siglip.py

# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
"""Implementation of SiglipVisionModel intended to be only used
within a vision language model."""

import math
from collections.abc import Iterable

import torch
from torch import nn
from transformers import SiglipVisionConfig

from vllm.attention.layer import MultiHeadAttention
from vllm.distributed import divide, get_tensor_model_parallel_world_size
from vllm.model_executor.layers.activation import get_act_fn
from vllm.model_executor.layers.linear import (
    ColumnParallelLinear,
    QKVParallelLinear,
    RowParallelLinear,
)
from vllm.model_executor.layers.quantization import QuantizationConfig
from vllm.model_executor.layers.vocab_parallel_embedding import VocabParallelEmbedding
from vllm.model_executor.model_loader.weight_utils import (
    default_weight_loader,
    maybe_remap_kv_scale_name,
)

from .vision import (
    VisionEncoderInfo,
    VisionFeatureSelectStrategy,
    resolve_visual_encoder_outputs,
)


class SiglipEncoderInfo(VisionEncoderInfo[SiglipVisionConfig]):
    def get_num_image_tokens(
        self,
        *,
        image_width: int,
        image_height: int,
    ) -> int:
        return self.get_patch_grid_length() ** 2

    def get_image_size(self) -> int:
        return self.vision_config.image_size

    def get_patch_size(self) -> int:
        return self.vision_config.patch_size

    def get_patch_grid_length(self) -> int:
        image_size, patch_size = self.get_image_size(), self.get_patch_size()
        return image_size // patch_size


# Adapted from https://github.com/huggingface/transformers/blob/v4.43.3/src/transformers/models/siglip/modeling_siglip.py#L249 # noqa
class SiglipVisionEmbeddings(nn.Module):
    def __init__(self, config: SiglipVisionConfig):
        super().__init__()
        self.config = config
        self.embed_dim = config.hidden_size
        self.image_size = config.image_size
        self.patch_size = config.patch_size

        self.patch_embedding = nn.Conv2d(
            in_channels=config.num_channels,
            out_channels=self.embed_dim,
            kernel_size=self.patch_size,
            stride=self.patch_size,
            padding="valid",
        )

        self.num_patches = (self.image_size // self.patch_size) ** 2
        self.num_positions = self.num_patches
        self.position_embedding = VocabParallelEmbedding(
            self.num_positions, self.embed_dim
        )
        self.register_buffer(
            "position_ids",
            torch.arange(self.num_positions, dtype=torch.int64).expand((1, -1)),
            persistent=False,
        )

    def interpolate_pos_encoding(
        self, embeddings: torch.Tensor, height: int, width: int
    ) -> torch.Tensor:
        """
        This method is an adapted method for SigLIP (due to SigLIP not having
        class embedding unlike other ViTs) that allows the model to interpolate
        the pre-trained position encodings such that it can be usable on higher
        resolution images.

        Source:
        https://github.com/facebookresearch/dino/blob/de9ee3df6cf39fac952ab558447af1fa1365362a/vision_transformer.py#L174
        """
        position_embeddings = self.position_embedding.weight.unsqueeze(0)
        num_patches = embeddings.shape[1]
        num_positions = position_embeddings.shape[1]
        if num_patches == num_positions and height == width:
            return position_embeddings

        dim = embeddings.shape[-1]
        height = height // self.patch_size
        width = width // self.patch_size
        # we add a small number to avoid floating point error
        # in the interpolation
        # see discussion at https://github.com/facebookresearch/dino/issues/8
        height, width = height + 0.1, width + 0.1

        patch_pos_embed = position_embeddings.reshape(
            1, int(math.sqrt(num_positions)), int(math.sqrt(num_positions)), dim
        )
        patch_pos_embed = patch_pos_embed.permute(0, 3, 1, 2)
        patch_pos_embed = nn.functional.interpolate(
            patch_pos_embed,
            scale_factor=(
                height / math.sqrt(num_positions),
                width / math.sqrt(num_positions),
            ),
            mode="bicubic",
            align_corners=False,
        )
        if (
            int(height) != patch_pos_embed.shape[-2]
            or int(width) != patch_pos_embed.shape[-1]
        ):
            raise ValueError(
                "Width or height does not match with "
                "the interpolated position embeddings"
            )

        patch_pos_embed = patch_pos_embed.permute(0, 2, 3, 1).view(1, -1, dim)
        return patch_pos_embed

    def forward(
        self, pixel_values: torch.Tensor, interpolate_pos_encoding: bool = False
    ) -> torch.Tensor:
        _, _, height, width = pixel_values.shape
        target_dtype = self.patch_embedding.weight.dtype
        patch_embeds = self.patch_embedding(
            pixel_values.to(dtype=target_dtype)
        )  # shape = [*, width, grid, grid]
        embeddings = patch_embeds.flatten(2).transpose(1, 2)

        if interpolate_pos_encoding:
            embeddings += self.interpolate_pos_encoding(embeddings, height, width)
        else:
            embeddings += self.position_embedding(self.position_ids)
        return embeddings


class SiglipAttention(nn.Module):
    def __init__(
        self,
        config: SiglipVisionConfig,
        quant_config: QuantizationConfig | None = None,
        prefix: str = "",
    ) -> None:
        super().__init__()

        self.config = config
        self.embed_dim = config.hidden_size
        self.num_heads = config.num_attention_heads
        self.head_dim = self.embed_dim // self.num_heads
        if self.head_dim * self.num_heads != self.embed_dim:
            raise ValueError(
                f"embed_dim must be divisible by num_heads (got "
                "`embed_dim`: {self.embed_dim} and `num_heads`:"
                f" {self.num_heads})."
            )

        self.scale = self.head_dim**-0.5
        self.dropout = config.attention_dropout
        self.qkv_proj = QKVParallelLinear(
            hidden_size=self.embed_dim,
            head_size=self.head_dim,
            total_num_heads=self.num_heads,
            quant_config=quant_config,
            prefix=f"{prefix}.qkv_proj",
        )

        self.out_proj = RowParallelLinear(
            input_size=self.embed_dim,
            output_size=self.embed_dim,
            quant_config=quant_config,
            prefix=f"{prefix}.out_proj",
        )

        self.tp_size = get_tensor_model_parallel_world_size()
        self.num_heads_per_partition = divide(self.num_heads, self.tp_size)

        self.attn = MultiHeadAttention(
            self.num_heads_per_partition, self.head_dim, self.scale
        )

    def forward(
        self,
        hidden_states: torch.Tensor,
    ) -> torch.Tensor:
        """Input shape: Batch x Time x Channel"""
        qkv_states, _ = self.qkv_proj(hidden_states)
        query_states, key_states, value_states = qkv_states.chunk(3, dim=-1)

        out = self.attn(query_states, key_states, value_states)
        attn_output, _ = self.out_proj(out)

        return attn_output, None


class SiglipMLP(nn.Module):
    def __init__(
        self,
        config: SiglipVisionConfig,
        quant_config: QuantizationConfig | None = None,
        prefix: str = "",
    ) -> None:
        super().__init__()

        self.config = config
        self.activation_fn = get_act_fn(config.hidden_act)
        # Special handling for BNB and torchao quantization
        if quant_config and quant_config.get_name() in ["bitsandbytes", "torchao"]:
            quantizable = True
        else:
            # For other quantization, we require the hidden size to be a
            # multiple of 64
            quantizable = (
                config.hidden_size % 64 == 0 and config.intermediate_size % 64 == 0
            )
        self.fc1 = ColumnParallelLinear(
            config.hidden_size,
            config.intermediate_size,
            quant_config=quant_config if quantizable else None,
            prefix=f"{prefix}.fc1",
        )
        self.fc2 = RowParallelLinear(
            config.intermediate_size,
            config.hidden_size,
            quant_config=quant_config if quantizable else None,
            prefix=f"{prefix}.fc2",
        )

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        hidden_states, _ = self.fc1(hidden_states)
        hidden_states = self.activation_fn(hidden_states)
        hidden_states, _ = self.fc2(hidden_states)
        return hidden_states


class SiglipEncoderLayer(nn.Module):
    def __init__(
        self,
        config: SiglipVisionConfig,
        quant_config: QuantizationConfig | None = None,
        prefix: str = "",
    ) -> None:
        super().__init__()

        self.embed_dim = config.hidden_size

        self.self_attn = SiglipAttention(
            config,
            quant_config=quant_config,
            prefix=f"{prefix}.self_attn",
        )
        self.layer_norm1 = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_eps)
        self.mlp = SiglipMLP(
            config,
            quant_config=quant_config,
            prefix=f"{prefix}.mlp",
        )
        self.layer_norm2 = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_eps)

    def forward(
        self,
        hidden_states: torch.Tensor,
    ) -> tuple[torch.Tensor, None]:
        residual = hidden_states

        hidden_states = self.layer_norm1(hidden_states)
        hidden_states, _ = self.self_attn(hidden_states=hidden_states)
        hidden_states += residual

        residual = hidden_states
        hidden_states = self.layer_norm2(hidden_states)
        hidden_states = self.mlp(hidden_states)
        hidden_states += residual

        return hidden_states, None


class SiglipEncoder(nn.Module):
    def __init__(
        self,
        config: SiglipVisionConfig,
        quant_config: QuantizationConfig | None = None,
        num_hidden_layers_override: int | None = None,
        prefix: str = "",
    ) -> None:
        super().__init__()

        self.config = config

        if num_hidden_layers_override is None:
            num_hidden_layers = config.num_hidden_layers
        else:
            num_hidden_layers = num_hidden_layers_override

        self.layers = nn.ModuleList(
            [
                SiglipEncoderLayer(
                    config,
                    quant_config=quant_config,
                    prefix=f"{prefix}.layers.{layer_idx}",
                )
                for layer_idx in range(num_hidden_layers)
            ]
        )

    def forward(
        self,
        inputs_embeds: torch.Tensor,
        return_all_hidden_states: bool,
    ) -> torch.Tensor | list[torch.Tensor]:
        hidden_states_pool = [inputs_embeds]
        hidden_states = inputs_embeds

        for encoder_layer in self.layers:
            hidden_states, _ = encoder_layer(hidden_states)
            if return_all_hidden_states:
                hidden_states_pool.append(hidden_states)
        # If we have multiple feature sample layers, we return all hidden
        # states in order and grab the ones we need by index.
        if return_all_hidden_states:
            return hidden_states_pool
        return hidden_states


class SiglipMultiheadAttentionPoolingHead(nn.Module):
    """Multihead Attention Pooling."""

    def __init__(
        self,
        config: SiglipVisionConfig,
        quant_config: QuantizationConfig | None = None,
        prefix: str = "",
    ) -> None:
        super().__init__()

        self.probe = nn.Parameter(torch.randn(1, 1, config.hidden_size))
        # TODO(ChristopherCho): Implement vLLM version of MultiheadAttention
        self.attention = torch.nn.MultiheadAttention(
            config.hidden_size, config.num_attention_heads, batch_first=True
        )
        self.layernorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.mlp = SiglipMLP(
            config=config, quant_config=quant_config, prefix=f"{prefix}.mlp"
        )

    def forward(self, hidden_state: torch.Tensor) -> torch.Tensor:
        batch_size = hidden_state.shape[0]
        probe = self.probe.repeat(batch_size, 1, 1)

        hidden_state = self.attention(probe, hidden_state, hidden_state)[0]

        residual = hidden_state
        hidden_state = self.layernorm(hidden_state)
        hidden_state = self.mlp(hidden_state)
        hidden_state += residual

        return hidden_state[:, 0]


class SiglipVisionTransformer(nn.Module):
    def __init__(
        self,
        config: SiglipVisionConfig,
        quant_config: QuantizationConfig | None = None,
        *,
        num_hidden_layers_override: int | None = None,
        require_post_norm: bool | None = None,
        prefix: str = "",
    ) -> None:
        super().__init__()

        self.config = config
        embed_dim = config.hidden_size

        self.embeddings = SiglipVisionEmbeddings(config)

        self.encoder = SiglipEncoder(
            config,
            quant_config=quant_config,
            num_hidden_layers_override=num_hidden_layers_override,
            prefix=f"{prefix}.encoder",
        )

        num_hidden_layers = config.num_hidden_layers
        if len(self.encoder.layers) > config.num_hidden_layers:
            raise ValueError(
                f"The original encoder only has {num_hidden_layers} "
                f"layers, but you requested {len(self.encoder.layers)} layers."
            )

        # If possible, skip post_layernorm to conserve memory
        if require_post_norm is None:
            require_post_norm = len(self.encoder.layers) == num_hidden_layers

        if require_post_norm:
            self.post_layernorm = nn.LayerNorm(embed_dim, eps=config.layer_norm_eps)
        else:
            self.post_layernorm = None

        self.use_head = (
            True if not hasattr(config, "vision_use_head") else config.vision_use_head
        )
        if self.use_head:
            self.head = SiglipMultiheadAttentionPoolingHead(
                config=config,
                quant_config=quant_config,
                prefix=f"{prefix}.head",
            )

    def forward(
        self,
        pixel_values: torch.Tensor,
        *,
        interpolate_pos_encoding: bool = False,
        select_layers: list[int] | None = None,
        feature_select_strategy: VisionFeatureSelectStrategy | None = None,
    ) -> torch.Tensor:
        hidden_states = self.embeddings(
            pixel_values,
            interpolate_pos_encoding=interpolate_pos_encoding,
        )

        # Produces either the last layer output or all of the hidden states,
        # depending on if we have select_layers or not
        encoder_outputs = self.encoder(
            inputs_embeds=hidden_states,
            return_all_hidden_states=select_layers is not None,
        )

        # Handle post-norm (if applicable) and stacks feature layers if needed
        encoder_outputs = resolve_visual_encoder_outputs(
            encoder_outputs,
            self.post_layernorm,
            select_layers=select_layers,
            max_possible_layers=self.config.num_hidden_layers,
            feature_select_strategy=feature_select_strategy,
        )

        # TODO: add this back when pooled_output is used in inference.
        # if self.use_head:
        # pooled_output = self.head(encoder_outputs)

        return encoder_outputs


class SiglipVisionModel(nn.Module):
    config_class = SiglipVisionConfig
    main_input_name = "pixel_values"

    def __init__(
        self,
        config: SiglipVisionConfig,
        quant_config: QuantizationConfig | None = None,
        *,
        num_hidden_layers_override: int | None = None,
        require_post_norm: bool | None = None,
        prefix: str = "",
    ) -> None:
        super().__init__()

        self.vision_model = SiglipVisionTransformer(
            config,
            quant_config,
            num_hidden_layers_override=num_hidden_layers_override,
            require_post_norm=require_post_norm,
            prefix=f"{prefix}.vision_model",
        )

    def get_input_embeddings(self) -> nn.Module:
        return self.vision_model.embeddings.patch_embedding

    @property
    def dtype(self):
        return self.get_input_embeddings().weight.dtype

    def forward(
        self,
        pixel_values: torch.Tensor,
        interpolate_pos_encoding: bool = False,
        select_layers: list[int] | None = None,
        feature_select_strategy: VisionFeatureSelectStrategy | None = None,
    ) -> torch.Tensor:
        return self.vision_model(
            pixel_values=pixel_values,
            interpolate_pos_encoding=interpolate_pos_encoding,
            select_layers=select_layers,
            feature_select_strategy=feature_select_strategy,
        )

    def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
        stacked_params_mapping = [
            # (param_name, shard_name, shard_id)
            ("qkv_proj", "q_proj", "q"),
            ("qkv_proj", "k_proj", "k"),
            ("qkv_proj", "v_proj", "v"),
        ]
        params_dict = dict(self.named_parameters())
        loaded_params: set[str] = set()
        layer_count = len(self.vision_model.encoder.layers)

        for name, loaded_weight in weights:
            # post_layernorm is optional in SiglipVisionModel
            if (
                name.startswith("vision_model.post_layernorm")
                and self.vision_model.post_layernorm is None
            ):
                continue

            # omit layers when num_hidden_layers_override is set
            if name.startswith("vision_model.encoder.layers"):
                layer_idx = int(name.split(".")[3])
                if layer_idx >= layer_count:
                    continue

            # Check if this is a scale parameter that needs remapping first
            if name.endswith((".k_scale", ".v_scale", ".q_scale", ".prob_scale")):
                # Try to remap the scale name first
                remapped_name = maybe_remap_kv_scale_name(name, params_dict)
                if remapped_name is not None and remapped_name in params_dict:
                    # Successfully remapped, use the remapped name
                    param = params_dict[remapped_name]
                    weight_loader = getattr(
                        param, "weight_loader", default_weight_loader
                    )
                    weight_loader(param, loaded_weight)
                    loaded_params.add(remapped_name)
                    continue
                # If remapping failed, continue with normal processing

            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)

                param = params_dict[name]
                weight_loader = param.weight_loader
                weight_loader(param, loaded_weight, shard_id)
                break
            else:
                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