Vendor in the single module used from taming_transformers; remove taming_transformers dependency

from torch.optim.lr_scheduler import LambdaLR

from ldm.modules.ema import LitEma

from taming.modules.vqvae.quantize import VectorQuantizer2 as VectorQuantizer

from vqvae_quantize import VectorQuantizer2 as VectorQuantizer

from ldm.modules.diffusionmodules.model import Encoder, Decoder

from ldm.util import instantiate_from_config

# Vendored from https://raw.githubusercontent.com/CompVis/taming-transformers/24268930bf1dce879235a7fddd0b2355b84d7ea6/taming/modules/vqvae/quantize.py,

# where the license is as follows:

#

# Copyright (c) 2020 Patrick Esser and Robin Rombach and Björn Ommer

#

# Permission is hereby granted, free of charge, to any person obtaining a copy

# of this software and associated documentation files (the "Software"), to deal

# in the Software without restriction, including without limitation the rights

# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell

# copies of the Software, and to permit persons to whom the Software is

# furnished to do so, subject to the following conditions:

#

# The above copyright notice and this permission notice shall be included in all

# copies or substantial portions of the Software.

#

# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,

# EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF

# MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.

# IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,

# DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR

# OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE

# OR OTHER DEALINGS IN THE SOFTWARE./

import torch

import torch.nn as nn

import numpy as np

from einops import rearrange

class VectorQuantizer2(nn.Module):

    """

    Improved version over VectorQuantizer, can be used as a drop-in replacement. Mostly

    avoids costly matrix multiplications and allows for post-hoc remapping of indices.

    """

    # NOTE: due to a bug the beta term was applied to the wrong term. for

    # backwards compatibility we use the buggy version by default, but you can

    # specify legacy=False to fix it.

    def __init__(self, n_e, e_dim, beta, remap=None, unknown_index="random",

                 sane_index_shape=False, legacy=True):

        super().__init__()

        self.n_e = n_e

        self.e_dim = e_dim

        self.beta = beta

        self.legacy = legacy

        self.embedding = nn.Embedding(self.n_e, self.e_dim)

        self.embedding.weight.data.uniform_(-1.0 / self.n_e, 1.0 / self.n_e)

        self.remap = remap

        if self.remap is not None:

            self.register_buffer("used", torch.tensor(np.load(self.remap)))

            self.re_embed = self.used.shape[0]

            self.unknown_index = unknown_index  # "random" or "extra" or integer

            if self.unknown_index == "extra":

                self.unknown_index = self.re_embed

                self.re_embed = self.re_embed + 1

            print(f"Remapping {self.n_e} indices to {self.re_embed} indices. "

                  f"Using {self.unknown_index} for unknown indices.")

        else:

            self.re_embed = n_e

        self.sane_index_shape = sane_index_shape

    def remap_to_used(self, inds):

        ishape = inds.shape

        assert len(ishape) > 1

        inds = inds.reshape(ishape[0], -1)

        used = self.used.to(inds)

        match = (inds[:, :, None] == used[None, None, ...]).long()

        new = match.argmax(-1)

        unknown = match.sum(2) < 1

        if self.unknown_index == "random":

            new[unknown] = torch.randint(0, self.re_embed, size=new[unknown].shape).to(device=new.device)

        else:

            new[unknown] = self.unknown_index

        return new.reshape(ishape)

    def unmap_to_all(self, inds):

        ishape = inds.shape

        assert len(ishape) > 1

        inds = inds.reshape(ishape[0], -1)

        used = self.used.to(inds)

        if self.re_embed > self.used.shape[0]:  # extra token

            inds[inds >= self.used.shape[0]] = 0  # simply set to zero

        back = torch.gather(used[None, :][inds.shape[0] * [0], :], 1, inds)

        return back.reshape(ishape)

    def forward(self, z, temp=None, rescale_logits=False, return_logits=False):

        assert temp is None or temp == 1.0, "Only for interface compatible with Gumbel"

        assert rescale_logits is False, "Only for interface compatible with Gumbel"

        assert return_logits is False, "Only for interface compatible with Gumbel"

        # reshape z -> (batch, height, width, channel) and flatten

        z = rearrange(z, 'b c h w -> b h w c').contiguous()

        z_flattened = z.view(-1, self.e_dim)

        # distances from z to embeddings e_j (z - e)^2 = z^2 + e^2 - 2 e * z

        d = torch.sum(z_flattened ** 2, dim=1, keepdim=True) + \

            torch.sum(self.embedding.weight ** 2, dim=1) - 2 * \

            torch.einsum('bd,dn->bn', z_flattened, rearrange(self.embedding.weight, 'n d -> d n'))

        min_encoding_indices = torch.argmin(d, dim=1)

        z_q = self.embedding(min_encoding_indices).view(z.shape)

        perplexity = None

        min_encodings = None

        # compute loss for embedding

        if not self.legacy:

            loss = self.beta * torch.mean((z_q.detach() - z) ** 2) + \

                   torch.mean((z_q - z.detach()) ** 2)

        else:

            loss = torch.mean((z_q.detach() - z) ** 2) + self.beta * \

                   torch.mean((z_q - z.detach()) ** 2)

        # preserve gradients

        z_q = z + (z_q - z).detach()

        # reshape back to match original input shape

        z_q = rearrange(z_q, 'b h w c -> b c h w').contiguous()

        if self.remap is not None:

            min_encoding_indices = min_encoding_indices.reshape(z.shape[0], -1)  # add batch axis

            min_encoding_indices = self.remap_to_used(min_encoding_indices)

            min_encoding_indices = min_encoding_indices.reshape(-1, 1)  # flatten

        if self.sane_index_shape:

            min_encoding_indices = min_encoding_indices.reshape(

                z_q.shape[0], z_q.shape[2], z_q.shape[3])

        return z_q, loss, (perplexity, min_encodings, min_encoding_indices)

    def get_codebook_entry(self, indices, shape):

        # shape specifying (batch, height, width, channel)

        if self.remap is not None:

            indices = indices.reshape(shape[0], -1)  # add batch axis

            indices = self.unmap_to_all(indices)

            indices = indices.reshape(-1)  # flatten again

        # get quantized latent vectors

        z_q = self.embedding(indices)

        if shape is not None:

            z_q = z_q.view(shape)

            # reshape back to match original input shape

            z_q = z_q.permute(0, 3, 1, 2).contiguous()

        return z_q

    openclip_package = os.environ.get('OPENCLIP_PACKAGE', "https://github.com/mlfoundations/open_clip/archive/bb6e834e9c70d9c27d0dc3ecedeebeaeb1ffad6b.zip")

    stable_diffusion_repo = os.environ.get('STABLE_DIFFUSION_REPO', "https://github.com/Stability-AI/stablediffusion.git")

    taming_transformers_repo = os.environ.get('TAMING_TRANSFORMERS_REPO', "https://github.com/CompVis/taming-transformers.git")

    k_diffusion_repo = os.environ.get('K_DIFFUSION_REPO', 'https://github.com/crowsonkb/k-diffusion.git')

    codeformer_repo = os.environ.get('CODEFORMER_REPO', 'https://github.com/sczhou/CodeFormer.git')

    blip_repo = os.environ.get('BLIP_REPO', 'https://github.com/salesforce/BLIP.git')

    stable_diffusion_commit_hash = os.environ.get('STABLE_DIFFUSION_COMMIT_HASH', "cf1d67a6fd5ea1aa600c4df58e5b47da45f6bdbf")

    taming_transformers_commit_hash = os.environ.get('TAMING_TRANSFORMERS_COMMIT_HASH', "24268930bf1dce879235a7fddd0b2355b84d7ea6")

    k_diffusion_commit_hash = os.environ.get('K_DIFFUSION_COMMIT_HASH', "c9fe758757e022f05ca5a53fa8fac28889e4f1cf")

    codeformer_commit_hash = os.environ.get('CODEFORMER_COMMIT_HASH', "c5b4593074ba6214284d6acd5f1719b6c5d739af")

    blip_commit_hash = os.environ.get('BLIP_COMMIT_HASH', "48211a1594f1321b00f14c9f7a5b4813144b2fb9")

    os.makedirs(os.path.join(script_path, dir_repos), exist_ok=True)

    git_clone(stable_diffusion_repo, repo_dir('stable-diffusion-stability-ai'), "Stable Diffusion", stable_diffusion_commit_hash)

    git_clone(taming_transformers_repo, repo_dir('taming-transformers'), "Taming Transformers", taming_transformers_commit_hash)

    git_clone(k_diffusion_repo, repo_dir('k-diffusion'), "K-diffusion", k_diffusion_commit_hash)

    git_clone(codeformer_repo, repo_dir('CodeFormer'), "CodeFormer", codeformer_commit_hash)

    git_clone(blip_repo, repo_dir('BLIP'), "BLIP", blip_commit_hash)

path_dirs = [

    (sd_path, 'ldm', 'Stable Diffusion', []),

    (os.path.join(sd_path, '../taming-transformers'), 'taming', 'Taming Transformers', []),

    (os.path.join(sd_path, '../CodeFormer'), 'inference_codeformer.py', 'CodeFormer', []),

    (os.path.join(sd_path, '../BLIP'), 'models/blip.py', 'BLIP', []),

    (os.path.join(sd_path, '../k-diffusion'), 'k_diffusion/sampling.py', 'k_diffusion', ["atstart"]),

# Fixed git commits

#export STABLE_DIFFUSION_COMMIT_HASH=""

#export TAMING_TRANSFORMERS_COMMIT_HASH=""

#export CODEFORMER_COMMIT_HASH=""

#export BLIP_COMMIT_HASH=""