lavis/models/clip_vit.py

from collections import OrderedDict
from itertools import repeat
import collections.abc
import math

import torch
import torch.nn.functional as F
from torch import nn

from fairscale.nn.checkpoint.checkpoint_activations import checkpoint_wrapper

from lavis.models.eva_vit import convert_weights_to_fp16
from lavis.common.dist_utils import download_cached_file

class Bottleneck(nn.Module):
    expansion = 4

    def __init__(self, inplanes, planes, stride=1):
        super().__init__()

        # all conv layers have stride 1. an avgpool is performed after the second convolution when stride > 1
        self.conv1 = nn.Conv2d(inplanes, planes, 1, bias=False)
        self.bn1 = nn.BatchNorm2d(planes)
        self.relu1 = nn.ReLU(inplace=True)

        self.conv2 = nn.Conv2d(planes, planes, 3, padding=1, bias=False)
        self.bn2 = nn.BatchNorm2d(planes)
        self.relu2 = nn.ReLU(inplace=True)

        self.avgpool = nn.AvgPool2d(stride) if stride > 1 else nn.Identity()

        self.conv3 = nn.Conv2d(planes, planes * self.expansion, 1, bias=False)
        self.bn3 = nn.BatchNorm2d(planes * self.expansion)
        self.relu3 = nn.ReLU(inplace=True)

        self.downsample = None
        self.stride = stride

        if stride > 1 or inplanes != planes * Bottleneck.expansion:
            # downsampling layer is prepended with an avgpool, and the subsequent convolution has stride 1
            self.downsample = nn.Sequential(OrderedDict([
                ("-1", nn.AvgPool2d(stride)),
                ("0", nn.Conv2d(inplanes, planes * self.expansion, 1, stride=1, bias=False)),
                ("1", nn.BatchNorm2d(planes * self.expansion))
            ]))

    def forward(self, x: torch.Tensor):
        identity = x

        out = self.relu1(self.bn1(self.conv1(x)))
        out = self.relu2(self.bn2(self.conv2(out)))
        out = self.avgpool(out)
        out = self.bn3(self.conv3(out))

        if self.downsample is not None:
            identity = self.downsample(x)

        out += identity
        out = self.relu3(out)
        return out


class AttentionPool2d(nn.Module):
    def __init__(self, spacial_dim: int, embed_dim: int, num_heads: int, output_dim: int = None):
        super().__init__()
        self.positional_embedding = nn.Parameter(torch.randn(spacial_dim ** 2 + 1, embed_dim) / embed_dim ** 0.5)
        self.k_proj = nn.Linear(embed_dim, embed_dim)
        self.q_proj = nn.Linear(embed_dim, embed_dim)
        self.v_proj = nn.Linear(embed_dim, embed_dim)
        self.c_proj = nn.Linear(embed_dim, output_dim or embed_dim)
        self.num_heads = num_heads

    def forward(self, x):
        x = x.reshape(x.shape[0], x.shape[1], x.shape[2] * x.shape[3]).permute(2, 0, 1)  # NCHW -> (HW)NC
        x = torch.cat([x.mean(dim=0, keepdim=True), x], dim=0)  # (HW+1)NC
        x = x + self.positional_embedding[:, None, :].to(x.dtype)  # (HW+1)NC
        x, _ = F.multi_head_attention_forward(
            query=x, key=x, value=x,
            embed_dim_to_check=x.shape[-1],
            num_heads=self.num_heads,
            q_proj_weight=self.q_proj.weight,
            k_proj_weight=self.k_proj.weight,
            v_proj_weight=self.v_proj.weight,
            in_proj_weight=None,
            in_proj_bias=torch.cat([self.q_proj.bias, self.k_proj.bias, self.v_proj.bias]),
            bias_k=None,
            bias_v=None,
            add_zero_attn=False,
            dropout_p=0,
            out_proj_weight=self.c_proj.weight,
            out_proj_bias=self.c_proj.bias,
            use_separate_proj_weight=True,
            training=self.training,
            need_weights=False
        )

        return x[0]
    

class LayerNorm(nn.LayerNorm):
    """Subclass torch's LayerNorm to handle fp16."""

    def forward(self, x: torch.Tensor):
        orig_type = x.dtype
        ret = super().forward(x.type(torch.float32))
        return ret.type(orig_type)


class QuickGELU(nn.Module):
    def forward(self, x: torch.Tensor):
        return x * torch.sigmoid(1.702 * x)


class ResidualAttentionBlock(nn.Module):
    def __init__(self, d_model: int, n_head: int, attn_mask: torch.Tensor = None, use_grad_checkpointing=False):
        super().__init__()

        self.attn = nn.MultiheadAttention(d_model, n_head)
        self.ln_1 = LayerNorm(d_model)
        self.mlp = nn.Sequential(OrderedDict([
            ("c_fc", nn.Linear(d_model, d_model * 4)),
            ("gelu", QuickGELU()),
            ("c_proj", nn.Linear(d_model * 4, d_model))
        ]))
        self.ln_2 = LayerNorm(d_model)
        self.attn_mask = attn_mask

        if use_grad_checkpointing:
            self.attn = checkpoint_wrapper(self.attn)
            self.mlp = checkpoint_wrapper(self.mlp)
            
    def attention(self, x: torch.Tensor):
        self.attn_mask = self.attn_mask.to(dtype=x.dtype, device=x.device) if self.attn_mask is not None else None
        return self.attn(x, x, x, need_weights=False, attn_mask=self.attn_mask)[0]

    def forward(self, x: torch.Tensor):
        x = x + self.attention(self.ln_1(x))
        x = x + self.mlp(self.ln_2(x))
        return x


class Transformer(nn.Module):
    def __init__(self, width: int, layers: int, heads: int, attn_mask: torch.Tensor = None, use_grad_checkpointing=False):
        super().__init__()
        self.width = width
        self.layers = layers
        self.resblocks = nn.Sequential(*[ResidualAttentionBlock(width, heads, attn_mask, use_grad_checkpointing and i>12) for i in range(layers)])

    def forward(self, x: torch.Tensor):
        return self.resblocks(x)


class VisionTransformer(nn.Module):
    def __init__(self, input_resolution: int, patch_size: int, width: int, layers: int, heads: int, use_grad_checkpointing: bool):
        super().__init__()
        self.input_resolution = input_resolution
        self.num_features = width
        self.num_heads = heads
        self.num_patches = (input_resolution // patch_size) ** 2
        self.conv1 = nn.Conv2d(in_channels=3, out_channels=width, kernel_size=patch_size, stride=patch_size, bias=False)

        scale = width ** -0.5
        self.class_embedding = nn.Parameter(scale * torch.randn(width))
        self.positional_embedding = nn.Parameter(scale * torch.randn(self.num_patches + 1, width))
        self.ln_pre = LayerNorm(width)
        
        self.transformer = Transformer(width, layers, heads, use_grad_checkpointing=use_grad_checkpointing)
           
#         self.ln_final = LayerNorm(width)

    def forward(self, x: torch.Tensor):

        x = self.conv1(x)  # shape = [*, width, grid, grid]
        x = x.reshape(x.shape[0], x.shape[1], -1)  # shape = [*, width, grid ** 2]
        x = x.permute(0, 2, 1)  # shape = [*, grid ** 2, width]
        x = torch.cat([self.class_embedding.to(x.dtype) + torch.zeros(x.shape[0], 1, x.shape[-1], dtype=x.dtype, device=x.device), x], dim=1)  # shape = [*, grid ** 2 + 1, width]
        x = x + self.positional_embedding.to(x.dtype)
        x = self.ln_pre(x)

        x = x.permute(1, 0, 2)  # NLD -> LND
        x = self.transformer(x)
        x = x.permute(1, 0, 2)  # LND -> NLD
        
#         x = self.ln_final(x)
        return x
    
    def get_num_layer(self, var_name=""):
        if var_name in ("class_embedding", "positional_embedding", "conv1", "ln_pre"):
            return 0
        elif var_name.startswith("transformer.resblocks"):
            layer_id = int(var_name.split('.')[2])
            return layer_id + 1
        else:
            return len(self.transformer.resblocks)    
            
            
# From PyTorch internals
def _ntuple(n):
    def parse(x):
        if isinstance(x, collections.abc.Iterable):
            return x
        return tuple(repeat(x, n))
    return parse
to_2tuple = _ntuple(2)    
    
def interpolate_pos_embed(model, state_dict, interpolation: str = 'bicubic', seq_dim=1):
    # Rescale the grid of position embeddings when loading from state_dict
    old_pos_embed = state_dict.get('positional_embedding', None)
    
    grid_size = round((model.positional_embedding.shape[0] - 1) ** 0.5)
    if old_pos_embed is None:
        return
    grid_size = to_2tuple(grid_size)
    extra_tokens = 1  # FIXME detect different token configs (ie no class token, or more)
    new_seq_len = grid_size[0] * grid_size[1] + extra_tokens
    if new_seq_len == old_pos_embed.shape[0]:
        return

    if extra_tokens:
        pos_emb_tok, pos_emb_img = old_pos_embed[:extra_tokens], old_pos_embed[extra_tokens:]
    else:
        pos_emb_tok, pos_emb_img = None, old_pos_embed
        
    old_grid_size = to_2tuple(int(math.sqrt(len(pos_emb_img))))

    print('Resizing position embedding grid-size from %s to %s', old_grid_size, grid_size)
    pos_emb_img = pos_emb_img.reshape(1, old_grid_size[0], old_grid_size[1], -1).permute(0, 3, 1, 2)
    pos_emb_img = F.interpolate(
        pos_emb_img,
        size=grid_size,
        mode=interpolation,
        align_corners=True,
    )
    pos_emb_img = pos_emb_img.permute(0, 2, 3, 1).reshape(1, grid_size[0] * grid_size[1], -1)[0]
    if pos_emb_tok is not None:
        new_pos_embed = torch.cat([pos_emb_tok, pos_emb_img], dim=0)
    else:
        new_pos_embed = pos_emb_img
    state_dict['positional_embedding'] = new_pos_embed
    
    
def create_clip_vit_L(img_size=224,use_checkpoint=False,precision="fp16"):
    model = VisionTransformer(
            input_resolution=img_size,
            patch_size=14,
            width=1024,
            layers=23,
            heads=16,
            use_grad_checkpointing=use_checkpoint,
        )         
    url = "https://storage.googleapis.com/sfr-vision-language-research/LAVIS/models/BLIP2/clip_vit_L.pth"
    cached_file = download_cached_file(
        url, check_hash=False, progress=True
    )
    state_dict = torch.load(cached_file, map_location="cpu")    
    interpolate_pos_embed(model,state_dict)
    
    incompatible_keys = model.load_state_dict(state_dict, strict=False)
    # print(incompatible_keys)
    
    if precision == "fp16":
        convert_weights_to_fp16(model)
    return model