models.py

# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.

# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
# --------------------------------------------------------
# References:
# GLIDE: https://github.com/openai/glide-text2im
# MAE: https://github.com/facebookresearch/mae/blob/main/models_mae.py
# --------------------------------------------------------

import torch
import torch.nn as nn
import numpy as np
import math
from timm.models.vision_transformer import PatchEmbed,  Mlp
from torch.jit import Final
from timm.layers.helpers import to_2tuple
from functools import partial
from einops import rearrange
from dynamic_model import TokenSelect, DiffRate


def round_to_nearest(input_size, width_mult, num_heads, min_value=1):
    new_width_mult = round(num_heads * width_mult) * 1.0 / num_heads
    input_size = int(new_width_mult * input_size)

    new_input_size = max(min_value, input_size)
    return new_input_size


class DynaLinear(nn.Linear):
    def __init__(self, in_features, out_features, num_heads, head_dim, bias=True, dyna_dim=[True, True], width_mult=1.):
        super(DynaLinear, self).__init__(
            in_features, out_features, bias=bias)
        self.in_features_max = in_features
        self.out_features_max = out_features
        self.num_heads = num_heads
        self.head_dim = head_dim
        self.width_mult = width_mult
        self.dyna_dim = dyna_dim

    def forward(self, input_x):
        if self.dyna_dim[0]:
            self.in_features = round_to_nearest(self.in_features_max, self.width_mult, self.num_heads)
        if self.dyna_dim[1]:
            self.out_features = round_to_nearest(self.out_features_max, self.width_mult, self.num_heads)
        weight = self.weight[:self.out_features, :self.in_features]
        if self.bias is not None:
            bias = self.bias[:self.out_features]
        else:
            bias = self.bias
        return nn.functional.linear(input_x, weight, bias)



class DynaQKVLinear(nn.Linear):
    def __init__(self, in_features, out_features, num_heads, head_dim, bias=True, dyna_dim=[True, True], width_mult=1.):
        super(DynaQKVLinear, self).__init__(
            in_features, out_features, bias=bias)
        self.in_features_max = in_features
        self.out_features_max = out_features // 3 # Q K V
        self.num_heads = num_heads
        self.head_dim = head_dim
        self.width_mult = width_mult
        self.dyna_dim = dyna_dim

    def forward(self, input_x):

        if self.dyna_dim[0]:
            self.in_features = round_to_nearest(self.in_features_max, self.width_mult, self.num_heads)
        if self.dyna_dim[1]:
            self.out_features = round_to_nearest(self.out_features_max, self.width_mult, self.num_heads)

        weight = rearrange(self.weight, "(qkv in_c) out_c -> qkv in_c out_c", qkv=3)
        weight = weight[:, :self.out_features, :self.in_features]
        weight = rearrange(weight, "qkv in_c out_c -> (qkv in_c) out_c")

        
        if self.bias is not None:
            bias = rearrange(self.bias, "(qkv out_c) -> qkv out_c", qkv=3)
            bias = bias[:, :self.out_features]
            bias = rearrange(bias, "qkv out_c -> (qkv out_c)")

        else:
            bias = self.bias
        return nn.functional.linear(input_x, weight, bias)
    
    
class Attention(nn.Module):

    def __init__(
            self,
            dim: int,
            num_heads: int = 8,
            qkv_bias: bool = False,
            qk_norm: bool = False,
            attn_drop: float = 0.,
            proj_drop: float = 0.,
            norm_layer: nn.Module = nn.LayerNorm,
    ) -> None:
        super().__init__()
        assert dim % num_heads == 0, 'dim should be divisible by num_heads'
        self.num_heads = num_heads
        self.head_dim = dim // num_heads
        self.scale = self.head_dim ** -0.5

        # self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias) # TODO
        self.qkv = DynaQKVLinear(in_features=dim, out_features=dim * 3, num_heads=self.num_heads, head_dim=self.head_dim, bias=qkv_bias, dyna_dim=[False, True])
        self.q_norm = norm_layer(self.head_dim) if qk_norm else nn.Identity()
        self.k_norm = norm_layer(self.head_dim) if qk_norm else nn.Identity()
        self.attn_drop = nn.Dropout(attn_drop)
        self.proj = nn.Linear(dim, dim)
        self.proj = DynaLinear(in_features=dim, out_features=dim, num_heads=self.num_heads,  head_dim=self.head_dim, bias=True, dyna_dim=[True, False])
       
        self.proj_drop = nn.Dropout(proj_drop)
        
        # self.dyna_num_heads = round(self.num_heads)

    def forward(self, x: torch.Tensor, channel_mask=None) -> torch.Tensor:
        B, N, C = x.shape
        qkv = self.qkv(x)
        qkv = qkv.reshape(B, N, 3, self.num_heads, self.head_dim).permute(2, 0, 3, 1, 4)
        q, k, v = qkv.unbind(0)
        q, k = self.q_norm(q), self.k_norm(k)


        q = q * self.scale
        attn = q @ k.transpose(-2, -1)
        attn = attn.softmax(dim=-1)
        attn = self.attn_drop(attn)
        x = attn @ v

        x = x.transpose(1, 2).reshape(B, N, -1)
        
        if channel_mask is not None:
            channel_mask = channel_mask.unsqueeze(dim=2).repeat(1, 1, self.head_dim).flatten(1).unsqueeze(1)
            x = channel_mask * x
                    
        x = self.proj(x)
        x = self.proj_drop(x)
        return x


class Mlp(nn.Module):
    """ MLP as used in Vision Transformer, MLP-Mixer and related networks
    """
    def __init__(
            self,
            in_features,
            hidden_features=None,
            out_features=None,
            act_layer=nn.GELU,
            norm_layer=None,
            bias=True,
            drop=0.,
            use_conv=False,
            num_heads=8
    ):
        super().__init__()
        out_features = out_features or in_features
        hidden_features = hidden_features or in_features
        bias = to_2tuple(bias)
        drop_probs = to_2tuple(drop)
        linear_layer = partial(nn.Conv2d, kernel_size=1) if use_conv else nn.Linear
        

        # self.fc1 = linear_layer(in_features, hidden_features, bias=bias[0])
        self.fc1 = DynaLinear(in_features=in_features, out_features=hidden_features, num_heads=num_heads, head_dim=(hidden_features // num_heads), bias=bias[0], dyna_dim=[False, True])
        self.act = act_layer()
        self.drop1 = nn.Dropout(drop_probs[0])
        self.norm = norm_layer(hidden_features) if norm_layer is not None else nn.Identity()
        # self.fc2 = linear_layer(hidden_features, out_features, bias=bias[1])
        self.fc2 = DynaLinear(in_features=hidden_features, out_features=out_features, num_heads=num_heads, head_dim=(hidden_features // num_heads), bias=bias[1], dyna_dim=[True, False])
        self.drop2 = nn.Dropout(drop_probs[1])
        
        self.head_dim = hidden_features // num_heads

    def forward(self, x, channel_mask=None):
        x = self.fc1(x)
        
        if channel_mask is not None:
            channel_mask = channel_mask.unsqueeze(dim=2).repeat(1, 1, self.head_dim).flatten(1).unsqueeze(1)
            x = channel_mask * x
            
        x = self.act(x)
        x = self.drop1(x)
        x = self.norm(x)
        x = self.fc2(x)
        x = self.drop2(x)
        return x

    
def modulate(x, shift, scale):
    return x * (1 + scale.unsqueeze(1)) + shift.unsqueeze(1)


#################################################################################
#               Embedding Layers for Timesteps and Class Labels                 #
#################################################################################

class TimestepEmbedder(nn.Module):
    """
    Embeds scalar timesteps into vector representations.
    """
    def __init__(self, hidden_size, frequency_embedding_size=256):
        super().__init__()
        self.mlp = nn.Sequential(
            nn.Linear(frequency_embedding_size, hidden_size, bias=True),
            nn.SiLU(),
            nn.Linear(hidden_size, hidden_size, bias=True),
        )
        self.frequency_embedding_size = frequency_embedding_size

    @staticmethod
    def timestep_embedding(t, dim, max_period=10000):
        """
        Create sinusoidal timestep embeddings.
        :param t: a 1-D Tensor of N indices, one per batch element.
                          These may be fractional.
        :param dim: the dimension of the output.
        :param max_period: controls the minimum frequency of the embeddings.
        :return: an (N, D) Tensor of positional embeddings.
        """
        # https://github.com/openai/glide-text2im/blob/main/glide_text2im/nn.py
        half = dim // 2
        freqs = torch.exp(
            -math.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32) / half
        ).to(device=t.device)
        args = t[:, None].float() * freqs[None]
        embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1)
        if dim % 2:
            embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1)
        return embedding

    def forward(self, t):
        t_freq = self.timestep_embedding(t, self.frequency_embedding_size)
        t_emb = self.mlp(t_freq)
        return t_emb


class LabelEmbedder(nn.Module):
    """
    Embeds class labels into vector representations. Also handles label dropout for classifier-free guidance.
    """
    def __init__(self, num_classes, hidden_size, dropout_prob):
        super().__init__()
        use_cfg_embedding = dropout_prob > 0
        self.embedding_table = nn.Embedding(num_classes + use_cfg_embedding, hidden_size)
        self.num_classes = num_classes
        self.dropout_prob = dropout_prob

    def token_drop(self, labels, force_drop_ids=None):
        """
        Drops labels to enable classifier-free guidance.
        """
        if force_drop_ids is None:
            drop_ids = torch.rand(labels.shape[0], device=labels.device) < self.dropout_prob
        else:
            drop_ids = force_drop_ids == 1
        labels = torch.where(drop_ids, self.num_classes, labels)
        return labels

    def forward(self, labels, train, force_drop_ids=None):
        use_dropout = self.dropout_prob > 0
        if (train and use_dropout) or (force_drop_ids is not None):
            labels = self.token_drop(labels, force_drop_ids)
        embeddings = self.embedding_table(labels)
        return embeddings


#################################################################################
#                                 Core DiT Model                                #
#################################################################################

class DiTBlock(nn.Module):
    """
    A DiT block with adaptive layer norm zero (adaLN-Zero) conditioning.
    """
    def __init__(self, hidden_size, num_heads, mlp_ratio=4.0, **block_kwargs):
        super().__init__()
        self.norm1 = nn.LayerNorm(hidden_size, elementwise_affine=False, eps=1e-6)
        self.attn = Attention(hidden_size, num_heads=num_heads, qkv_bias=True, **block_kwargs)
        self.norm2 = nn.LayerNorm(hidden_size, elementwise_affine=False, eps=1e-6)
        mlp_hidden_dim = int(hidden_size * mlp_ratio)
        approx_gelu = lambda: nn.GELU(approximate="tanh")
        self.mlp = Mlp(in_features=hidden_size, hidden_features=mlp_hidden_dim, act_layer=approx_gelu, drop=0, num_heads=num_heads)
        self.adaLN_modulation = nn.Sequential(
            nn.SiLU(),
            nn.Linear(hidden_size, 6 * hidden_size, bias=True)
        )
        self.attn_rate = DiffRate(dim=hidden_size, channel_number=num_heads)
        self.mlp_rate = DiffRate(dim=hidden_size, channel_number=num_heads)
        self.token_selection = TokenSelect(dim_in=hidden_size, num_sub_layer=1)

    def forward(self, x, c, t, complete_model):
        shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = self.adaLN_modulation(c).chunk(6, dim=1)
        
        
        if not complete_model:
            attn_weight_mask = self.attn_rate(t)
        else:
            attn_weight_mask = None
        x = x + gate_msa.unsqueeze(1) * self.attn(modulate(self.norm1(x), shift_msa, scale_msa), attn_weight_mask)
        

        if not complete_model:
            token_select, _ = self.token_selection(x)
            mlp_weight_mask = self.mlp_rate(t)
            mlp_x = gate_mlp.unsqueeze(1) * self.mlp(modulate(self.norm2(x), shift_mlp, scale_mlp), mlp_weight_mask)
            x = x + token_select * mlp_x
        else:
            token_select = None
            mlp_weight_mask = None
            x = x + gate_mlp.unsqueeze(1) * self.mlp(modulate(self.norm2(x), shift_mlp, scale_mlp))
        
    
        return x, attn_weight_mask, mlp_weight_mask, token_select


class FinalLayer(nn.Module):
    """
    The final layer of DiT.
    """
    def __init__(self, hidden_size, patch_size, out_channels):
        super().__init__()
        self.norm_final = nn.LayerNorm(hidden_size, elementwise_affine=False, eps=1e-6)
        self.linear = nn.Linear(hidden_size, patch_size * patch_size * out_channels, bias=True)
        self.adaLN_modulation = nn.Sequential(
            nn.SiLU(),
            nn.Linear(hidden_size, 2 * hidden_size, bias=True)
        )

    def forward(self, x, c):
        shift, scale = self.adaLN_modulation(c).chunk(2, dim=1)
        x = modulate(self.norm_final(x), shift, scale)
        x = self.linear(x)
        return x


class DiT(nn.Module):
    """
    Diffusion model with a Transformer backbone.
    """
    def __init__(
        self,
        input_size=32,
        patch_size=2,
        in_channels=4,
        hidden_size=1152,
        depth=28,
        num_heads=16,
        mlp_ratio=4.0,
        class_dropout_prob=0.1,
        num_classes=1000,
        learn_sigma=True,
    ):
        super().__init__()
        self.learn_sigma = learn_sigma
        self.in_channels = in_channels
        self.out_channels = in_channels * 2 if learn_sigma else in_channels
        self.patch_size = patch_size
        self.num_heads = num_heads
        self.depth = depth
        self.x_embedder = PatchEmbed(input_size, patch_size, in_channels, hidden_size, bias=True)
        self.t_embedder = TimestepEmbedder(hidden_size)
        self.y_embedder = LabelEmbedder(num_classes, hidden_size, class_dropout_prob)
        num_patches = self.x_embedder.num_patches
        # Will use fixed sin-cos embedding:
        self.pos_embed = nn.Parameter(torch.zeros(1, num_patches, hidden_size), requires_grad=False)

        self.blocks = nn.ModuleList([
            DiTBlock(hidden_size, num_heads, mlp_ratio=mlp_ratio) for _ in range(depth)
        ])
        self.final_layer = FinalLayer(hidden_size, patch_size, self.out_channels)
        self.initialize_weights()

    def initialize_weights(self):
        # Initialize transformer layers:
        def _basic_init(module):
            if isinstance(module, nn.Linear):
                torch.nn.init.xavier_uniform_(module.weight)
                if module.bias is not None:
                    nn.init.constant_(module.bias, 0)
        self.apply(_basic_init)

        # Initialize (and freeze) pos_embed by sin-cos embedding:
        pos_embed = get_2d_sincos_pos_embed(self.pos_embed.shape[-1], int(self.x_embedder.num_patches ** 0.5))
        self.pos_embed.data.copy_(torch.from_numpy(pos_embed).float().unsqueeze(0))

        # Initialize patch_embed like nn.Linear (instead of nn.Conv2d):
        w = self.x_embedder.proj.weight.data
        nn.init.xavier_uniform_(w.view([w.shape[0], -1]))
        nn.init.constant_(self.x_embedder.proj.bias, 0)

        # Initialize label embedding table:
        nn.init.normal_(self.y_embedder.embedding_table.weight, std=0.02)

        # Initialize timestep embedding MLP:
        nn.init.normal_(self.t_embedder.mlp[0].weight, std=0.02)
        nn.init.normal_(self.t_embedder.mlp[2].weight, std=0.02)

        # Zero-out adaLN modulation layers in DiT blocks:
        for block in self.blocks:
            nn.init.constant_(block.adaLN_modulation[-1].weight, 0)
            nn.init.constant_(block.adaLN_modulation[-1].bias, 0)

        # Zero-out output layers:
        nn.init.constant_(self.final_layer.adaLN_modulation[-1].weight, 0)
        nn.init.constant_(self.final_layer.adaLN_modulation[-1].bias, 0)
        nn.init.constant_(self.final_layer.linear.weight, 0)
        nn.init.constant_(self.final_layer.linear.bias, 0)

        for block in self.blocks:
            nn.init.constant_(block.attn_rate.router.bias, 15.0)
            nn.init.constant_(block.mlp_rate.router.bias, 15.0)
            nn.init.constant_(block.token_selection.mlp_head[2].bias, 35.0)



    def unpatchify(self, x):
        """
        x: (N, T, patch_size**2 * C)
        imgs: (N, H, W, C)
        """
        c = self.out_channels
        p = self.x_embedder.patch_size[0]
        h = w = int(x.shape[1] ** 0.5)
        assert h * w == x.shape[1]

        x = x.reshape(shape=(x.shape[0], h, w, p, p, c))
        x = torch.einsum('nhwpqc->nchpwq', x)
        imgs = x.reshape(shape=(x.shape[0], c, h * p, h * p))
        return imgs

    def forward(self, x, t, y, complete_model=False):
        """
        Forward pass of DiT.
        x: (N, C, H, W) tensor of spatial inputs (images or latent representations of images)
        t: (N,) tensor of diffusion timesteps
        y: (N,) tensor of class labels
        """
        x = self.x_embedder(x) + self.pos_embed  # (N, T, D), where T = H * W / patch_size ** 2
        t = self.t_embedder(t)                   # (N, D)
        y = self.y_embedder(y, self.training)    # (N, D)
        c = t + y                                # (N, D)
        token_select = []
        attn_weight_masks = []
        mlp_weight_masks = []
        for block in self.blocks:
            x, attn_weight_mask, mlp_weight_mask, token = block(x, c, t, complete_model)
            attn_weight_masks.append(attn_weight_mask)
            mlp_weight_masks.append(mlp_weight_mask)               
            token_select.append(token) 
        x = self.final_layer(x, c)                # (N, T, patch_size ** 2 * out_channels)
        x = self.unpatchify(x)                   # (N, out_channels, H, W)
        
        if not complete_model:
            attn_weight_masks = torch.stack(attn_weight_masks, dim=1)
            mlp_weight_masks = torch.stack(mlp_weight_masks, dim=1)
            token_select = torch.stack(token_select, dim=1)
            return x, attn_weight_masks, mlp_weight_masks, token_select
        else:
            return x, None, None, None

    def forward_with_cfg(self, x, t, y, cfg_scale):
        """
        Forward pass of DiT, but also batches the unconditional forward pass for classifier-free guidance.
        """
        # https://github.com/openai/glide-text2im/blob/main/notebooks/text2im.ipynb
        half = x[: len(x) // 2]
        combined = torch.cat([half, half], dim=0)
        model_out, attn_weight_masks, mlp_weight_masks, token_select  = self.forward(combined, t, y)

        print("Attn Channel", attn_weight_masks.mean())
        print("MLP Channel", mlp_weight_masks.mean())
        print("MLP token", token_select.mean())
        # For exact reproducibility reasons, we apply classifier-free guidance on only
        # three channels by default. The standard approach to cfg applies it to all channels.
        # This can be done by uncommenting the following line and commenting-out the line following that.
        # eps, rest = model_out[:, :self.in_channels], model_out[:, self.in_channels:]
        eps, rest = model_out[:, :3], model_out[:, 3:]
        cond_eps, uncond_eps = torch.split(eps, len(eps) // 2, dim=0)
        half_eps = uncond_eps + cfg_scale * (cond_eps - uncond_eps)
        eps = torch.cat([half_eps, half_eps], dim=0)
        return torch.cat([eps, rest], dim=1)


#################################################################################
#                   Sine/Cosine Positional Embedding Functions                  #
#################################################################################
# https://github.com/facebookresearch/mae/blob/main/util/pos_embed.py

def get_2d_sincos_pos_embed(embed_dim, grid_size, cls_token=False, extra_tokens=0):
    """
    grid_size: int of the grid height and width
    return:
    pos_embed: [grid_size*grid_size, embed_dim] or [1+grid_size*grid_size, embed_dim] (w/ or w/o cls_token)
    """
    grid_h = np.arange(grid_size, dtype=np.float32)
    grid_w = np.arange(grid_size, dtype=np.float32)
    grid = np.meshgrid(grid_w, grid_h)  # here w goes first
    grid = np.stack(grid, axis=0)

    grid = grid.reshape([2, 1, grid_size, grid_size])
    pos_embed = get_2d_sincos_pos_embed_from_grid(embed_dim, grid)
    if cls_token and extra_tokens > 0:
        pos_embed = np.concatenate([np.zeros([extra_tokens, embed_dim]), pos_embed], axis=0)
    return pos_embed


def get_2d_sincos_pos_embed_from_grid(embed_dim, grid):
    assert embed_dim % 2 == 0

    # use half of dimensions to encode grid_h
    emb_h = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[0])  # (H*W, D/2)
    emb_w = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[1])  # (H*W, D/2)

    emb = np.concatenate([emb_h, emb_w], axis=1) # (H*W, D)
    return emb


def get_1d_sincos_pos_embed_from_grid(embed_dim, pos):
    """
    embed_dim: output dimension for each position
    pos: a list of positions to be encoded: size (M,)
    out: (M, D)
    """
    assert embed_dim % 2 == 0
    omega = np.arange(embed_dim // 2, dtype=np.float64)
    omega /= embed_dim / 2.
    omega = 1. / 10000**omega  # (D/2,)

    pos = pos.reshape(-1)  # (M,)
    out = np.einsum('m,d->md', pos, omega)  # (M, D/2), outer product

    emb_sin = np.sin(out) # (M, D/2)
    emb_cos = np.cos(out) # (M, D/2)

    emb = np.concatenate([emb_sin, emb_cos], axis=1)  # (M, D)
    return emb


#################################################################################
#                                   DiT Configs                                  #
#################################################################################

def DiT_XL_2(**kwargs):
    return DiT(depth=28, hidden_size=1152, patch_size=2, num_heads=16, **kwargs)

def DiT_XL_4(**kwargs):
    return DiT(depth=28, hidden_size=1152, patch_size=4, num_heads=16, **kwargs)

def DiT_XL_8(**kwargs):
    return DiT(depth=28, hidden_size=1152, patch_size=8, num_heads=16, **kwargs)

def DiT_L_2(**kwargs):
    return DiT(depth=24, hidden_size=1024, patch_size=2, num_heads=16, **kwargs)

def DiT_L_4(**kwargs):
    return DiT(depth=24, hidden_size=1024, patch_size=4, num_heads=16, **kwargs)

def DiT_L_8(**kwargs):
    return DiT(depth=24, hidden_size=1024, patch_size=8, num_heads=16, **kwargs)

def DiT_B_2(**kwargs):
    return DiT(depth=12, hidden_size=768, patch_size=2, num_heads=12, **kwargs)

def DiT_B_4(**kwargs):
    return DiT(depth=12, hidden_size=768, patch_size=4, num_heads=12, **kwargs)

def DiT_B_8(**kwargs):
    return DiT(depth=12, hidden_size=768, patch_size=8, num_heads=12, **kwargs)

def DiT_S_2(**kwargs):
    return DiT(depth=12, hidden_size=384, patch_size=2, num_heads=6, **kwargs)

def DiT_S_4(**kwargs):
    return DiT(depth=12, hidden_size=384, patch_size=4, num_heads=6, **kwargs)

def DiT_S_8(**kwargs):
    return DiT(depth=12, hidden_size=384, patch_size=8, num_heads=6, **kwargs)


DiT_models = {
    'DiT-XL/2': DiT_XL_2,  'DiT-XL/4': DiT_XL_4,  'DiT-XL/8': DiT_XL_8,
    'DiT-L/2':  DiT_L_2,   'DiT-L/4':  DiT_L_4,   'DiT-L/8':  DiT_L_8,
    'DiT-B/2':  DiT_B_2,   'DiT-B/4':  DiT_B_4,   'DiT-B/8':  DiT_B_8,
    'DiT-S/2':  DiT_S_2,   'DiT-S/4':  DiT_S_4,   'DiT-S/8':  DiT_S_8,
}