sglang/python/sglang/srt/models/qwen2_vl.py

# coding=utf-8
# Adapted from
# https://github.com/huggingface/transformers/blob/19e6e80e10118f855137b90740936c0b11ac397f/src/transformers/models/qwen2_vl/modeling_qwen2_vl.py
# Copyright 2024 The Qwen team.
# Copyright 2023 The vLLM team.
# Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
#
# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
# and OPT implementations in this library. It has been modified from its
# original forms to accommodate minor architectural differences compared
# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Inference-only Qwen2-VL model compatible with HuggingFace weights."""
import logging
from functools import lru_cache, partial
from typing import Iterable, List, Optional, Tuple, Type, TypedDict

import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
from einops import rearrange, repeat
from vllm.model_executor.layers.activation import QuickGELU

from sglang.srt.configs import Qwen2VLConfig, Qwen2VLVisionConfig
from sglang.srt.distributed import parallel_state
from sglang.srt.distributed import utils as dist_utils
from sglang.srt.hf_transformers_utils import get_processor
from sglang.srt.layers.attention.vision import VisionAttention
from sglang.srt.layers.linear import ColumnParallelLinear, RowParallelLinear
from sglang.srt.layers.logits_processor import LogitsProcessor
from sglang.srt.layers.pooler import Pooler, PoolingType
from sglang.srt.layers.quantization.base_config import QuantizationConfig
from sglang.srt.layers.vocab_parallel_embedding import ParallelLMHead
from sglang.srt.managers.schedule_batch import ImageInputs
from sglang.srt.model_executor.forward_batch_info import ForwardBatch
from sglang.srt.model_loader.weight_utils import default_weight_loader
from sglang.srt.models.qwen2 import Qwen2Model

logger = logging.getLogger(__name__)


# === Vision Inputs === #


class Qwen2VLImageInputs(TypedDict):
    pixel_values: torch.Tensor
    """Shape:
    `(num_patches, num_channels * patch_size * patch_size)`
    """

    image_grid_thw: torch.Tensor
    """Shape: `(num_images, 3)`

    This should be in `(grid_t, grid_h, grid_w)` format.
    """


class Qwen2VLVideoInputs(TypedDict):
    pixel_values_videos: torch.Tensor
    """Shape:
    `(num_patches,
      num_channels * temporal_patch_size * patch_size * patch_size)`
    """

    video_grid_thw: torch.Tensor
    """Shape: `(num_videos, 3)`

    This should be in `(grid_t, grid_h, grid_w)` format.
    """


# === Vision Encoder === #


class Qwen2VisionMLP(nn.Module):

    def __init__(
        self,
        in_features: int,
        hidden_features: int = None,
        act_layer: Type[nn.Module] = QuickGELU,
        quant_config: Optional[QuantizationConfig] = None,
    ):
        super().__init__()
        self.fc1 = ColumnParallelLinear(
            in_features, hidden_features, quant_config=quant_config
        )
        self.act = act_layer()
        self.fc2 = RowParallelLinear(
            hidden_features, in_features, quant_config=quant_config
        )

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        x_parallel, _ = self.fc1(x)
        x_parallel = self.act(x_parallel)
        x, _ = self.fc2(x_parallel)
        return x


class Qwen2VisionBlock(nn.Module):

    def __init__(
        self,
        dim: int,
        num_heads: int,
        mlp_ratio: float,
        act_layer: Type[nn.Module] = QuickGELU,
        norm_layer: Type[nn.Module] = None,
        quant_config: Optional[QuantizationConfig] = None,
    ) -> None:
        super().__init__()
        if norm_layer is None:
            norm_layer = partial(nn.LayerNorm, eps=1e-6)
        self.norm1 = norm_layer(dim)
        self.norm2 = norm_layer(dim)
        mlp_hidden_dim = int(dim * mlp_ratio)

        self.attn = VisionAttention(
            embed_dim=dim,
            num_heads=num_heads,
            projection_size=dim,
            use_qkv_parallel=False,
            quant_config=quant_config,
        )
        self.mlp = Qwen2VisionMLP(
            dim, mlp_hidden_dim, act_layer=act_layer, quant_config=quant_config
        )

    def forward(
        self, x: torch.Tensor, cu_seqlens: torch.Tensor, rotary_pos_emb: torch.Tensor
    ) -> torch.Tensor:
        hidden_states = self.norm1(x)
        hidden_states = rearrange(hidden_states, "s b ... -> b s ...")
        attn = self.attn(
            hidden_states, cu_seqlens=cu_seqlens, rotary_pos_emb=rotary_pos_emb
        )
        attn = rearrange(attn, "b s ... -> s b ...")
        x = x + attn
        x = x + self.mlp(self.norm2(x))
        return x


class Qwen2VisionPatchEmbed(nn.Module):

    def __init__(
        self,
        patch_size: int = 14,
        temporal_patch_size: int = 2,
        in_chans: int = 3,
        embed_dim: int = 1152,
    ) -> None:
        super().__init__()
        self.patch_size = patch_size
        self.temporal_patch_size = temporal_patch_size
        self.embed_dim = embed_dim

        kernel_size = [temporal_patch_size, patch_size, patch_size]
        self.proj = nn.Conv3d(
            in_chans, embed_dim, kernel_size=kernel_size, stride=kernel_size, bias=False
        )

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        L, C = x.shape
        x = x.view(L, -1, self.temporal_patch_size, self.patch_size, self.patch_size)
        x = self.proj(x).view(L, self.embed_dim)
        return x


class Qwen2VisionPatchMerger(nn.Module):

    def __init__(
        self,
        d_model: int,
        context_dim: int,
        norm_layer: Type[nn.Module] = None,
        spatial_merge_size: int = 2,
        quant_config: Optional[QuantizationConfig] = None,
    ) -> None:
        super().__init__()
        self.hidden_size = context_dim * (spatial_merge_size**2)
        if norm_layer is None:
            norm_layer = partial(nn.LayerNorm, eps=1e-6)
        self.ln_q = norm_layer(context_dim)
        self.mlp = nn.ModuleList(
            [
                ColumnParallelLinear(
                    self.hidden_size,
                    self.hidden_size,
                    bias=True,
                    quant_config=quant_config,
                ),
                nn.GELU(),
                RowParallelLinear(
                    self.hidden_size, d_model, bias=True, quant_config=quant_config
                ),
            ]
        )

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        x = self.ln_q(x)
        x = x.view(-1, self.hidden_size)

        mlp_fc1, mlp_act, mlp_fc2 = self.mlp
        x_parallel, _ = mlp_fc1(x)
        x_parallel = mlp_act(x_parallel)
        out, _ = mlp_fc2(x_parallel)
        return out


class Qwen2VisionRotaryEmbedding(nn.Module):

    def __init__(self, dim: int, theta: float = 10000.0) -> None:
        super().__init__()
        self.dim = dim
        self.theta = theta
        inv_freq = 1.0 / (theta ** (torch.arange(0, dim, 2, dtype=torch.float) / dim))
        self.register_buffer("inv_freq", inv_freq, persistent=False)
        self._seq_len_cached = 0
        self._freqs_cached = None

    def update_freqs_cache(self, seqlen: int) -> None:
        if seqlen > self._seq_len_cached:
            seqlen *= 2
            self._seq_len_cached = seqlen
            self.inv_freq = 1.0 / (
                self.theta
                ** (
                    torch.arange(
                        0, self.dim, 2, dtype=torch.float, device=self.inv_freq.device
                    )
                    / self.dim
                )
            )
            seq = torch.arange(
                seqlen, device=self.inv_freq.device, dtype=self.inv_freq.dtype
            )
            freqs = torch.outer(seq, self.inv_freq)
            self._freqs_cached = freqs

    def forward(self, seqlen: int) -> torch.Tensor:
        self.update_freqs_cache(seqlen)
        return self._freqs_cached[:seqlen]


class Qwen2VisionTransformer(nn.Module):

    def __init__(
        self,
        vision_config: Qwen2VLVisionConfig,
        norm_eps: float = 1e-6,
        quant_config: Optional[QuantizationConfig] = None,
    ) -> None:
        super().__init__()

        patch_size: int = vision_config.patch_size
        temporal_patch_size: int = vision_config.temporal_patch_size
        spatial_merge_size: int = vision_config.spatial_merge_size
        in_chans: int = vision_config.in_chans
        hidden_size: int = vision_config.hidden_size
        embed_dim: int = vision_config.embed_dim
        depth: int = vision_config.depth
        num_heads: int = vision_config.num_heads
        mlp_ratio: float = vision_config.mlp_ratio

        self.spatial_merge_size = spatial_merge_size

        self.patch_embed = Qwen2VisionPatchEmbed(
            patch_size=patch_size,
            temporal_patch_size=temporal_patch_size,
            in_chans=in_chans,
            embed_dim=embed_dim,
        )

        norm_layer = partial(nn.LayerNorm, eps=norm_eps)
        head_dim = embed_dim // num_heads
        self.rotary_pos_emb = Qwen2VisionRotaryEmbedding(head_dim // 2)

        self.blocks = nn.ModuleList(
            [
                Qwen2VisionBlock(
                    dim=embed_dim,
                    num_heads=num_heads,
                    mlp_ratio=mlp_ratio,
                    norm_layer=norm_layer,
                    quant_config=quant_config,
                )
                for _ in range(depth)
            ]
        )
        self.merger = Qwen2VisionPatchMerger(
            d_model=hidden_size,
            context_dim=embed_dim,
            norm_layer=norm_layer,
            quant_config=quant_config,
        )

    @property
    def dtype(self) -> torch.dtype:
        return self.blocks[0].mlp.fc2.weight.dtype

    @property
    def device(self) -> torch.device:
        return self.blocks[0].mlp.fc2.weight.device

    def rot_pos_emb(self, grid_thw: torch.Tensor) -> torch.Tensor:
        pos_ids = []
        for t, h, w in grid_thw:
            hpos_ids = torch.arange(h).unsqueeze(1).expand(-1, w)
            wpos_ids = torch.arange(w).unsqueeze(0).expand(h, -1)
            hpos_ids = (
                hpos_ids.reshape(
                    h // self.spatial_merge_size,
                    self.spatial_merge_size,
                    w // self.spatial_merge_size,
                    self.spatial_merge_size,
                )
                .permute(0, 2, 1, 3)
                .flatten()
            )
            wpos_ids = (
                wpos_ids.reshape(
                    h // self.spatial_merge_size,
                    self.spatial_merge_size,
                    w // self.spatial_merge_size,
                    self.spatial_merge_size,
                )
                .permute(0, 2, 1, 3)
                .flatten()
            )
            pos_ids.append(torch.stack([hpos_ids, wpos_ids], dim=-1).repeat(t, 1))
        pos_ids = torch.cat(pos_ids, dim=0)
        max_grid_size = grid_thw[:, 1:].max()
        rotary_pos_emb_full = self.rotary_pos_emb(max_grid_size)
        rotary_pos_emb = rotary_pos_emb_full[pos_ids].flatten(1)
        return rotary_pos_emb

    def forward(
        self,
        x: torch.Tensor,
        grid_thw: torch.Tensor,
    ) -> torch.Tensor:
        # patchify
        x = x.to(device=self.device, dtype=self.dtype)
        x = self.patch_embed(x)

        # compute position embedding
        rotary_pos_emb = self.rot_pos_emb(grid_thw)

        # compute cu_seqlens
        cu_seqlens = torch.repeat_interleave(
            grid_thw[:, 1] * grid_thw[:, 2], grid_thw[:, 0]
        ).cumsum(dim=0, dtype=torch.int32)
        cu_seqlens = F.pad(cu_seqlens, (1, 0), "constant", 0)

        # transformers
        x = x.unsqueeze(1)
        for blk in self.blocks:
            x = blk(x, cu_seqlens=cu_seqlens, rotary_pos_emb=rotary_pos_emb)

        # adapter
        x = self.merger(x)
        return x


cached_get_processor = lru_cache(get_processor)


class Qwen2VLForConditionalGeneration(nn.Module):
    def calculate_num_image_tokens(self, image_grid_thw: Tuple[int, int, int]):
        processor = cached_get_processor(self.config._name_or_path)
        grid_t, grid_h, grid_w = image_grid_thw
        num_image_tokens = (
            grid_t
            * grid_h
            * grid_w
            // processor.image_processor.merge_size
            // processor.image_processor.merge_size
        )
        return num_image_tokens

    # Use grid_t * grid_w * grid_h to pad tokens for each image
    # add replaced padding by unique image hash
    def pad_input_ids(self, input_ids: List[int], image_inputs: ImageInputs):
        image_grid_thws = image_inputs.image_grid_thws
        pad_values = image_inputs.pad_values

        image_indices = [
            idx
            for idx, token in enumerate(input_ids)
            if token == self.config.image_token_id
        ]
        image_inputs.image_offsets = []

        input_ids_with_image = []
        for image_cnt, _ in enumerate(image_grid_thws):
            num_image_tokens = self.calculate_num_image_tokens(
                image_grid_thws[image_cnt]
            )
            if image_cnt == 0:
                non_image_tokens = input_ids[: image_indices[image_cnt]]
            else:
                non_image_tokens = input_ids[
                    image_indices[image_cnt - 1] + 1 : image_indices[image_cnt]
                ]
            input_ids_with_image.extend(non_image_tokens)
            image_inputs.image_offsets.append(len(input_ids_with_image))
            pad_ids = pad_values * (
                (num_image_tokens + len(pad_values)) // len(pad_values)
            )
            input_ids_with_image.extend(pad_ids[:num_image_tokens])
        input_ids_with_image.extend(input_ids[image_indices[-1] + 1 :])

        return input_ids_with_image

    def __init__(
        self,
        config: Qwen2VLConfig,
        quant_config: Optional[QuantizationConfig] = None,
    ) -> None:
        super().__init__()

        self.config = config
        self.visual = Qwen2VisionTransformer(
            config.vision_config,
            norm_eps=getattr(config, "rms_norm_eps", 1e-6),
            # NOTE: Qwen2-VL vision encoder does not support any
            # quantization method now.
            quant_config=None,
        )

        self.model = Qwen2Model(config, quant_config)

        if config.tie_word_embeddings:
            self.lm_head = self.model.embed_tokens
        else:
            self.lm_head = ParallelLMHead(
                config.vocab_size, config.hidden_size, quant_config=quant_config
            )

        self.logits_processor = LogitsProcessor(config)
        self.pooler = Pooler(pooling_type=PoolingType.LAST, normalize=True)

    def _process_image_input(self, image_input: Qwen2VLImageInputs) -> torch.Tensor:
        pixel_values = image_input["pixel_values"].type(self.visual.dtype)
        image_embeds = self.visual(pixel_values, grid_thw=image_input["image_grid_thw"])
        return image_embeds

    def _process_video_input(self, video_input: Qwen2VLVideoInputs) -> torch.Tensor:
        pixel_values_videos = video_input["pixel_values_videos"].type(self.visual.dtype)
        video_embeds = self.visual(
            pixel_values_videos, grid_thw=video_input["video_grid_thw"]
        )
        return video_embeds

    def forward(
        self,
        input_ids: torch.Tensor,
        positions: torch.Tensor,
        forward_batch: ForwardBatch,
        get_embedding: bool = False,
    ):
        """Run forward pass for Qwen2-VL.

        Args:
            input_ids: Flattened (concatenated) input_ids corresponding to a
                batch.
            positions: Flattened (concatenated) position ids corresponding to a
                batch.
                **NOTE**: If mrope is enabled (default setting for Qwen2-VL
                opensource models), the shape will be `(3, seq_len)`,
                otherwise it will be `(seq_len,).
                (Use input_metadata.mrope_positions to replace it)
            pixel_values: Pixel values to be fed to a model.
                `None` if no images are passed.
            image_grid_thw: Tensor `(n_images, 3)` of image 3D grid in LLM.
                `None` if no images are passed.
        """
        if getattr(self.config, "rope_scaling", {}).get("type", None) == "mrope":
            positions = forward_batch.mrope_positions

        image_inputs = None
        if forward_batch.image_inputs is not None:
            image_inputs = [
                img for img in forward_batch.image_inputs if img is not None
            ]

        if (
            forward_batch.forward_mode.is_decode()
            or image_inputs is None
            or len(image_inputs) == 0
        ):
            inputs_embeds = self.model.embed_tokens(input_ids)
        else:
            if getattr(self.config, "rope_scaling", {}).get("type", None) == "mrope":
                assert positions.ndim == 2 and positions.size(0) == 3, (
                    "multimodal section rotary embedding requires "
                    f"(3, seq_len) positions, but got {positions.size()}"
                )

            # Clamp input ids. This is because the input_ids for the image tokens are
            # filled with the hash values of the image for the prefix matching in the radix attention.
            # There values are useless because their embeddings will be replaced by vision embeddings anyway.
            input_ids.clamp_(min=0, max=self.config.vocab_size - 1)

            inputs_embeds = self.model.embed_tokens(input_ids)
            extend_start_loc_cpu = forward_batch.extend_start_loc.cpu().numpy()
            prefix_lens_cpu = forward_batch.extend_prefix_lens_cpu
            for i, image in enumerate(forward_batch.image_inputs):
                if image is None:
                    continue
                start_idx = extend_start_loc_cpu[i]
                prefix_len = prefix_lens_cpu[i]

                pixel_values = torch.tensor(image.pixel_values, device="cuda")
                image_grid_thws = torch.tensor(
                    np.array(image.image_grid_thws), device="cuda"
                )
                image_offsets = image.image_offsets
                image_input = Qwen2VLImageInputs(
                    pixel_values=pixel_values, image_grid_thw=image_grid_thws
                )
                image_embeds = self._process_image_input(image_input)

                image_embeds_offset = 0
                for idx, image_offset in enumerate(image_offsets):
                    if image_offset < prefix_len:
                        continue
                    num_image_tokens = self.calculate_num_image_tokens(
                        image_grid_thws[idx]
                    )
                    left_idx = start_idx + (image_offset - prefix_len)
                    right_idx = (
                        start_idx + (image_offset - prefix_len) + num_image_tokens
                    )
                    inputs_embeds[left_idx:right_idx] = image_embeds[
                        image_embeds_offset : image_embeds_offset + num_image_tokens
                    ]
                    image_embeds_offset += num_image_tokens

        hidden_states = self.model(
            input_ids=input_ids,
            positions=positions,
            forward_batch=forward_batch,
            input_embeds=inputs_embeds,
        )

        if not get_embedding:
            return self.logits_processor(
                input_ids, hidden_states, self.lm_head, forward_batch
            )
        else:
            return self.pooler(hidden_states, forward_batch)

    def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
        stacked_params_mapping = [
            # (param_name, shard_name, shard_id)
            ("qkv_proj", "q_proj", "q"),
            ("qkv_proj", "k_proj", "k"),
            ("qkv_proj", "v_proj", "v"),
            ("gate_up_proj", "up_proj", 1),
            ("gate_up_proj", "gate_proj", 0),
        ]
        params_dict = dict(self.named_parameters(remove_duplicate=False))
        for name, loaded_weight in weights:
            if "rotary_emb.inv_freq" 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:

                if "visual" in name and "qkv.weight" in name:
                    visual_num_heads = self.config.vision_config.num_heads
                    visual_embed_dim = self.config.vision_config.embed_dim
                    head_size = visual_embed_dim // visual_num_heads
                    loaded_weight = loaded_weight.view(
                        3, visual_num_heads, head_size, visual_embed_dim
                    )
                    loaded_weight = loaded_weight.transpose(0, 1)
                    loaded_weight = loaded_weight.reshape(-1, visual_embed_dim)
                elif "visual" in name and "qkv.bias" in name:
                    visual_num_heads = self.config.vision_config.num_heads
                    visual_embed_dim = self.config.vision_config.embed_dim
                    head_size = visual_embed_dim // visual_num_heads
                    loaded_weight = loaded_weight.view(3, visual_num_heads, head_size)
                    loaded_weight = loaded_weight.transpose(0, 1)
                    loaded_weight = loaded_weight.reshape(-1)

                if "visual" in name:
                    # adapt to VisionAttention
                    name = name.replace(r"attn.qkv.", r"attn.qkv_proj.")

                try:
                    # Skip loading extra bias for GPTQ models.
                    if name.endswith(".bias") and name not in params_dict:
                        continue
                    param = params_dict[name]
                except KeyError:
                    print(params_dict.keys())
                    raise

                weight_loader = getattr(param, "weight_loader", default_weight_loader)
                weight_loader(param, loaded_weight)


EntryClass = Qwen2VLForConditionalGeneration