models/attention_processor.py

# Copyright 2023 The HuggingFace Team. All rights reserved.
#
# 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.
import warnings
from typing import Callable, Optional, Union

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

from diffusers.utils import deprecate, logging, maybe_allow_in_graph

logger = logging.get_logger(__name__)  # pylint: disable=invalid-name

@maybe_allow_in_graph
class Attention(nn.Module):
    r"""
    A cross attention layer.

    Parameters:
        query_dim (`int`): The number of channels in the query.
        cross_attention_dim (`int`, *optional*):
            The number of channels in the encoder_hidden_states. If not given, defaults to `query_dim`.
        heads (`int`,  *optional*, defaults to 8): The number of heads to use for multi-head attention.
        dim_head (`int`,  *optional*, defaults to 64): The number of channels in each head.
        dropout (`float`, *optional*, defaults to 0.0): The dropout probability to use.
        bias (`bool`, *optional*, defaults to False):
            Set to `True` for the query, key, and value linear layers to contain a bias parameter.
    """

    def __init__(
        self,
        query_dim: int,
        cross_attention_dim: Optional[int] = None,
        heads: int = 8,
        dim_head: int = 64,
        dropout: float = 0.0,
        bias=False,
        upcast_attention: bool = False,
        upcast_softmax: bool = False,
        cross_attention_norm: Optional[str] = None,
        cross_attention_norm_num_groups: int = 32,
        added_kv_proj_dim: Optional[int] = None,
        norm_num_groups: Optional[int] = None,
        spatial_norm_dim: Optional[int] = None,
        out_bias: bool = True,
        scale_qk: bool = True,
        only_cross_attention: bool = False,
        eps: float = 1e-5,
        rescale_output_factor: float = 1.0,
        residual_connection: bool = False,
        _from_deprecated_attn_block=False,
        processor: Optional["AttnProcessor"] = None,
    ):
        super().__init__()
        inner_dim = dim_head * heads
        cross_attention_dim = cross_attention_dim if cross_attention_dim is not None else query_dim
        self.upcast_attention = upcast_attention
        self.upcast_softmax = upcast_softmax
        self.rescale_output_factor = rescale_output_factor
        self.residual_connection = residual_connection

        # we make use of this private variable to know whether this class is loaded
        # with an deprecated state dict so that we can convert it on the fly
        self._from_deprecated_attn_block = _from_deprecated_attn_block

        self.scale_qk = scale_qk
        self.scale = dim_head**-0.5 if self.scale_qk else 1.0

        self.heads = heads
        # for slice_size > 0 the attention score computation
        # is split across the batch axis to save memory
        # You can set slice_size with `set_attention_slice`
        self.sliceable_head_dim = heads

        self.added_kv_proj_dim = added_kv_proj_dim
        self.only_cross_attention = only_cross_attention

        if self.added_kv_proj_dim is None and self.only_cross_attention:
            raise ValueError(
                "`only_cross_attention` can only be set to True if `added_kv_proj_dim` is not None. Make sure to set either `only_cross_attention=False` or define `added_kv_proj_dim`."
            )

        if norm_num_groups is not None:
            self.group_norm = nn.GroupNorm(num_channels=query_dim, num_groups=norm_num_groups, eps=eps, affine=True)
        else:
            self.group_norm = None

        if spatial_norm_dim is not None:
            self.spatial_norm = SpatialNorm(f_channels=query_dim, zq_channels=spatial_norm_dim)
        else:
            self.spatial_norm = None

        if cross_attention_norm is None:
            self.norm_cross = None
        elif cross_attention_norm == "layer_norm":
            self.norm_cross = nn.LayerNorm(cross_attention_dim)
        elif cross_attention_norm == "group_norm":
            if self.added_kv_proj_dim is not None:
                # The given `encoder_hidden_states` are initially of shape
                # (batch_size, seq_len, added_kv_proj_dim) before being projected
                # to (batch_size, seq_len, cross_attention_dim). The norm is applied
                # before the projection, so we need to use `added_kv_proj_dim` as
                # the number of channels for the group norm.
                norm_cross_num_channels = added_kv_proj_dim
            else:
                norm_cross_num_channels = cross_attention_dim

            self.norm_cross = nn.GroupNorm(
                num_channels=norm_cross_num_channels, num_groups=cross_attention_norm_num_groups, eps=1e-5, affine=True
            )
        else:
            raise ValueError(
                f"unknown cross_attention_norm: {cross_attention_norm}. Should be None, 'layer_norm' or 'group_norm'"
            )

        self.to_q = nn.Linear(query_dim, inner_dim, bias=bias)

        if not self.only_cross_attention:
            # only relevant for the `AddedKVProcessor` classes
            self.to_k = nn.Linear(cross_attention_dim, inner_dim, bias=bias)
            self.to_v = nn.Linear(cross_attention_dim, inner_dim, bias=bias)
        else:
            self.to_k = None
            self.to_v = None

        if self.added_kv_proj_dim is not None:
            self.add_k_proj = nn.Linear(added_kv_proj_dim, inner_dim)
            self.add_v_proj = nn.Linear(added_kv_proj_dim, inner_dim)

        self.to_out = nn.ModuleList([])
        self.to_out.append(nn.Linear(inner_dim, query_dim, bias=out_bias))
        self.to_out.append(nn.Dropout(dropout))

        # set attention processor
        # We use the AttnProcessor2_0 by default when torch 2.x is used which uses
        # torch.nn.functional.scaled_dot_product_attention for native Flash/memory_efficient_attention
        # but only if it has the default `scale` argument. TODO remove scale_qk check when we move to torch 2.1
        if processor is None:
            # processor = (
            #     AttnProcessor2_0() if hasattr(F, "scaled_dot_product_attention") and self.scale_qk else AttnProcessor()
            # )
            # Note: efficient attention is not used. We can use efficient attention to speed up.
            processor = AttnProcessor()
        self.set_processor(processor)

    def set_processor(self, processor: "AttnProcessor"):
        # if current processor is in `self._modules` and if passed `processor` is not, we need to
        # pop `processor` from `self._modules`
        if (
            hasattr(self, "processor")
            and isinstance(self.processor, torch.nn.Module)
            and not isinstance(processor, torch.nn.Module)
        ):
            logger.info(f"You are removing possibly trained weights of {self.processor} with {processor}")
            self._modules.pop("processor")

        self.processor = processor

    def forward(self, hidden_states, encoder_hidden_states=None, attention_mask=None, return_attntion_probs=False, **cross_attention_kwargs):
        # The `Attention` class can call different attention processors / attention functions
        # here we simply pass along all tensors to the selected processor class
        # For standard processors that are defined here, `**cross_attention_kwargs` is empty
        return self.processor(
            self,
            hidden_states,
            encoder_hidden_states=encoder_hidden_states,
            attention_mask=attention_mask,
            return_attntion_probs=return_attntion_probs,
            **cross_attention_kwargs,
        )

    def batch_to_head_dim(self, tensor):
        head_size = self.heads
        batch_size, seq_len, dim = tensor.shape
        tensor = tensor.reshape(batch_size // head_size, head_size, seq_len, dim)
        tensor = tensor.permute(0, 2, 1, 3).reshape(batch_size // head_size, seq_len, dim * head_size)
        return tensor

    def head_to_batch_dim(self, tensor, out_dim=3):
        head_size = self.heads
        batch_size, seq_len, dim = tensor.shape
        tensor = tensor.reshape(batch_size, seq_len, head_size, dim // head_size)
        tensor = tensor.permute(0, 2, 1, 3)

        if out_dim == 3:
            tensor = tensor.reshape(batch_size * head_size, seq_len, dim // head_size)

        return tensor

    def get_attention_scores(self, query, key, attention_mask=None):
        dtype = query.dtype
        if self.upcast_attention:
            query = query.float()
            key = key.float()

        if attention_mask is None:
            baddbmm_input = torch.empty(
                query.shape[0], query.shape[1], key.shape[1], dtype=query.dtype, device=query.device
            )
            beta = 0
        else:
            baddbmm_input = attention_mask
            beta = 1

        attention_scores = torch.baddbmm(
            baddbmm_input,
            query,
            key.transpose(-1, -2),
            beta=beta,
            alpha=self.scale,
        )
        del baddbmm_input

        if self.upcast_softmax:
            attention_scores = attention_scores.float()

        attention_probs = attention_scores.softmax(dim=-1)
        del attention_scores

        attention_probs = attention_probs.to(dtype)

        return attention_probs

    def prepare_attention_mask(self, attention_mask, target_length, batch_size=None, out_dim=3):
        if batch_size is None:
            deprecate(
                "batch_size=None",
                "0.0.15",
                (
                    "Not passing the `batch_size` parameter to `prepare_attention_mask` can lead to incorrect"
                    " attention mask preparation and is deprecated behavior. Please make sure to pass `batch_size` to"
                    " `prepare_attention_mask` when preparing the attention_mask."
                ),
            )
            batch_size = 1

        head_size = self.heads
        if attention_mask is None:
            return attention_mask

        current_length: int = attention_mask.shape[-1]
        if current_length != target_length:
            if attention_mask.device.type == "mps":
                # HACK: MPS: Does not support padding by greater than dimension of input tensor.
                # Instead, we can manually construct the padding tensor.
                padding_shape = (attention_mask.shape[0], attention_mask.shape[1], target_length)
                padding = torch.zeros(padding_shape, dtype=attention_mask.dtype, device=attention_mask.device)
                attention_mask = torch.cat([attention_mask, padding], dim=2)
            else:
                # TODO: for pipelines such as stable-diffusion, padding cross-attn mask:
                #       we want to instead pad by (0, remaining_length), where remaining_length is:
                #       remaining_length: int = target_length - current_length
                # TODO: re-enable tests/models/test_models_unet_2d_condition.py#test_model_xattn_padding
                attention_mask = F.pad(attention_mask, (0, target_length), value=0.0)

        if out_dim == 3:
            if attention_mask.shape[0] < batch_size * head_size:
                attention_mask = attention_mask.repeat_interleave(head_size, dim=0)
        elif out_dim == 4:
            attention_mask = attention_mask.unsqueeze(1)
            attention_mask = attention_mask.repeat_interleave(head_size, dim=1)

        return attention_mask

    def norm_encoder_hidden_states(self, encoder_hidden_states):
        assert self.norm_cross is not None, "self.norm_cross must be defined to call self.norm_encoder_hidden_states"

        if isinstance(self.norm_cross, nn.LayerNorm):
            encoder_hidden_states = self.norm_cross(encoder_hidden_states)
        elif isinstance(self.norm_cross, nn.GroupNorm):
            # Group norm norms along the channels dimension and expects
            # input to be in the shape of (N, C, *). In this case, we want
            # to norm along the hidden dimension, so we need to move
            # (batch_size, sequence_length, hidden_size) ->
            # (batch_size, hidden_size, sequence_length)
            encoder_hidden_states = encoder_hidden_states.transpose(1, 2)
            encoder_hidden_states = self.norm_cross(encoder_hidden_states)
            encoder_hidden_states = encoder_hidden_states.transpose(1, 2)
        else:
            assert False

        return encoder_hidden_states


class AttnProcessor:
    r"""
    Processor for implementing scaled dot-product attention (enabled by default if you're using PyTorch 2.0).
    """

    def __init__(self):
        if not hasattr(F, "scaled_dot_product_attention"):
            raise ImportError("AttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.")

    def __call_fast__(
        self,
        attn: Attention,
        hidden_states,
        encoder_hidden_states=None,
        attention_mask=None,
        temb=None,
    ):
        residual = hidden_states

        if attn.spatial_norm is not None:
            hidden_states = attn.spatial_norm(hidden_states, temb)

        input_ndim = hidden_states.ndim

        if input_ndim == 4:
            batch_size, channel, height, width = hidden_states.shape
            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)

        batch_size, sequence_length, _ = (
            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
        )
        inner_dim = hidden_states.shape[-1]

        if attention_mask is not None:
            attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
            # scaled_dot_product_attention expects attention_mask shape to be
            # (batch, heads, source_length, target_length)
            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])

        if attn.group_norm is not None:
            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)

        query = attn.to_q(hidden_states)

        if encoder_hidden_states is None:
            encoder_hidden_states = hidden_states
        elif attn.norm_cross:
            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)

        key = attn.to_k(encoder_hidden_states)
        value = attn.to_v(encoder_hidden_states)

        head_dim = inner_dim // attn.heads
        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

        # the output of sdp = (batch, num_heads, seq_len, head_dim)
        # TODO: add support for attn.scale when we move to Torch 2.1
        hidden_states = F.scaled_dot_product_attention(
            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
        )

        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
        hidden_states = hidden_states.to(query.dtype)

        # linear proj
        hidden_states = attn.to_out[0](hidden_states)
        # dropout
        hidden_states = attn.to_out[1](hidden_states)

        if input_ndim == 4:
            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)

        if attn.residual_connection:
            hidden_states = hidden_states + residual

        hidden_states = hidden_states / attn.rescale_output_factor

        return hidden_states

    def __call__(
        self,
        attn: Attention,
        hidden_states,
        encoder_hidden_states=None,
        attention_mask=None,
        temb=None,
        return_attntion_probs=False,
        attn_key=None,
        attn_process_fn=None,
        return_cond_ca_only=False,
        return_token_ca_only=None,
        offload_cross_attn_to_cpu=False,
        save_attn_to_dict=None,
        save_keys=None,
        enable_flash_attn=True,
    ):
        """
        attn_key: current key (a tuple of hierarchy index (up/mid/down, stage id, block id, sub-block id), sub block id should always be 0 in SD UNet)
        save_attn_to_dict: pass in a dict to save to dict
        """
        cross_attn = encoder_hidden_states is not None
        
        if (not cross_attn) or (
            (attn_process_fn is None) 
            and not (save_attn_to_dict is not None and (save_keys is None or (tuple(attn_key) in save_keys))) 
            and not return_attntion_probs):
            with torch.backends.cuda.sdp_kernel(enable_flash=enable_flash_attn, enable_math=True, enable_mem_efficient=enable_flash_attn):
                return self.__call_fast__(attn, hidden_states, encoder_hidden_states, attention_mask, temb)
        
        residual = hidden_states

        if attn.spatial_norm is not None:
            hidden_states = attn.spatial_norm(hidden_states, temb)

        input_ndim = hidden_states.ndim

        if input_ndim == 4:
            batch_size, channel, height, width = hidden_states.shape
            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)

        batch_size, sequence_length, _ = (
            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
        )
        attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)

        if attn.group_norm is not None:
            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)

        query = attn.to_q(hidden_states)

        if encoder_hidden_states is None:
            encoder_hidden_states = hidden_states
        elif attn.norm_cross:
            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)

        key = attn.to_k(encoder_hidden_states)
        value = attn.to_v(encoder_hidden_states)

        query = attn.head_to_batch_dim(query)
        key = attn.head_to_batch_dim(key)
        value = attn.head_to_batch_dim(value)

        attention_probs = attn.get_attention_scores(query, key, attention_mask)
        # Currently only process cross-attention
        if attn_process_fn is not None and cross_attn:
            attention_probs_before_process = attention_probs.clone()
            attention_probs = attn_process_fn(attention_probs, query, key, value, attn_key=attn_key, cross_attn=cross_attn, batch_size=batch_size, heads=attn.heads)
        else:
            attention_probs_before_process = attention_probs
        hidden_states = torch.bmm(attention_probs, value)
        hidden_states = attn.batch_to_head_dim(hidden_states)

        # linear proj
        hidden_states = attn.to_out[0](hidden_states)
        # dropout
        hidden_states = attn.to_out[1](hidden_states)

        if input_ndim == 4:
            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)

        if attn.residual_connection:
            hidden_states = hidden_states + residual

        hidden_states = hidden_states / attn.rescale_output_factor

        if return_attntion_probs or save_attn_to_dict is not None:
            # Recover batch dimension: (batch_size, heads, flattened_2d, text_tokens)
            attention_probs_unflattened = attention_probs_before_process.unflatten(dim=0, sizes=(batch_size, attn.heads))
            if return_token_ca_only is not None:
                # (batch size, n heads, 2d dimension, num text tokens)
                if isinstance(return_token_ca_only, int):
                    # return_token_ca_only: an integer
                    attention_probs_unflattened = attention_probs_unflattened[:, :, :, return_token_ca_only:return_token_ca_only+1]
                else:
                    # return_token_ca_only: A 1d index tensor
                    attention_probs_unflattened = attention_probs_unflattened[:, :, :, return_token_ca_only]
            if return_cond_ca_only:
                assert batch_size % 2 == 0, f"Samples are not in pairs: {batch_size} samples"
                attention_probs_unflattened = attention_probs_unflattened[batch_size // 2:]
            if offload_cross_attn_to_cpu:
                attention_probs_unflattened = attention_probs_unflattened.cpu()
            if save_attn_to_dict is not None and (save_keys is None or (tuple(attn_key) in save_keys)):
                save_attn_to_dict[tuple(attn_key)] = attention_probs_unflattened
            if return_attntion_probs:
                return hidden_states, attention_probs_unflattened
        return hidden_states

# For typing
AttentionProcessor = AttnProcessor

class SpatialNorm(nn.Module):
    """
    Spatially conditioned normalization as defined in https://arxiv.org/abs/2209.09002
    """

    def __init__(
        self,
        f_channels,
        zq_channels,
    ):
        super().__init__()
        self.norm_layer = nn.GroupNorm(num_channels=f_channels, num_groups=32, eps=1e-6, affine=True)
        self.conv_y = nn.Conv2d(zq_channels, f_channels, kernel_size=1, stride=1, padding=0)
        self.conv_b = nn.Conv2d(zq_channels, f_channels, kernel_size=1, stride=1, padding=0)

    def forward(self, f, zq):
        f_size = f.shape[-2:]
        zq = F.interpolate(zq, size=f_size, mode="nearest")
        norm_f = self.norm_layer(f)
        new_f = norm_f * self.conv_y(zq) + self.conv_b(zq)
        return new_f