forward_context().attn_metadata
        assert hasattr(attn_metadata, "seq_lens_tensor")
        hidden_states = self.embeddings(
            input_ids=input_ids,
            seq_lens=attn_metadata.seq_lens_tensor,
            position_ids=position_ids,
            token_type_ids=token_type_ids)
    return self.encoder(hidden_states)

load_weights 

load_weights(
    weights: Iterable[tuple[str, Tensor]],
) -> set[str]
Source code in vllm/model_executor/models/bert.py 
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
def load_weights(self, weights: Iterable[tuple[str,
                                               torch.Tensor]]) -> set[str]:
    stacked_params_mapping = [
        # (param_name, shard_name, shard_id)
        ("qkv_proj", "query", "q"),
        ("qkv_proj", "key", "k"),
        ("qkv_proj", "value", "v"),
    ]

    params_dict = dict(self.named_parameters())
    loaded_params: set[str] = set()
    for name, loaded_weight in weights:
        if self.pooler is None and "pooler" in name:
            continue
        for (param_name, weight_name, shard_id) in stacked_params_mapping:
            if weight_name not in name:
                continue
            name = name.replace(weight_name, param_name)
            # Skip loading extra bias for GPTQ models.
            if name.endswith(".bias") and name not in params_dict:
                continue
            param = params_dict[name]
            weight_loader = param.weight_loader
            weight_loader(param, loaded_weight, shard_id)
            break
        else:
            # Skip loading extra bias for GPTQ models.
            if name.endswith(".bias") and name not in params_dict:
                continue
            param = params_dict[name]
            weight_loader = getattr(param, "weight_loader",
                                    default_weight_loader)
            weight_loader(param, loaded_weight)
        loaded_params.add(name)
    return loaded_params

BertOutput

Bases: Module

Source code in vllm/model_executor/models/bert.py 
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
class BertOutput(nn.Module):

    def __init__(self,
                 hidden_size: int,
                 intermediate_size: int,
                 layer_norm_eps: float,
                 quant_config: Optional[QuantizationConfig] = None,
                 prefix: str = ""):
        super().__init__()

        self.dense = RowParallelLinear(input_size=intermediate_size,
                                       output_size=hidden_size,
                                       bias=True,
                                       quant_config=quant_config,
                                       prefix=f"{prefix}.dense")

        self.LayerNorm = nn.LayerNorm(hidden_size, eps=layer_norm_eps)

    def forward(self, hidden_states: torch.Tensor,
                input_tensor: torch.Tensor) -> torch.Tensor:
        hidden_states, _ = self.dense(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states

LayerNorm instance-attribute 

LayerNorm = LayerNorm(hidden_size, eps=layer_norm_eps)

dense instance-attribute 

dense = RowParallelLinear(
    input_size=intermediate_size,
    output_size=hidden_size,
    bias=True,
    quant_config=quant_config,
    prefix=f"{prefix}.dense",
)

__init__ 

__init__(
    hidden_size: int,
    intermediate_size: int,
    layer_norm_eps: float,
    quant_config: Optional[QuantizationConfig] = None,
    prefix: str = "",
)
Source code in vllm/model_executor/models/bert.py 
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
def __init__(self,
             hidden_size: int,
             intermediate_size: int,
             layer_norm_eps: float,
             quant_config: Optional[QuantizationConfig] = None,
             prefix: str = ""):
    super().__init__()

    self.dense = RowParallelLinear(input_size=intermediate_size,
                                   output_size=hidden_size,
                                   bias=True,
                                   quant_config=quant_config,
                                   prefix=f"{prefix}.dense")

    self.LayerNorm = nn.LayerNorm(hidden_size, eps=layer_norm_eps)

forward 

forward(
    hidden_states: Tensor, input_tensor: Tensor
) -> Tensor
Source code in vllm/model_executor/models/bert.py 
311
312
313
314
315
def forward(self, hidden_states: torch.Tensor,
            input_tensor: torch.Tensor) -> torch.Tensor:
    hidden_states, _ = self.dense(hidden_states)
    hidden_states = self.LayerNorm(hidden_states + input_tensor)
    return hidden_states

BertPooler

Bases: Module

Source code in vllm/model_executor/models/bert.py 
84
85
86
87
88
89
90
91
92
93
94
95
96
97
class BertPooler(nn.Module):

    def __init__(self, config: BertConfig):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.activation = nn.Tanh()

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        # We "pool" the model by simply taking the hidden state corresponding
        # to the first token.
        first_token_tensor = hidden_states[0, :]
        pooled_output = self.dense(first_token_tensor)
        pooled_output = self.activation(pooled_output)
        return pooled_output

activation instance-attribute 

activation = Tanh()

dense instance-attribute 

dense = Linear(hidden_size, hidden_size)

__init__ 

__init__(config: BertConfig)
Source code in vllm/model_executor/models/bert.py 
86
87
88
89
def __init__(self, config: BertConfig):
    super().__init__()
    self.dense = nn.Linear(config.hidden_size, config.hidden_size)
    self.activation = nn.Tanh()

forward 

forward(hidden_states: Tensor) -> Tensor
Source code in vllm/model_executor/models/bert.py 
91
92
93
94
95
96
97
def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
    # We "pool" the model by simply taking the hidden state corresponding
    # to the first token.
    first_token_tensor = hidden_states[0, :]
    pooled_output = self.dense(first_token_tensor)
    pooled_output = self.activation(pooled_output)
    return pooled_output

BertSelfAttention

Bases: Module

Source code in vllm/model_executor/models/bert.py 
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
class BertSelfAttention(nn.Module):

    def __init__(
        self,
        hidden_size: int,
        num_attention_heads: int,
        cache_config: Optional[CacheConfig] = None,
        quant_config: Optional[QuantizationConfig] = None,
        prefix: str = "",
    ):
        super().__init__()
        self.hidden_size = hidden_size
        tp_size = get_tensor_model_parallel_world_size()

        self.total_num_heads = num_attention_heads
        assert self.total_num_heads % tp_size == 0

        self.num_heads = self.total_num_heads // tp_size
        self.total_num_kv_heads = self.total_num_heads
        self.head_dim = self.hidden_size // self.total_num_heads
        assert self.head_dim * self.total_num_heads == self.hidden_size

        self.num_kv_heads = max(1, self.total_num_kv_heads // tp_size)

        self.q_size = self.num_heads * self.head_dim
        self.kv_size = self.num_kv_heads * self.head_dim
        self.scaling = self.head_dim**-0.5
        self.qkv_proj = QKVParallelLinear(
            hidden_size=self.hidden_size,
            head_size=self.head_dim,
            total_num_heads=self.total_num_heads,
            total_num_kv_heads=self.total_num_kv_heads,
            bias=True,
            quant_config=quant_config,
            prefix=f"{prefix}.qkv_proj")

        self.attn = Attention(num_heads=self.num_heads,
                              head_size=self.head_dim,
                              scale=self.scaling,
                              num_kv_heads=self.num_kv_heads,
                              cache_config=cache_config,
                              quant_config=quant_config,
                              prefix=f"{prefix}.attn",
                              attn_type=AttentionType.ENCODER_ONLY)

    def forward(
        self,
        hidden_states: torch.Tensor,
    ) -> torch.Tensor:
        qkv, _ = self.qkv_proj(hidden_states)
        q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
        output = self.attn(q, k, v)
        return output

attn instance-attribute 

attn = Attention(
    num_heads=num_heads,
    head_size=head_dim,
    scale=scaling,
    num_kv_heads=num_kv_heads,
    cache_config=cache_config,
    quant_config=quant_config,
    prefix=f"{prefix}.attn",
    attn_type=ENCODER_ONLY,
)

head_dim instance-attribute 

head_dim = hidden_size // total_num_heads

hidden_size instance-attribute 

hidden_size = hidden_size

kv_size instance-attribute 

kv_size = num_kv_heads * head_dim

num_heads instance-attribute 

num_heads = total_num_heads // tp_size

num_kv_heads instance-attribute 

num_kv_heads = max(1, total_num_kv_heads // tp_size)

q_size instance-attribute 

q_size = num_heads * head_dim

qkv_proj instance-attribute 

qkv_proj = QKVParallelLinear(
    hidden_size=hidden_size,
    head_size=head_dim,
    total_num_heads=total_num_heads,
    total_num_kv_heads=total_num_kv_heads,
    bias=True,
    quant_config=quant_config,
    prefix=f"{prefix}.qkv_proj",
)

scaling instance-attribute 

scaling = head_dim ** -0.5

total_num_heads instance-attribute 

total_num_heads = num_attention_heads

total_num_kv_heads instance-attribute 

total_num_kv_heads = total_num_heads

__init__ 

__init__(
    hidden_size: int,
    num_attention_heads: int,
    cache_config: Optional[CacheConfig] = None,
    quant_config: Optional[QuantizationConfig] = None,
    prefix: str = "",
)
Source code in vllm/model_executor/models/bert.py 
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
def __init__(
    self,
    hidden_size: int,
    num_attention_heads: int,
    cache_config: Optional[CacheConfig] = None,
    quant_config: Optional[QuantizationConfig] = None,
    prefix: str = "",
):
    super().__init__()
    self.hidden_size = hidden_size
    tp_size = get_tensor_model_parallel_world_size()

    self.total_num_heads = num_attention_heads
    assert self.total_num_heads % tp_size == 0

    self.num_heads = self.total_num_heads // tp_size
    self.total_num_kv_heads = self.total_num_heads
    self.head_dim = self.hidden_size // self.total_num_heads
    assert self.head_dim * self.total_num_heads == self.hidden_size

    self.num_kv_heads = max(1, self.total_num_kv_heads // tp_size)

    self.q_size = self.num_heads * self.head_dim
    self.kv_size = self.num_kv_heads * self.head_dim
    self.scaling = self.head_dim**-0.5
    self.qkv_proj = QKVParallelLinear(
        hidden_size=self.hidden_size,
        head_size=self.head_dim,
        total_num_heads=self.total_num_heads,
        total_num_kv_heads=self.total_num_kv_heads,
        bias=True,
        quant_config=quant_config,
        prefix=f"{prefix}.qkv_proj")

    self.attn = Attention(num_heads=self.num_heads,
                          head_size=self.head_dim,
                          scale=self.scaling,
                          num_kv_heads=self.num_kv_heads,
                          cache_config=cache_config,
                          quant_config=quant_config,
                          prefix=f"{prefix}.attn",
                          attn_type=AttentionType.ENCODER_ONLY)

forward 

forward(hidden_states: Tensor) -> Tensor
Source code in vllm/model_executor/models/bert.py 
239
240
241
242
243
244
245
246
def forward(
    self,
    hidden_states: torch.Tensor,
) -> torch.Tensor:
    qkv, _ = self.qkv_proj(hidden_states)
    q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
    output = self.attn(q, k, v)
    return output

BertSelfOutput

Bases: Module

Source code in vllm/model_executor/models/bert.py 
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
class BertSelfOutput(nn.Module):

    def __init__(self,
                 hidden_size: int,
                 layer_norm_eps: float,
                 quant_config: Optional[QuantizationConfig] = None,
                 prefix: str = ""):
        super().__init__()
        self.dense = RowParallelLinear(input_size=hidden_size,
                                       output_size=hidden_size,
                                       bias=True,
                                       quant_config=quant_config,
                                       prefix=f"{prefix}.dense")
        self.LayerNorm = nn.LayerNorm(hidden_size, eps=layer_norm_eps)

    def forward(self, hidden_states: torch.Tensor,
                input_tensor: torch.Tensor) -> torch.Tensor:
        hidden_states, _ = self.dense(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states

LayerNorm instance-attribute 

LayerNorm = LayerNorm(hidden_size, eps=layer_norm_eps)

dense instance-attribute 

dense = RowParallelLinear(
    input_size=hidden_size,
    output_size=hidden_size,
    bias=True,
    quant_config=quant_config,
    prefix=f"{prefix}.dense",
)

__init__ 

__init__(
    hidden_size: int,
    layer_norm_eps: float,
    quant_config: Optional[QuantizationConfig] = None,
    prefix: str = "",
)
Source code in vllm/model_executor/models/bert.py 
251
252
253
254
255
256
257
258
259
260
261
262
def __init__(self,
             hidden_size: int,
             layer_norm_eps: float,
             quant_config: Optional[QuantizationConfig] = None,
             prefix: str = ""):
    super().__init__()
    self.dense = RowParallelLinear(input_size=hidden_size,
                                   output_size=hidden_size,
                                   bias=True,
                                   quant_config=quant_config,
                                   prefix=f"{prefix}.dense")
    self.LayerNorm = nn.LayerNorm(hidden_size, eps=layer_norm_eps)

forward 

forward(
    hidden_states: Tensor, input_tensor: Tensor
) -> Tensor
Source code in vllm/model_executor/models/bert.py 
264
265
266
267
268
def forward(self, hidden_states: torch.Tensor,
            input_tensor: torch.Tensor) -> torch.Tensor:
    hidden_states, _ = self.dense(hidden_states)
    hidden_states = self.LayerNorm(hidden_states + input_tensor)
    return hidden_states