vllm.model_executor.models.bert ¶

class BertAttention(nn.Module):
    def __init__(
        self,
        hidden_size: int,
        num_attention_heads: int,
        layer_norm_eps: float,
        cache_config: CacheConfig | None = None,
        quant_config: QuantizationConfig | None = 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)

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.register_buffer(
            "position_ids",
            torch.arange(config.max_position_embeddings).unsqueeze(0),
        )
        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,
        position_ids: torch.Tensor,
        inputs_embeds: torch.Tensor | None = None,
    ) -> torch.Tensor:
        token_type_ids = _decode_token_type_ids(input_ids)

        if inputs_embeds is None:
            inputs_embeds = self.word_embeddings(input_ids)

        position_embeddings = self.position_embeddings(position_ids)

        token_type_embeddings = self.token_type_embeddings(token_type_ids)

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

@default_pooling_type("CLS")
class BertEmbeddingModel(nn.Module, 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.
    """

    is_pooling_model = True

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

        pooler_config = vllm_config.model_config.pooler_config
        assert pooler_config is not None

        self.model = self._build_model(
            vllm_config=vllm_config, prefix=maybe_prefix(prefix, "model")
        )
        self.pooler = self._build_pooler(pooler_config)

    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
        return self.model.get_input_embeddings(input_ids)

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

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

        has_model_prefix = any(name.startswith("model.") for name, _ in weights_list)
        if not has_model_prefix:
            mapper = WeightsMapper(orig_to_new_prefix={"": "model."})

        loader = AutoWeightsLoader(self, skip_prefixes=["lm_head."])
        return loader.load_weights(weights_list, mapper=mapper)

    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 DispatchPooler(
            {
                "encode": Pooler.for_encode(pooler_config),
                "embed": Pooler.for_embed(pooler_config),
            }
        )

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

@default_pooling_type("CLS")
class BertForSequenceClassification(nn.Module, 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.
    """

    is_pooling_model = True

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

        self.num_labels = config.num_labels
        self.bert = BertPoolingModel(
            vllm_config=vllm_config,
            prefix=maybe_prefix(prefix, "bert"),
            embedding_class=BertEmbedding,
        )
        self.classifier = nn.Linear(
            config.hidden_size,
            config.num_labels,
            dtype=vllm_config.model_config.head_dtype,
        )

        pooler_config = vllm_config.model_config.pooler_config
        assert pooler_config is not None

        self.pooler = DispatchPooler(
            {
                "encode": Pooler.for_encode(pooler_config),
                "classify": ClassifierPooler(
                    pooling=self.bert.pooler,
                    classifier=self.classifier,
                    act_fn=ClassifierPooler.act_fn_for_seq_cls(
                        vllm_config.model_config
                    ),
                ),
                "score": ClassifierPooler(
                    pooling=self.bert.pooler,
                    classifier=self.classifier,
                    act_fn=ClassifierPooler.act_fn_for_cross_encoder(
                        vllm_config.model_config
                    ),
                ),
            }
        )

    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
        return self.bert.get_input_embeddings(input_ids)

    def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]):
        loader = AutoWeightsLoader(self)
        loaded_params = loader.load_weights(weights)
        return loaded_params

    def forward(
        self,
        input_ids: torch.Tensor | None,
        positions: torch.Tensor,
        intermediate_tensors: IntermediateTensors | None = None,
        inputs_embeds: torch.Tensor | None = None,
        token_type_ids: torch.Tensor | None = None,
    ) -> torch.Tensor:
        if token_type_ids is not None:
            assert self.bert.config.vocab_size < (1 << TOKEN_TYPE_SHIFT)
            assert input_ids is not None
            _encode_token_type_ids(input_ids, token_type_ids)

        return self.bert(
            input_ids=input_ids,
            positions=positions,
            inputs_embeds=inputs_embeds,
            intermediate_tensors=intermediate_tensors,
        )

@default_pooling_type("ALL")
class BertForTokenClassification(nn.Module):
    is_pooling_model = True

    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
        super().__init__()
        config = vllm_config.model_config.hf_config
        self.head_dtype = vllm_config.model_config.head_dtype
        self.num_labels = config.num_labels
        self.bert = BertModel(
            vllm_config=vllm_config,
            prefix=maybe_prefix(prefix, "bert"),
            embedding_class=BertEmbedding,
        )
        self.classifier = nn.Linear(
            config.hidden_size, config.num_labels, dtype=self.head_dtype
        )

        pooler_config = vllm_config.model_config.pooler_config
        assert pooler_config is not None

        self.pooler = DispatchPooler(
            {
                "encode": Pooler.for_encode(pooler_config),
            }
        )

    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
        return self.bert.get_input_embeddings(input_ids)

    def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]):
        loader = AutoWeightsLoader(self)
        loaded_params = loader.load_weights(weights)
        return loaded_params

    def forward(
        self,
        input_ids: torch.Tensor | None,
        positions: torch.Tensor,
        intermediate_tensors: IntermediateTensors | None = None,
        inputs_embeds: torch.Tensor | None = None,
        token_type_ids: torch.Tensor | None = None,
    ) -> torch.Tensor:
        if token_type_ids is not None:
            assert self.bert.config.vocab_size < (1 << TOKEN_TYPE_SHIFT)
            assert input_ids is not None
            _encode_token_type_ids(input_ids, token_type_ids)

        hidden_states = self.bert(
            input_ids=input_ids,
            positions=positions,
            inputs_embeds=inputs_embeds,
            intermediate_tensors=intermediate_tensors,
        )

        hidden_states = hidden_states.to(self.head_dtype)
        return self.classifier(hidden_states)

class BertIntermediate(nn.Module):
    def __init__(
        self,
        hidden_size: int,
        intermediate_size: int,
        hidden_act: str,
        quant_config: QuantizationConfig | None = 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

class BertLayer(nn.Module):
    def __init__(
        self,
        config: BertConfig,
        cache_config: CacheConfig | None = None,
        quant_config: QuantizationConfig | None = 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