modeling_navil_vit_anyres.py

# --------------------------------------------------------
# NaViL
# Copyright (c) 2025 OpenGVLab
# Licensed under The MIT License [see LICENSE for details]
# --------------------------------------------------------
from typing import Optional, Tuple, Union
from functools import partial

import torch
import torch.nn.functional as F
import torch.utils.checkpoint

from einops import rearrange
from timm.models.layers import DropPath
from torch import nn
from transformers.activations import ACT2FN
from transformers.modeling_outputs import (BaseModelOutput,
                                           BaseModelOutputWithPooling)
from transformers.modeling_utils import PreTrainedModel
from transformers.utils import logging

from .configuration_navil_vit import NaViLVisionConfig
from .modular_intern_vit import (
    InternVisionFlashAttention2,
    InternVisionSdpaAttention,
    InternMLP,
    NORM2FN,
    InternVisionRotaryEmbedding,
)

try:
    # from .flash_attention import FlashAttention
    from flash_attn import flash_attn_varlen_func
    from flash_attn.layers.rotary import apply_rotary_emb
    has_flash_attn = True
except:
    print('FlashAttention is not installed.')
    has_flash_attn = False

logger = logging.get_logger(__name__)


class NaViLVisionEmbeddingsAnyRes(nn.Module):
    def __init__(self, config: NaViLVisionConfig):
        super().__init__()
        self.config = config
        self.embed_dim = config.hidden_size
        self.image_size = config.image_size
        self.patch_size = config.patch_size
        self.merge_size = int(1.0 / config.downsample_ratio)

        self.patch_embedding = nn.Conv2d(
            in_channels=3, out_channels=self.embed_dim, kernel_size=self.patch_size, stride=self.patch_size
        )

        self.num_patches = (self.image_size // self.patch_size) ** 2
        self.num_positions = self.num_patches + 1

    def forward(self, pixel_values: torch.FloatTensor) -> torch.Tensor:
        target_dtype = self.patch_embedding.weight.dtype
        patch_embeds = self.patch_embedding(pixel_values)  # shape = [*, channel, width, height]
        batch_size, _, height, width = patch_embeds.shape

        return patch_embeds.flatten(1)


class NaViLVisionEncoderLayerAnyRes(nn.Module):
    def __init__(self, config: NaViLVisionConfig, drop_path_rate: float):
        super().__init__()
        self.embed_dim = config.hidden_size
        self.intermediate_size = config.intermediate_size
        self.norm_type = config.norm_type

        if has_flash_attn:
            self.attn = InternVisionFlashAttention2(config)
        else:
            self.attn = InternVisionSdpaAttention(config)
        self.mlp = InternMLP(config)
        self.norm1 = NORM2FN[self.norm_type](self.embed_dim, eps=config.layer_norm_eps)
        self.norm2 = NORM2FN[self.norm_type](self.embed_dim, eps=config.layer_norm_eps)

        self.ls1 = nn.Parameter(config.initializer_factor * torch.ones(self.embed_dim))
        self.ls2 = nn.Parameter(config.initializer_factor * torch.ones(self.embed_dim))
        self.drop_path1 = DropPath(drop_path_rate) if drop_path_rate > 0. else nn.Identity()
        self.drop_path2 = DropPath(drop_path_rate) if drop_path_rate > 0. else nn.Identity()

    def forward(
            self,
            hidden_states: torch.Tensor,
            cu_seqlens,
            rotary_pos_emb
    ) -> Tuple[torch.FloatTensor, Optional[torch.FloatTensor], Optional[Tuple[torch.FloatTensor]]]:
        """
        Args:
            hidden_states (`Tuple[torch.FloatTensor, Optional[torch.FloatTensor]]`): input to the layer of shape `(batch, seq_len, embed_dim)`
        """
        hidden_states = hidden_states + self.drop_path1(
            self.attn(
                self.norm1(hidden_states), 
                cu_seqlens=cu_seqlens,
                rotary_pos_emb=rotary_pos_emb,
                ) * self.ls1)

        hidden_states = hidden_states + self.drop_path2(self.mlp(self.norm2(hidden_states)) * self.ls2)

        return hidden_states


class NaViLVisionEncoderAnyRes(nn.Module):
    """
    Transformer encoder consisting of `config.num_hidden_layers` self attention layers. Each layer is a
    [`InternEncoderLayer`].

    Args:
        config (`InternConfig`):
            The corresponding vision configuration for the `InternEncoder`.
    """

    def __init__(self, config: NaViLVisionConfig):
        super().__init__()
        self.config = config
        # stochastic depth decay rule
        dpr = [x.item() for x in torch.linspace(0, config.drop_path_rate, config.num_hidden_layers)]
        self.layers = nn.ModuleList([
            NaViLVisionEncoderLayerAnyRes(config, dpr[idx]) for idx in range(config.num_hidden_layers)])
        self.gradient_checkpointing = True

        head_dim = config.hidden_size // config.num_attention_heads
        self.rotary_pos_emb = InternVisionRotaryEmbedding(head_dim // 2)

        self.merge_size = int(1.0 / config.downsample_ratio)
        self.merge_unit = self.merge_size * self.merge_size
        self.patch_size = config.patch_size
        self.fullatt_block_indexes = config.fullatt_block_indexes
        self.window_size = config.window_size
    
    def rot_pos_emb(self, grid_thw):
        pos_ids = []
        for t, h, w in grid_thw:
            hpos_ids = torch.arange(h).unsqueeze(1).expand(-1, w)
            hpos_ids = hpos_ids.reshape(
                h // self.merge_size,
                self.merge_size,
                w // self.merge_size,
                self.merge_size,
            )
            hpos_ids = hpos_ids.permute(0, 2, 1, 3)
            hpos_ids = hpos_ids.flatten()

            wpos_ids = torch.arange(w).unsqueeze(0).expand(h, -1)
            wpos_ids = wpos_ids.reshape(
                h // self.merge_size,
                self.merge_size,
                w // self.merge_size,
                self.merge_size,
            )
            wpos_ids = wpos_ids.permute(0, 2, 1, 3)
            wpos_ids = wpos_ids.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 get_window_index(self, grid_thw):
        window_index: list = []
        cu_window_seqlens: list = [0]
        window_index_id = 0
        vit_merger_window_size = self.window_size // self.merge_size
        assert vit_merger_window_size > 0

        for grid_t, grid_h, grid_w in grid_thw:
            llm_grid_h, llm_grid_w = (
                grid_h // self.merge_size,
                grid_w // self.merge_size,
            )
            index = torch.arange(grid_t * llm_grid_h * llm_grid_w).reshape(grid_t, llm_grid_h, llm_grid_w)
            pad_h = vit_merger_window_size - llm_grid_h % vit_merger_window_size
            pad_w = vit_merger_window_size - llm_grid_w % vit_merger_window_size
            num_windows_h = (llm_grid_h + pad_h) // vit_merger_window_size
            num_windows_w = (llm_grid_w + pad_w) // vit_merger_window_size
            index_padded = F.pad(index, (0, pad_w, 0, pad_h), "constant", -100)
            index_padded = index_padded.reshape(
                grid_t,
                num_windows_h,
                vit_merger_window_size,
                num_windows_w,
                vit_merger_window_size,
            )
            index_padded = index_padded.permute(0, 1, 3, 2, 4).reshape(
                grid_t,
                num_windows_h * num_windows_w,
                vit_merger_window_size,
                vit_merger_window_size,
            )
            seqlens = (index_padded != -100).sum([2, 3]).reshape(-1)
            index_padded = index_padded.reshape(-1)
            index_new = index_padded[index_padded != -100]
            window_index.append(index_new + window_index_id)
            cu_seqlens_tmp = seqlens.cumsum(0) * self.merge_unit + cu_window_seqlens[-1]
            cu_window_seqlens.extend(cu_seqlens_tmp.tolist())
            window_index_id += (grid_t * llm_grid_h * llm_grid_w).item()
        window_index = torch.cat(window_index, dim=0)

        return window_index, cu_window_seqlens

    def forward(
            self,
            inputs_embeds,
            output_hidden_states: Optional[bool] = None,
            return_dict: Optional[bool] = None,
            grid_thw: Optional[torch.Tensor] = None,
    ) -> Union[Tuple, BaseModelOutput]:
        r"""
        Args:
            inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
                Embedded representation of the inputs. Should be float, not int tokens.
            output_hidden_states (`bool`, *optional*):
                Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors
                for more detail.
            return_dict (`bool`, *optional*):
                Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
        """
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        encoder_states = () if output_hidden_states else None
        hidden_states = inputs_embeds

        rotary_pos_emb = self.rot_pos_emb(grid_thw)
        window_index, cu_window_seqlens = self.get_window_index(grid_thw)
        cu_window_seqlens = torch.tensor(
            cu_window_seqlens,
            device=hidden_states.device,
            dtype=grid_thw.dtype if torch.jit.is_tracing() else torch.int32,
        )
        cu_window_seqlens = torch.unique_consecutive(cu_window_seqlens)

        seq_len, _ = hidden_states.size()
        hidden_states = hidden_states.reshape(seq_len // self.merge_unit, self.merge_unit, -1)
        hidden_states = hidden_states[window_index, :, :]
        hidden_states = hidden_states.reshape(seq_len, -1)
        rotary_pos_emb = rotary_pos_emb.reshape(seq_len // self.merge_unit, self.merge_unit, -1)
        rotary_pos_emb = rotary_pos_emb[window_index, :, :]
        rotary_pos_emb = rotary_pos_emb.reshape(seq_len, -1)

        cu_seqlens = torch.repeat_interleave(grid_thw[:, 1] * grid_thw[:, 2], grid_thw[:, 0]).cumsum(
            dim=0,
            # Select dtype based on the following factors:
            #  - FA2 requires that cu_seqlens_q must have dtype int32
            #  - torch.onnx.export requires that cu_seqlens_q must have same dtype as grid_thw
            # See https://github.com/huggingface/transformers/pull/34852 for more information
            dtype=grid_thw.dtype if torch.jit.is_tracing() else torch.int32,
        )
        cu_seqlens = F.pad(cu_seqlens, (1, 0), value=0)


        for idx, encoder_layer in enumerate(self.layers):
            if (self.fullatt_block_indexes is None) or (idx in self.fullatt_block_indexes):
                cu_seqlens_now = cu_seqlens
            else:
                cu_seqlens_now = cu_window_seqlens
            if output_hidden_states:
                encoder_states = encoder_states + (hidden_states,)
            if self.gradient_checkpointing and self.training:
                layer_outputs = torch.utils.checkpoint.checkpoint(
                    partial(encoder_layer, cu_seqlens=cu_seqlens_now, rotary_pos_emb=rotary_pos_emb),
                    hidden_states)
            else:
                layer_outputs = encoder_layer(
                    hidden_states,
                    cu_seqlens=cu_seqlens_now,
                    rotary_pos_emb=rotary_pos_emb,
                )
            hidden_states = layer_outputs

        if output_hidden_states:
            encoder_states = encoder_states + (hidden_states,)

        if not return_dict:
            return tuple(v for v in [hidden_states, encoder_states] if v is not None)
        return BaseModelOutput(
            last_hidden_state=hidden_states, hidden_states=encoder_states
        )


class NaViLVisionModelAnyRes(PreTrainedModel):
    main_input_name = 'pixel_values'
    config_class = NaViLVisionConfig
    _no_split_modules = ['NaViLVisionEncoderLayerAnyRes']

    def __init__(self, config: NaViLVisionConfig):
        super().__init__(config)
        self.config = config
        
        self.merge_size = int(1.0 / config.downsample_ratio)
        self.embeddings = NaViLVisionEmbeddingsAnyRes(config)
        self.encoder = NaViLVisionEncoderAnyRes(config)
        
    def get_input_embeddings(self):
        return self.embeddings

    def forward(
        self,
        pixel_values: Optional[torch.FloatTensor] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        pixel_embeds: Optional[torch.FloatTensor] = None,
        grid_thw: Optional[torch.Tensor] = None,
    ) -> Union[Tuple, BaseModelOutputWithPooling]:
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        if pixel_values is None and pixel_embeds is None:
            raise ValueError('You have to specify pixel_values or pixel_embeds')

        if pixel_embeds is not None:
            hidden_states = pixel_embeds
        else:
            if len(pixel_values.shape) == 4:
                hidden_states = self.embeddings(pixel_values)
            else:
                raise ValueError(f'wrong pixel_values size: {pixel_values.shape}')
            
        encoder_outputs = self.encoder(
            inputs_embeds=hidden_states,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            grid_thw=grid_thw
        )
        last_hidden_state = encoder_outputs.last_hidden_state
        # pooled_output = last_hidden_state[:, 0, :]

        last_hidden_state = last_hidden_state.unsqueeze(1).reshape(-1, self.merge_size, self.merge_size, last_hidden_state.shape[-1])

        if not return_dict:
            return (last_hidden_state, ) + encoder_outputs[1:]
        
        return BaseModelOutputWithPooling(
            last_hidden_state=last_hidden_state,
            pooler_output=None,
            hidden_states=encoder_outputs.hidden_states,
            attentions=encoder_outputs.attentions,
        )