Upload 9 files

cldm/cldm.py

@@ -0,0 +1,470 @@
+import einops
+import torch
+import torch as th
+import torch.nn as nn
+
+from ldm.modules.diffusionmodules.util import (
+    conv_nd,
+    linear,
+    zero_module,
+    timestep_embedding,
+)
+
+from einops import rearrange, repeat
+from torchvision.utils import make_grid
+from ldm.modules.attention import SpatialTransformer
+from ldm.modules.diffusionmodules.openaimodel import UNetModel, TimestepEmbedSequential, ResBlock, Downsample, AttentionBlock
+from ldm.models.diffusion.ddpm import LatentDiffusion
+from ldm.util import log_txt_as_img, exists, instantiate_from_config
+from ldm.models.diffusion.ddim import DDIMSampler
+
+
+class ControlledUnetModel(UNetModel):
+    def forward(self, x, timesteps=None, context=None, control=None, only_mid_control=False, **kwargs):
+        hs = []
+        with torch.no_grad():
+            t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False)
+            emb = self.time_embed(t_emb)
+            h = x.type(self.dtype)
+            for module in self.input_blocks:
+                h = module(h, emb, context)
+                hs.append(h)
+            h = self.middle_block(h, emb, context)
+
+        if control is not None:
+            h += control.pop()
+
+        for i, module in enumerate(self.output_blocks):
+            if only_mid_control or control is None:
+                h = torch.cat([h, hs.pop()], dim=1)
+            else:
+                h = torch.cat([h, hs.pop() + control.pop()], dim=1)
+            h = module(h, emb, context)
+
+        h = h.type(x.dtype)
+        return self.out(h)
+
+
+class ControlNet(nn.Module):
+    def __init__(
+            self,
+            image_size,
+            in_channels,
+            model_channels,
+            hint_channels,
+            num_res_blocks,
+            attention_resolutions,
+            dropout=0,
+            channel_mult=(1, 2, 4, 8),
+            conv_resample=True,
+            dims=2,
+            use_checkpoint=False,
+            use_fp16=False,
+            num_heads=-1,
+            num_head_channels=-1,
+            num_heads_upsample=-1,
+            use_scale_shift_norm=False,
+            resblock_updown=False,
+            use_new_attention_order=False,
+            use_spatial_transformer=False,  # custom transformer support
+            transformer_depth=1,  # custom transformer support
+            context_dim=None,  # custom transformer support
+            n_embed=None,  # custom support for prediction of discrete ids into codebook of first stage vq model
+            legacy=True,
+            disable_self_attentions=None,
+            num_attention_blocks=None,
+            disable_middle_self_attn=False,
+            use_linear_in_transformer=False,
+            latent_control=False,
+            noise_control=False
+    ):
+        super().__init__()
+        if use_spatial_transformer:
+            assert context_dim is not None, 'Fool!! You forgot to include the dimension of your cross-attention conditioning...'
+
+        if context_dim is not None:
+            assert use_spatial_transformer, 'Fool!! You forgot to use the spatial transformer for your cross-attention conditioning...'
+            from omegaconf.listconfig import ListConfig
+            if type(context_dim) == ListConfig:
+                context_dim = list(context_dim)
+
+        if num_heads_upsample == -1:
+            num_heads_upsample = num_heads
+
+        if num_heads == -1:
+            assert num_head_channels != -1, 'Either num_heads or num_head_channels has to be set'
+
+        if num_head_channels == -1:
+            assert num_heads != -1, 'Either num_heads or num_head_channels has to be set'
+
+        self.dims = dims
+        self.image_size = image_size
+        self.in_channels = in_channels
+        self.model_channels = model_channels
+        if isinstance(num_res_blocks, int):
+            self.num_res_blocks = len(channel_mult) * [num_res_blocks]
+        else:
+            if len(num_res_blocks) != len(channel_mult):
+                raise ValueError("provide num_res_blocks either as an int (globally constant) or "
+                                 "as a list/tuple (per-level) with the same length as channel_mult")
+            self.num_res_blocks = num_res_blocks
+        if disable_self_attentions is not None:
+            # should be a list of booleans, indicating whether to disable self-attention in TransformerBlocks or not
+            assert len(disable_self_attentions) == len(channel_mult)
+        if num_attention_blocks is not None:
+            assert len(num_attention_blocks) == len(self.num_res_blocks)
+            assert all(map(lambda i: self.num_res_blocks[i] >= num_attention_blocks[i], range(len(num_attention_blocks))))
+            print(f"Constructor of UNetModel received num_attention_blocks={num_attention_blocks}. "
+                  f"This option has LESS priority than attention_resolutions {attention_resolutions}, "
+                  f"i.e., in cases where num_attention_blocks[i] > 0 but 2**i not in attention_resolutions, "
+                  f"attention will still not be set.")
+
+        self.attention_resolutions = attention_resolutions
+        self.dropout = dropout
+        self.channel_mult = channel_mult
+        self.conv_resample = conv_resample
+        self.use_checkpoint = use_checkpoint
+        self.dtype = th.float16 if use_fp16 else th.float32
+        self.num_heads = num_heads
+        self.num_head_channels = num_head_channels
+        self.num_heads_upsample = num_heads_upsample
+        self.predict_codebook_ids = n_embed is not None
+
+        time_embed_dim = model_channels * 4
+        self.time_embed = nn.Sequential(
+            linear(model_channels, time_embed_dim),
+            nn.SiLU(),
+            linear(time_embed_dim, time_embed_dim),
+        )
+
+        self.input_blocks = nn.ModuleList(
+            [
+                TimestepEmbedSequential(
+                    conv_nd(dims, in_channels, model_channels, 3, padding=1)
+                )
+            ]
+        )
+        self.latent_control = latent_control
+        self.noise_control = noise_control
+        self.zero_convs = nn.ModuleList([self.make_zero_conv(model_channels)])
+        if self.latent_control:
+            self.input_hint_block = TimestepEmbedSequential(
+                    zero_module(conv_nd(dims, hint_channels, model_channels, 3, padding=1))
+                )
+        else:
+            # 输入段，这一段参数都是需要训练的，要改成隐码输入，就要改变结构，这也是原设计中控制分支里面和主网络不对称的结构
+            # 原来的controlNet中，不管输入输出的图像是多大，control 都是 256x256
+            self.input_hint_block = TimestepEmbedSequential(
+                conv_nd(dims, hint_channels, 16, 3, padding=1),
+                nn.SiLU(),
+                conv_nd(dims, 16, 16, 3, padding=1),
+                nn.SiLU(),
+                conv_nd(dims, 16, 32, 3, padding=1, stride=2),  # 256 -> 128
+                nn.SiLU(),
+                conv_nd(dims, 32, 32, 3, padding=1),
+                nn.SiLU(),
+                conv_nd(dims, 32, 96, 3, padding=1, stride=2),  # 128 -> 64
+                nn.SiLU(),
+                conv_nd(dims, 96, 96, 3, padding=1),
+                nn.SiLU(),
+                conv_nd(dims, 96, 256, 3, padding=1, stride=2),  # 64 -> 32
+                nn.SiLU(),
+                zero_module(conv_nd(dims, 256, model_channels, 3, padding=1))
+            )
+
+        self._feature_size = model_channels
+        input_block_chans = [model_channels]
+        ch = model_channels
+        ds = 1
+        for level, mult in enumerate(channel_mult):
+            for nr in range(self.num_res_blocks[level]):
+                layers = [
+                    ResBlock(
+                        ch,
+                        time_embed_dim,
+                        dropout,
+                        out_channels=mult * model_channels,
+                        dims=dims,
+                        use_checkpoint=use_checkpoint,
+                        use_scale_shift_norm=use_scale_shift_norm,
+                    )
+                ]
+                ch = mult * model_channels
+                if ds in attention_resolutions:
+                    if num_head_channels == -1:
+                        dim_head = ch // num_heads
+                    else:
+                        num_heads = ch // num_head_channels
+                        dim_head = num_head_channels
+                    if legacy:
+                        # num_heads = 1
+                        dim_head = ch // num_heads if use_spatial_transformer else num_head_channels  # use_spatial_transformer=True
+                    if exists(disable_self_attentions):
+                        disabled_sa = disable_self_attentions[level]
+                    else:
+                        disabled_sa = False
+
+                    if not exists(num_attention_blocks) or nr < num_attention_blocks[level]:
+                        layers.append(
+                            AttentionBlock(
+                                ch,
+                                use_checkpoint=use_checkpoint,
+                                num_heads=num_heads,
+                                num_head_channels=dim_head,
+                                use_new_attention_order=use_new_attention_order,
+                            ) if not use_spatial_transformer else SpatialTransformer(
+                                ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim,
+                                disable_self_attn=disabled_sa, use_linear=use_linear_in_transformer,
+                                use_checkpoint=use_checkpoint
+                            )
+                        )
+                self.input_blocks.append(TimestepEmbedSequential(*layers))
+                self.zero_convs.append(self.make_zero_conv(ch))
+                self._feature_size += ch
+                input_block_chans.append(ch)
+            if level != len(channel_mult) - 1:
+                out_ch = ch
+                self.input_blocks.append(
+                    TimestepEmbedSequential(
+                        ResBlock(
+                            ch,
+                            time_embed_dim,
+                            dropout,
+                            out_channels=out_ch,
+                            dims=dims,
+                            use_checkpoint=use_checkpoint,
+                            use_scale_shift_norm=use_scale_shift_norm,
+                            down=True,
+                        )
+                        if resblock_updown
+                        else Downsample(
+                            ch, conv_resample, dims=dims, out_channels=out_ch
+                        )
+                    )
+                )
+                ch = out_ch
+                input_block_chans.append(ch)
+                self.zero_convs.append(self.make_zero_conv(ch))
+                ds *= 2
+                self._feature_size += ch
+
+        if num_head_channels == -1:
+            dim_head = ch // num_heads
+        else:
+            num_heads = ch // num_head_channels
+            dim_head = num_head_channels
+        if legacy:
+            # num_heads = 1
+            dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
+        self.middle_block = TimestepEmbedSequential(
+            ResBlock(
+                ch,
+                time_embed_dim,
+                dropout,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+            ),
+            AttentionBlock(
+                ch,
+                use_checkpoint=use_checkpoint,
+                num_heads=num_heads,
+                num_head_channels=dim_head,
+                use_new_attention_order=use_new_attention_order,
+            ) if not use_spatial_transformer else SpatialTransformer(  # always uses a self-attn
+                ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim,
+                disable_self_attn=disable_middle_self_attn, use_linear=use_linear_in_transformer,
+                use_checkpoint=use_checkpoint
+            ),
+            ResBlock(
+                ch,
+                time_embed_dim,
+                dropout,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+            ),
+        )
+        self.middle_block_out = self.make_zero_conv(ch)
+        self._feature_size += ch
+
+    def make_zero_conv(self, channels):
+        return TimestepEmbedSequential(zero_module(conv_nd(self.dims, channels, channels, 1, padding=0)))
+
+    def forward(self, x, hint, timesteps, context, **kwargs):
+        t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False)
+        emb = self.time_embed(t_emb)
+
+        if not self.noise_control:
+            guided_hint = self.input_hint_block(hint, emb, context)
+            h = x.type(self.dtype)
+        else:
+            guided_hint = None
+            h = hint.type(self.dtype)
+
+        outs = []
+        for module, zero_conv in zip(self.input_blocks, self.zero_convs):
+            if guided_hint is not None:
+                h = module(h, emb, context)
+                h += guided_hint
+                guided_hint = None
+            else:
+                h = module(h, emb, context)
+            outs.append(zero_conv(h, emb, context))
+
+        h = self.middle_block(h, emb, context)
+        outs.append(self.middle_block_out(h, emb, context))  # 分支先预测，把所有的outs都出来
+
+        return outs
+
+
+class ControlLDM(LatentDiffusion):
+
+    def __init__(self, control_stage_config, control_key, only_mid_control, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.control_model = instantiate_from_config(control_stage_config)
+        self.control_key = control_key
+        self.only_mid_control = only_mid_control
+        self.control_scales = [1.0] * 13
+
+
+    @torch.no_grad()
+    def get_input(self, batch, k, bs=None, *args, **kwargs):
+        x, c = super().get_input(batch, self.first_stage_key, *args, **kwargs)
+        control = batch[self.control_key]
+        if bs is not None:
+            control = control[:bs]
+        control = control.to(self.device)
+        control = einops.rearrange(control, 'b h w c -> b c h w')
+        control = control.to(memory_format=torch.contiguous_format).float()
+                # 之前用了一个预抽取的方法，不能兼容数据增强，所以改写
+        if self.control_model.latent_control:
+            control = (control * 2.0) - 1.0
+            control = self.encode_first_stage(control).mean
+        ctrl_loss_params = {}
+        if (self.l_coltrans_weight > 0 or self.l_mrcoltrans_weight > 0 or self.l_idcoltrans_weight > 0 or self.l_mridcoltrans_weight > 0) and "ctrl_mask" in batch.keys():  # 后面一个条件是为了兼容 test 和 sample 脚本
+            mask = batch["ctrl_mask"]
+            if bs is not None:
+                mask = mask[:bs]
+            mask = mask.to(self.device)
+            ctrl_loss_params["ctrl_mask"] = mask
+            tgt_rgb = batch["tgt"]
+            if bs is not None:
+                tgt_rgb = tgt_rgb[:bs]
+            tgt_rgb = tgt_rgb.to(self.device)
+            ctrl_loss_params["tgt_rgb"] = tgt_rgb
+            ctrl_rgb = batch["ctrl"]
+            if bs is not None:
+                ctrl_rgb = ctrl_rgb[:bs]
+            ctrl_rgb = ctrl_rgb.to(self.device)
+            ctrl_loss_params["ctrl_rgb"] = ctrl_rgb
+        return x, dict(c_crossattn=[c], c_concat=[control])
+
+    def apply_model(self, x_noisy, t, cond, *args, **kwargs):
+        assert isinstance(cond, dict)
+        diffusion_model = self.model.diffusion_model
+
+        cond_txt = torch.cat(cond['c_crossattn'], 1)
+
+        if cond['c_concat'] is None:
+            eps = diffusion_model(x=x_noisy, timesteps=t, context=cond_txt, control=None, only_mid_control=self.only_mid_control)
+        else:
+            control = self.control_model(x=x_noisy, hint=torch.cat(cond['c_concat'], 1), timesteps=t, context=cond_txt)
+            control = [c * scale for c, scale in zip(control, self.control_scales)]
+            eps = diffusion_model(x=x_noisy, timesteps=t, context=cond_txt, control=control, only_mid_control=self.only_mid_control)
+
+        return eps
+
+    @torch.no_grad()
+    def get_unconditional_conditioning(self, N):
+        return self.get_learned_conditioning([""] * N)
+
+    @torch.no_grad()
+    def log_images(self, batch, N=4, n_row=2, sample=False, ddim_steps=50, ddim_eta=0.0, return_keys=None,
+                   quantize_denoised=True, inpaint=True, plot_denoise_rows=False, plot_progressive_rows=True,
+                   plot_diffusion_rows=False, unconditional_guidance_scale=9.0, unconditional_guidance_label=None,
+                   use_ema_scope=True,
+                   **kwargs):
+        use_ddim = ddim_steps is not None
+
+        log = dict()
+        z, c = self.get_input(batch, self.first_stage_key, bs=N)
+        c_cat, c = c["c_concat"][0][:N], c["c_crossattn"][0][:N]
+        N = min(z.shape[0], N)
+        n_row = min(z.shape[0], n_row)
+        log["reconstruction"] = self.decode_first_stage(z)
+        log["control"] = c_cat * 2.0 - 1.0
+        log["conditioning"] = log_txt_as_img((512, 512), batch[self.cond_stage_key], size=16)
+
+        if plot_diffusion_rows:
+            # get diffusion row
+            diffusion_row = list()
+            z_start = z[:n_row]
+            for t in range(self.num_timesteps):
+                if t % self.log_every_t == 0 or t == self.num_timesteps - 1:
+                    t = repeat(torch.tensor([t]), '1 -> b', b=n_row)
+                    t = t.to(self.device).long()
+                    noise = torch.randn_like(z_start)
+                    z_noisy = self.q_sample(x_start=z_start, t=t, noise=noise)
+                    diffusion_row.append(self.decode_first_stage(z_noisy))
+
+            diffusion_row = torch.stack(diffusion_row)  # n_log_step, n_row, C, H, W
+            diffusion_grid = rearrange(diffusion_row, 'n b c h w -> b n c h w')
+            diffusion_grid = rearrange(diffusion_grid, 'b n c h w -> (b n) c h w')
+            diffusion_grid = make_grid(diffusion_grid, nrow=diffusion_row.shape[0])
+            log["diffusion_row"] = diffusion_grid
+
+        if sample:
+            # get denoise row
+            samples, z_denoise_row = self.sample_log(cond={"c_concat": [c_cat], "c_crossattn": [c]},
+                                                     batch_size=N, ddim=use_ddim,
+                                                     ddim_steps=ddim_steps, eta=ddim_eta)
+            x_samples = self.decode_first_stage(samples)
+            log["samples"] = x_samples
+            if plot_denoise_rows:
+                denoise_grid = self._get_denoise_row_from_list(z_denoise_row)
+                log["denoise_row"] = denoise_grid
+
+        if unconditional_guidance_scale > 1.0:
+            uc_cross = self.get_unconditional_conditioning(N)
+            uc_cat = c_cat  # torch.zeros_like(c_cat)
+            uc_full = {"c_concat": [uc_cat], "c_crossattn": [uc_cross]}
+            samples_cfg, _ = self.sample_log(cond={"c_concat": [c_cat], "c_crossattn": [c]},
+                                             batch_size=N, ddim=use_ddim,
+                                             ddim_steps=ddim_steps, eta=ddim_eta,
+                                             unconditional_guidance_scale=unconditional_guidance_scale,
+                                             unconditional_conditioning=uc_full,
+                                             )
+            x_samples_cfg = self.decode_first_stage(samples_cfg)
+            log[f"samples_cfg_scale_{unconditional_guidance_scale:.2f}"] = x_samples_cfg
+
+        return log
+
+    @torch.no_grad()
+    def sample_log(self, cond, batch_size, ddim, ddim_steps, **kwargs):
+        ddim_sampler = DDIMSampler(self)
+        b, c, h, w = cond["c_concat"][0].shape
+        shape = (self.channels, h // 8, w // 8)
+        samples, intermediates = ddim_sampler.sample(ddim_steps, batch_size, shape, cond, verbose=False, **kwargs)
+        return samples, intermediates
+
+    def configure_optimizers(self):
+        lr = self.learning_rate
+        params = list(self.control_model.parameters())
+        if not self.sd_locked:
+            params += list(self.model.diffusion_model.output_blocks.parameters())
+            params += list(self.model.diffusion_model.out.parameters())
+        opt = torch.optim.AdamW(params, lr=lr)
+        return opt
+
+    def low_vram_shift(self, is_diffusing):
+        if is_diffusing:
+            self.model = self.model.cuda()
+            self.control_model = self.control_model.cuda()
+            self.first_stage_model = self.first_stage_model.cpu()
+            self.cond_stage_model = self.cond_stage_model.cpu()
+        else:
+            self.model = self.model.cpu()
+            self.control_model = self.control_model.cpu()
+            self.first_stage_model = self.first_stage_model.cuda()
+            self.cond_stage_model = self.cond_stage_model.cuda()

cldm/ddim_hacked.py

@@ -0,0 +1,317 @@
+"""SAMPLING ONLY."""
+
+import torch
+import numpy as np
+from tqdm import tqdm
+
+from ldm.modules.diffusionmodules.util import make_ddim_sampling_parameters, make_ddim_timesteps, noise_like, extract_into_tensor
+
+
+class DDIMSampler(object):
+    def __init__(self, model, schedule="linear", **kwargs):
+        super().__init__()
+        self.model = model
+        self.ddpm_num_timesteps = model.num_timesteps
+        self.schedule = schedule
+
+    def register_buffer(self, name, attr):
+        if type(attr) == torch.Tensor:
+            if attr.device != torch.device("cuda"):
+                attr = attr.to(torch.device("cuda"))
+        setattr(self, name, attr)
+
+    def make_schedule(self, ddim_num_steps, ddim_discretize="uniform", ddim_eta=0., verbose=True):
+        self.ddim_timesteps = make_ddim_timesteps(ddim_discr_method=ddim_discretize, num_ddim_timesteps=ddim_num_steps,
+                                                  num_ddpm_timesteps=self.ddpm_num_timesteps,verbose=verbose)
+        alphas_cumprod = self.model.alphas_cumprod
+        assert alphas_cumprod.shape[0] == self.ddpm_num_timesteps, 'alphas have to be defined for each timestep'
+        to_torch = lambda x: x.clone().detach().to(torch.float32).to(self.model.device)
+
+        self.register_buffer('betas', to_torch(self.model.betas))
+        self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod))
+        self.register_buffer('alphas_cumprod_prev', to_torch(self.model.alphas_cumprod_prev))
+
+        # calculations for diffusion q(x_t | x_{t-1}) and others
+        self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod.cpu())))
+        self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod.cpu())))
+        self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod.cpu())))
+        self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu())))
+        self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu() - 1)))
+
+        # ddim sampling parameters
+        ddim_sigmas, ddim_alphas, ddim_alphas_prev = make_ddim_sampling_parameters(alphacums=alphas_cumprod.cpu(),
+                                                                                   ddim_timesteps=self.ddim_timesteps,
+                                                                                   eta=ddim_eta,verbose=verbose)
+        self.register_buffer('ddim_sigmas', ddim_sigmas)
+        self.register_buffer('ddim_alphas', ddim_alphas)
+        self.register_buffer('ddim_alphas_prev', ddim_alphas_prev)
+        self.register_buffer('ddim_sqrt_one_minus_alphas', np.sqrt(1. - ddim_alphas))
+        sigmas_for_original_sampling_steps = ddim_eta * torch.sqrt(
+            (1 - self.alphas_cumprod_prev) / (1 - self.alphas_cumprod) * (
+                        1 - self.alphas_cumprod / self.alphas_cumprod_prev))
+        self.register_buffer('ddim_sigmas_for_original_num_steps', sigmas_for_original_sampling_steps)
+
+    @torch.no_grad()
+    def sample(self,
+               S,
+               batch_size,
+               shape,
+               conditioning=None,
+               callback=None,
+               normals_sequence=None,
+               img_callback=None,
+               quantize_x0=False,
+               eta=0.,
+               mask=None,
+               x0=None,
+               temperature=1.,
+               noise_dropout=0.,
+               score_corrector=None,
+               corrector_kwargs=None,
+               verbose=True,
+               x_T=None,
+               log_every_t=100,
+               unconditional_guidance_scale=1.,
+               unconditional_conditioning=None, # this has to come in the same format as the conditioning, # e.g. as encoded tokens, ...
+               dynamic_threshold=None,
+               ucg_schedule=None,
+               **kwargs
+               ):
+        if conditioning is not None:
+            if isinstance(conditioning, dict):
+                ctmp = conditioning[list(conditioning.keys())[0]]
+                while isinstance(ctmp, list): ctmp = ctmp[0]
+                cbs = ctmp.shape[0]
+                if cbs != batch_size:
+                    print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}")
+
+            elif isinstance(conditioning, list):
+                for ctmp in conditioning:
+                    if ctmp.shape[0] != batch_size:
+                        print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}")
+
+            else:
+                if conditioning.shape[0] != batch_size:
+                    print(f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}")
+
+        self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=verbose)
+        # sampling
+        C, H, W = shape
+        size = (batch_size, C, H, W)
+        print(f'Data shape for DDIM sampling is {size}, eta {eta}')
+
+        samples, intermediates = self.ddim_sampling(conditioning, size,
+                                                    callback=callback,
+                                                    img_callback=img_callback,
+                                                    quantize_denoised=quantize_x0,
+                                                    mask=mask, x0=x0,
+                                                    ddim_use_original_steps=False,
+                                                    noise_dropout=noise_dropout,
+                                                    temperature=temperature,
+                                                    score_corrector=score_corrector,
+                                                    corrector_kwargs=corrector_kwargs,
+                                                    x_T=x_T,
+                                                    log_every_t=log_every_t,
+                                                    unconditional_guidance_scale=unconditional_guidance_scale,
+                                                    unconditional_conditioning=unconditional_conditioning,
+                                                    dynamic_threshold=dynamic_threshold,
+                                                    ucg_schedule=ucg_schedule
+                                                    )
+        return samples, intermediates
+
+    @torch.no_grad()
+    def ddim_sampling(self, cond, shape,
+                      x_T=None, ddim_use_original_steps=False,
+                      callback=None, timesteps=None, quantize_denoised=False,
+                      mask=None, x0=None, img_callback=None, log_every_t=100,
+                      temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
+                      unconditional_guidance_scale=1., unconditional_conditioning=None, dynamic_threshold=None,
+                      ucg_schedule=None):
+        device = self.model.betas.device
+        b = shape[0]
+        if x_T is None:
+            img = torch.randn(shape, device=device)
+        else:
+            img = x_T
+
+        if timesteps is None:
+            timesteps = self.ddpm_num_timesteps if ddim_use_original_steps else self.ddim_timesteps
+        elif timesteps is not None and not ddim_use_original_steps:
+            subset_end = int(min(timesteps / self.ddim_timesteps.shape[0], 1) * self.ddim_timesteps.shape[0]) - 1
+            timesteps = self.ddim_timesteps[:subset_end]
+
+        intermediates = {'x_inter': [img], 'pred_x0': [img]}
+        time_range = reversed(range(0,timesteps)) if ddim_use_original_steps else np.flip(timesteps)
+        total_steps = timesteps if ddim_use_original_steps else timesteps.shape[0]
+        print(f"Running DDIM Sampling with {total_steps} timesteps")
+
+        iterator = tqdm(time_range, desc='DDIM Sampler', total=total_steps)
+
+        for i, step in enumerate(iterator):
+            index = total_steps - i - 1
+            ts = torch.full((b,), step, device=device, dtype=torch.long)
+
+            if mask is not None:
+                assert x0 is not None
+                img_orig = self.model.q_sample(x0, ts)  # TODO: deterministic forward pass?
+                img = img_orig * mask + (1. - mask) * img
+
+            if ucg_schedule is not None:
+                assert len(ucg_schedule) == len(time_range)
+                unconditional_guidance_scale = ucg_schedule[i]
+
+            outs = self.p_sample_ddim(img, cond, ts, index=index, use_original_steps=ddim_use_original_steps,
+                                      quantize_denoised=quantize_denoised, temperature=temperature,
+                                      noise_dropout=noise_dropout, score_corrector=score_corrector,
+                                      corrector_kwargs=corrector_kwargs,
+                                      unconditional_guidance_scale=unconditional_guidance_scale,
+                                      unconditional_conditioning=unconditional_conditioning,
+                                      dynamic_threshold=dynamic_threshold)
+            img, pred_x0 = outs
+            if callback: callback(i)
+            if img_callback: img_callback(pred_x0, i)
+
+            if index % log_every_t == 0 or index == total_steps - 1:
+                intermediates['x_inter'].append(img)
+                intermediates['pred_x0'].append(pred_x0)
+
+        return img, intermediates
+
+    @torch.no_grad()
+    def p_sample_ddim(self, x, c, t, index, repeat_noise=False, use_original_steps=False, quantize_denoised=False,
+                      temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
+                      unconditional_guidance_scale=1., unconditional_conditioning=None,
+                      dynamic_threshold=None):
+        b, *_, device = *x.shape, x.device
+
+        if unconditional_conditioning is None or unconditional_guidance_scale == 1.:
+            model_output = self.model.apply_model(x, t, c)
+        else:
+            model_t = self.model.apply_model(x, t, c)
+            model_uncond = self.model.apply_model(x, t, unconditional_conditioning)
+            model_output = model_uncond + unconditional_guidance_scale * (model_t - model_uncond)
+
+        if self.model.parameterization == "v":
+            e_t = self.model.predict_eps_from_z_and_v(x, t, model_output)
+        else:
+            e_t = model_output
+
+        if score_corrector is not None:
+            assert self.model.parameterization == "eps", 'not implemented'
+            e_t = score_corrector.modify_score(self.model, e_t, x, t, c, **corrector_kwargs)
+
+        alphas = self.model.alphas_cumprod if use_original_steps else self.ddim_alphas
+        alphas_prev = self.model.alphas_cumprod_prev if use_original_steps else self.ddim_alphas_prev
+        sqrt_one_minus_alphas = self.model.sqrt_one_minus_alphas_cumprod if use_original_steps else self.ddim_sqrt_one_minus_alphas
+        sigmas = self.model.ddim_sigmas_for_original_num_steps if use_original_steps else self.ddim_sigmas
+        # select parameters corresponding to the currently considered timestep
+        a_t = torch.full((b, 1, 1, 1), alphas[index], device=device)
+        a_prev = torch.full((b, 1, 1, 1), alphas_prev[index], device=device)
+        sigma_t = torch.full((b, 1, 1, 1), sigmas[index], device=device)
+        sqrt_one_minus_at = torch.full((b, 1, 1, 1), sqrt_one_minus_alphas[index],device=device)
+
+        # current prediction for x_0
+        if self.model.parameterization != "v":
+            pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt()
+        else:
+            pred_x0 = self.model.predict_start_from_z_and_v(x, t, model_output)
+
+        if quantize_denoised:
+            pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0)
+
+        if dynamic_threshold is not None:
+            raise NotImplementedError()
+
+        # direction pointing to x_t
+        dir_xt = (1. - a_prev - sigma_t**2).sqrt() * e_t
+        noise = sigma_t * noise_like(x.shape, device, repeat_noise) * temperature
+        if noise_dropout > 0.:
+            noise = torch.nn.functional.dropout(noise, p=noise_dropout)
+        x_prev = a_prev.sqrt() * pred_x0 + dir_xt + noise
+        return x_prev, pred_x0
+
+    @torch.no_grad()
+    def encode(self, x0, c, t_enc, use_original_steps=False, return_intermediates=None,
+               unconditional_guidance_scale=1.0, unconditional_conditioning=None, callback=None):
+        timesteps = np.arange(self.ddpm_num_timesteps) if use_original_steps else self.ddim_timesteps
+        num_reference_steps = timesteps.shape[0]
+
+        assert t_enc <= num_reference_steps
+        num_steps = t_enc
+
+        if use_original_steps:
+            alphas_next = self.alphas_cumprod[:num_steps]
+            alphas = self.alphas_cumprod_prev[:num_steps]
+        else:
+            alphas_next = self.ddim_alphas[:num_steps]
+            alphas = torch.tensor(self.ddim_alphas_prev[:num_steps])
+
+        x_next = x0
+        intermediates = []
+        inter_steps = []
+        for i in tqdm(range(num_steps), desc='Encoding Image'):
+            t = torch.full((x0.shape[0],), timesteps[i], device=self.model.device, dtype=torch.long)
+            if unconditional_guidance_scale == 1.:
+                noise_pred = self.model.apply_model(x_next, t, c)
+            else:
+                assert unconditional_conditioning is not None
+                e_t_uncond, noise_pred = torch.chunk(
+                    self.model.apply_model(torch.cat((x_next, x_next)), torch.cat((t, t)),
+                                           torch.cat((unconditional_conditioning, c))), 2)
+                noise_pred = e_t_uncond + unconditional_guidance_scale * (noise_pred - e_t_uncond)
+
+            xt_weighted = (alphas_next[i] / alphas[i]).sqrt() * x_next
+            weighted_noise_pred = alphas_next[i].sqrt() * (
+                    (1 / alphas_next[i] - 1).sqrt() - (1 / alphas[i] - 1).sqrt()) * noise_pred
+            x_next = xt_weighted + weighted_noise_pred
+            if return_intermediates and i % (
+                    num_steps // return_intermediates) == 0 and i < num_steps - 1:
+                intermediates.append(x_next)
+                inter_steps.append(i)
+            elif return_intermediates and i >= num_steps - 2:
+                intermediates.append(x_next)
+                inter_steps.append(i)
+            if callback: callback(i)
+
+        out = {'x_encoded': x_next, 'intermediate_steps': inter_steps}
+        if return_intermediates:
+            out.update({'intermediates': intermediates})
+        return x_next, out
+
+    @torch.no_grad()
+    def stochastic_encode(self, x0, t, use_original_steps=False, noise=None):
+        # fast, but does not allow for exact reconstruction
+        # t serves as an index to gather the correct alphas
+        if use_original_steps:
+            sqrt_alphas_cumprod = self.sqrt_alphas_cumprod
+            sqrt_one_minus_alphas_cumprod = self.sqrt_one_minus_alphas_cumprod
+        else:
+            sqrt_alphas_cumprod = torch.sqrt(self.ddim_alphas)
+            sqrt_one_minus_alphas_cumprod = self.ddim_sqrt_one_minus_alphas
+
+        if noise is None:
+            noise = torch.randn_like(x0)
+        return (extract_into_tensor(sqrt_alphas_cumprod, t, x0.shape) * x0 +
+                extract_into_tensor(sqrt_one_minus_alphas_cumprod, t, x0.shape) * noise)
+
+    @torch.no_grad()
+    def decode(self, x_latent, cond, t_start, unconditional_guidance_scale=1.0, unconditional_conditioning=None,
+               use_original_steps=False, callback=None):
+
+        timesteps = np.arange(self.ddpm_num_timesteps) if use_original_steps else self.ddim_timesteps
+        timesteps = timesteps[:t_start]
+
+        time_range = np.flip(timesteps)
+        total_steps = timesteps.shape[0]
+        print(f"Running DDIM Sampling with {total_steps} timesteps")
+
+        iterator = tqdm(time_range, desc='Decoding image', total=total_steps)
+        x_dec = x_latent
+        for i, step in enumerate(iterator):
+            index = total_steps - i - 1
+            ts = torch.full((x_latent.shape[0],), step, device=x_latent.device, dtype=torch.long)
+            x_dec, _ = self.p_sample_ddim(x_dec, cond, ts, index=index, use_original_steps=use_original_steps,
+                                          unconditional_guidance_scale=unconditional_guidance_scale,
+                                          unconditional_conditioning=unconditional_conditioning)
+            if callback: callback(i)
+        return x_dec

cldm/hack.py

@@ -0,0 +1,111 @@
+import torch
+import einops
+
+import ldm.modules.encoders.modules
+import ldm.modules.attention
+
+from transformers import logging
+from ldm.modules.attention import default
+
+
+def disable_verbosity():
+    logging.set_verbosity_error()
+    print('logging improved.')
+    return
+
+
+def enable_sliced_attention():
+    ldm.modules.attention.CrossAttention.forward = _hacked_sliced_attentin_forward
+    print('Enabled sliced_attention.')
+    return
+
+
+def hack_everything(clip_skip=0):
+    disable_verbosity()
+    ldm.modules.encoders.modules.FrozenCLIPEmbedder.forward = _hacked_clip_forward
+    ldm.modules.encoders.modules.FrozenCLIPEmbedder.clip_skip = clip_skip
+    print('Enabled clip hacks.')
+    return
+
+
+# Written by Lvmin
+def _hacked_clip_forward(self, text):
+    PAD = self.tokenizer.pad_token_id
+    EOS = self.tokenizer.eos_token_id
+    BOS = self.tokenizer.bos_token_id
+
+    def tokenize(t):
+        return self.tokenizer(t, truncation=False, add_special_tokens=False)["input_ids"]
+
+    def transformer_encode(t):
+        if self.clip_skip > 1:
+            rt = self.transformer(input_ids=t, output_hidden_states=True)
+            return self.transformer.text_model.final_layer_norm(rt.hidden_states[-self.clip_skip])
+        else:
+            return self.transformer(input_ids=t, output_hidden_states=False).last_hidden_state
+
+    def split(x):
+        return x[75 * 0: 75 * 1], x[75 * 1: 75 * 2], x[75 * 2: 75 * 3]
+
+    def pad(x, p, i):
+        return x[:i] if len(x) >= i else x + [p] * (i - len(x))
+
+    raw_tokens_list = tokenize(text)
+    tokens_list = []
+
+    for raw_tokens in raw_tokens_list:
+        raw_tokens_123 = split(raw_tokens)
+        raw_tokens_123 = [[BOS] + raw_tokens_i + [EOS] for raw_tokens_i in raw_tokens_123]
+        raw_tokens_123 = [pad(raw_tokens_i, PAD, 77) for raw_tokens_i in raw_tokens_123]
+        tokens_list.append(raw_tokens_123)
+
+    tokens_list = torch.IntTensor(tokens_list).to(self.device)
+
+    feed = einops.rearrange(tokens_list, 'b f i -> (b f) i')
+    y = transformer_encode(feed)
+    z = einops.rearrange(y, '(b f) i c -> b (f i) c', f=3)
+
+    return z
+
+
+# Stolen from https://github.com/basujindal/stable-diffusion/blob/main/optimizedSD/splitAttention.py
+def _hacked_sliced_attentin_forward(self, x, context=None, mask=None):
+    h = self.heads
+
+    q = self.to_q(x)
+    context = default(context, x)
+    k = self.to_k(context)
+    v = self.to_v(context)
+    del context, x
+
+    q, k, v = map(lambda t: einops.rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q, k, v))
+
+    limit = k.shape[0]
+    att_step = 1
+    q_chunks = list(torch.tensor_split(q, limit // att_step, dim=0))
+    k_chunks = list(torch.tensor_split(k, limit // att_step, dim=0))
+    v_chunks = list(torch.tensor_split(v, limit // att_step, dim=0))
+
+    q_chunks.reverse()
+    k_chunks.reverse()
+    v_chunks.reverse()
+    sim = torch.zeros(q.shape[0], q.shape[1], v.shape[2], device=q.device)
+    del k, q, v
+    for i in range(0, limit, att_step):
+        q_buffer = q_chunks.pop()
+        k_buffer = k_chunks.pop()
+        v_buffer = v_chunks.pop()
+        sim_buffer = torch.einsum('b i d, b j d -> b i j', q_buffer, k_buffer) * self.scale
+
+        del k_buffer, q_buffer
+        # attention, what we cannot get enough of, by chunks
+
+        sim_buffer = sim_buffer.softmax(dim=-1)
+
+        sim_buffer = torch.einsum('b i j, b j d -> b i d', sim_buffer, v_buffer)
+        del v_buffer
+        sim[i:i + att_step, :, :] = sim_buffer
+
+        del sim_buffer
+    sim = einops.rearrange(sim, '(b h) n d -> b n (h d)', h=h)
+    return self.to_out(sim)

cldm/logger.py

@@ -0,0 +1,76 @@
+import os
+
+import numpy as np
+import torch
+import torchvision
+from PIL import Image
+from pytorch_lightning.callbacks import Callback
+from pytorch_lightning.utilities.distributed import rank_zero_only
+
+
+class ImageLogger(Callback):
+    def __init__(self, batch_frequency=2000, max_images=4, clamp=True, increase_log_steps=True,
+                 rescale=True, disabled=False, log_on_batch_idx=False, log_first_step=False,
+                 log_images_kwargs=None):
+        super().__init__()
+        self.rescale = rescale
+        self.batch_freq = batch_frequency
+        self.max_images = max_images
+        if not increase_log_steps:
+            self.log_steps = [self.batch_freq]
+        self.clamp = clamp
+        self.disabled = disabled
+        self.log_on_batch_idx = log_on_batch_idx
+        self.log_images_kwargs = log_images_kwargs if log_images_kwargs else {}
+        self.log_first_step = log_first_step
+
+    @rank_zero_only
+    def log_local(self, save_dir, split, images, global_step, current_epoch, batch_idx):
+        root = os.path.join(save_dir, "image_log", split)
+        for k in images:
+            grid = torchvision.utils.make_grid(images[k], nrow=4)
+            if self.rescale:
+                grid = (grid + 1.0) / 2.0  # -1,1 -> 0,1; c,h,w
+            grid = grid.transpose(0, 1).transpose(1, 2).squeeze(-1)
+            grid = grid.numpy()
+            grid = (grid * 255).astype(np.uint8)
+            filename = "{}_gs-{:06}_e-{:06}_b-{:06}.png".format(k, global_step, current_epoch, batch_idx)
+            path = os.path.join(root, filename)
+            os.makedirs(os.path.split(path)[0], exist_ok=True)
+            Image.fromarray(grid).save(path)
+
+    def log_img(self, pl_module, batch, batch_idx, split="train"):
+        check_idx = batch_idx  # if self.log_on_batch_idx else pl_module.global_step
+        if (self.check_frequency(check_idx) and  # batch_idx % self.batch_freq == 0
+                hasattr(pl_module, "log_images") and
+                callable(pl_module.log_images) and
+                self.max_images > 0):
+            logger = type(pl_module.logger)
+
+            is_train = pl_module.training
+            if is_train:
+                pl_module.eval()
+
+            with torch.no_grad():
+                images = pl_module.log_images(batch, split=split, **self.log_images_kwargs)
+
+            for k in images:
+                N = min(images[k].shape[0], self.max_images)
+                images[k] = images[k][:N]
+                if isinstance(images[k], torch.Tensor):
+                    images[k] = images[k].detach().cpu()
+                    if self.clamp:
+                        images[k] = torch.clamp(images[k], -1., 1.)
+
+            self.log_local(pl_module.logger.save_dir, split, images,
+                           pl_module.global_step, pl_module.current_epoch, batch_idx)
+
+            if is_train:
+                pl_module.train()
+
+    def check_frequency(self, check_idx):
+        return check_idx % self.batch_freq == 0
+
+    def on_train_batch_end(self, trainer, pl_module, outputs, batch, batch_idx, dataloader_idx):
+        if not self.disabled:
+            self.log_img(pl_module, batch, batch_idx, split="train")

cldm/model.py

@@ -0,0 +1,28 @@
+import os
+import torch
+
+from omegaconf import OmegaConf
+from ldm.util import instantiate_from_config
+
+
+def get_state_dict(d):
+    return d.get('state_dict', d)
+
+
+def load_state_dict(ckpt_path, location='cpu'):
+    _, extension = os.path.splitext(ckpt_path)
+    if extension.lower() == ".safetensors":
+        import safetensors.torch
+        state_dict = safetensors.torch.load_file(ckpt_path, device=location)
+    else:
+        state_dict = get_state_dict(torch.load(ckpt_path, map_location=torch.device(location)))
+    state_dict = get_state_dict(state_dict)
+    print(f'Loaded state_dict from [{ckpt_path}]')
+    return state_dict
+
+
+def create_model(config_path):
+    config = OmegaConf.load(config_path)
+    model = instantiate_from_config(config.model).cpu()
+    print(f'Loaded model config from [{config_path}]')
+    return model