pipeline.py

from diffusers import StableDiffusionPipeline
import torch
from typing import Union, List, Optional, Callable
import inspect
import re
import os
import copy

def load_learned_concepts(pipe, root_folder="selected_outputs/", num_scales=10):
    """Load multiresolution concepts from a directory.

        Args:
            pipe: A diffusers pipeline. Tested with StableDiffusionPipeline.
            root_folder: folder where the concepts are stored.
                Expected structure: {root_folder}/{experiment_name}.

        Returns:
            string_to_param_dict: A torch.nn.Parameter dictionary where the keys are 
                the token names and the values are the embeddings.
    """
    string_to_param_dict = torch.nn.ParameterDict()
    string_to_token_dict = {}
    tokens_to_add = []

    pure_names = []

    for exp_name in os.listdir(root_folder):
        # get everything up to the first numeric
        pure_names.append(exp_name)
        encoder = torch.load(os.path.join(root_folder, exp_name, "text_encoder/pytorch_model.bin"))
        embeddings = encoder["text_model.embeddings.token_embedding.weight"]
        param_value = embeddings[-10:]

        for t in range(num_scales):
            # Fixed Resolution: fix embedding to specific value throughout the generation.
            string_name = f"<{exp_name}|{t}|>"
            tokens_to_add.append(string_name)
            string_to_param_dict[string_name] = torch.nn.Parameter(param_value[t].unsqueeze(0).repeat([num_scales, 1]))
            
            # Fully Resolution: use appropriate time embedding for the whole generation time.
            string_name = f"<{exp_name}[{t}]>"
            tokens_to_add.append(string_name)
            repeats = t + 1
            rep_param = param_value[t].unsqueeze(0).repeat([repeats, 1])
            left = param_value[rep_param.shape[0]:]
            new_param = torch.cat([rep_param, left]) 
            string_to_param_dict[string_name] = torch.nn.Parameter(new_param)

            # Semi Resolution: use appropriate time embedding up to a certain time and then no conditioning.
            string_name = f"<{exp_name}({t})>"
            tokens_to_add.append(string_name)
            null_embedding = torch.zeros((param_value.shape[1],), device=param_value.device, dtype=param_value.dtype)
            rep_param = null_embedding.unsqueeze(0).repeat([t + 1, 1])
            left = param_value[rep_param.shape[0]:]
            new_param = torch.cat([rep_param, left]) 
            string_to_param_dict[string_name] = torch.nn.Parameter(new_param)

    pipe.tokenizer.add_tokens(tokens_to_add)
    pipe.text_encoder.resize_token_embeddings(len(pipe.tokenizer))
    return string_to_param_dict


class MultiResPipeline(StableDiffusionPipeline):
    """Extends StableDiffusionPipeline for multiresolution textual embeddings."""

    def update_embeddings_based_on_time(self, normalized_time, string_to_param_dict):
        num_embs = [x for x in string_to_param_dict.values()][0].shape[0]
        time_index = int(normalized_time * num_embs)
        
        old_dict = self.text_encoder.text_model.embeddings.token_embedding.state_dict()
        for token, token_val in string_to_param_dict.items():
            try:
                token_val_at_time = token_val[time_index]
            except:
                print(token)
            old_id = self.tokenizer(token, add_special_tokens=False)["input_ids"][0]
            old_dict["weight"][old_id] = token_val_at_time
        # update values
        self.text_encoder.text_model.embeddings.token_embedding.load_state_dict(old_dict)

    
    def get_embeddings(self, text_input_ids, batch_size, prompt, negative_prompt, guidance_scale, num_images_per_prompt):
        text_embeddings = self.text_encoder(text_input_ids.to(self.device))[0]


        # duplicate text embeddings for each generation per prompt, using mps friendly method
        bs_embed, seq_len, _ = text_embeddings.shape
        text_embeddings = text_embeddings.repeat(1, num_images_per_prompt, 1)
        text_embeddings = text_embeddings.view(bs_embed * num_images_per_prompt, seq_len, -1)

        # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
        # corresponds to doing no classifier free guidance.
        do_classifier_free_guidance = guidance_scale > 1.0
        # get unconditional embeddings for classifier free guidance
        if do_classifier_free_guidance:
            uncond_tokens: List[str]
            if negative_prompt is None:
                uncond_tokens = [""] * batch_size
            elif type(prompt) is not type(negative_prompt):
                raise TypeError(
                    f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
                    f" {type(prompt)}."
                )
            elif isinstance(negative_prompt, str):
                uncond_tokens = [negative_prompt]
            elif batch_size != len(negative_prompt):
                raise ValueError(
                    f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
                    f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
                    " the batch size of `prompt`."
                )
            else:
                uncond_tokens = negative_prompt

            max_length = text_input_ids.shape[-1]
            uncond_input = self.tokenizer(
                uncond_tokens,
                padding="max_length",
                max_length=max_length,
                truncation=True,
                return_tensors="pt",
            )
            uncond_embeddings = self.text_encoder(uncond_input.input_ids.to(self.device))[0]

            # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
            seq_len = uncond_embeddings.shape[1]
            uncond_embeddings = uncond_embeddings.repeat(1, num_images_per_prompt, 1)
            uncond_embeddings = uncond_embeddings.view(batch_size * num_images_per_prompt, seq_len, -1)

            # For classifier free guidance, we need to do two forward passes.
            # Here we concatenate the unconditional and text embeddings into a single batch
            # to avoid doing two forward passes
            text_embeddings = torch.cat([uncond_embeddings, text_embeddings])
        return text_embeddings

    @torch.no_grad()
    def __call__(
        self,
        prompt: Union[str, List[str]],
        string_to_param_dict = None,
        height: int = 512,
        width: int = 512,
        num_inference_steps: int = 50,
        guidance_scale: float = 7.5,
        negative_prompt: Optional[Union[str, List[str]]] = None,
        num_images_per_prompt: Optional[int] = 1,
        eta: float = 0.0,
        generator: Optional[torch.Generator] = None,
        latents: Optional[torch.FloatTensor] = None,
        output_type: Optional[str] = "pil",
        return_dict: bool = True,
        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
        callback_steps: Optional[int] = 1,
        **kwargs,
    ):
        r"""
        Function invoked when calling the pipeline for generation.
        Args:
            prompt (`str` or `List[str]`):
                The prompt or prompts to guide the image generation.
            height (`int`, *optional*, defaults to 512):
                The height in pixels of the generated image.
            width (`int`, *optional*, defaults to 512):
                The width in pixels of the generated image.
            num_inference_steps (`int`, *optional*, defaults to 50):
                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
                expense of slower inference.
            guidance_scale (`float`, *optional*, defaults to 7.5):
                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
                `guidance_scale` is defined as `w` of equation 2. of [Imagen
                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
                usually at the expense of lower image quality.
            negative_prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts not to guide the image generation. Ignored when not using guidance (i.e., ignored
                if `guidance_scale` is less than `1`).
            num_images_per_prompt (`int`, *optional*, defaults to 1):
                The number of images to generate per prompt.
            eta (`float`, *optional*, defaults to 0.0):
                Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
                [`schedulers.DDIMScheduler`], will be ignored for others.
            generator (`torch.Generator`, *optional*):
                A [torch generator](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make generation
                deterministic.
            latents (`torch.FloatTensor`, *optional*):
                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
                tensor will ge generated by sampling using the supplied random `generator`.
            output_type (`str`, *optional*, defaults to `"pil"`):
                The output format of the generate image. Choose between
                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
            return_dict (`bool`, *optional*, defaults to `True`):
                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
                plain tuple.
            callback (`Callable`, *optional*):
                A function that will be called every `callback_steps` steps during inference. The function will be
                called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
            callback_steps (`int`, *optional*, defaults to 1):
                The frequency at which the `callback` function will be called. If not specified, the callback will be
                called at every step.
        Returns:
            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple.
            When returning a tuple, the first element is a list with the generated images, and the second element is a
            list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
            (nsfw) content, according to the `safety_checker`.
        """
        if isinstance(prompt, str):
            batch_size = 1
        elif isinstance(prompt, list):
            batch_size = len(prompt)
        else:
            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")

        if height % 8 != 0 or width % 8 != 0:
            raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")

        if (callback_steps is None) or (
            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
        ):
            raise ValueError(
                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
                f" {type(callback_steps)}."
            )

        # get prompt text embeddings
        text_inputs = self.tokenizer(
            prompt,
            padding="max_length",
            max_length=self.tokenizer.model_max_length,
            return_tensors="pt",
        )
        text_input_ids = text_inputs.input_ids

        if text_input_ids.shape[-1] > self.tokenizer.model_max_length:
            removed_text = self.tokenizer.batch_decode(text_input_ids[:, self.tokenizer.model_max_length :])
            print(
                "The following part of your input was truncated because CLIP can only handle sequences up to"
                f" {self.tokenizer.model_max_length} tokens: {removed_text}"
            )
            text_input_ids = text_input_ids[:, : self.tokenizer.model_max_length]
        
        do_classifier_free_guidance = guidance_scale > 1.0
        # get the initial random noise unless the user supplied it

        # Unlike in other pipelines, latents need to be generated in the target device
        # for 1-to-1 results reproducibility with the CompVis implementation.
        # However this currently doesn't work in `mps`.
        latents_shape = (batch_size * num_images_per_prompt, self.unet.in_channels, height // 8, width // 8)
        latents_dtype = self.text_encoder.dtype
        if latents is None:
            if self.device.type == "mps":
                # randn does not work reproducibly on mps
                latents = torch.randn(latents_shape, generator=generator, device="cpu", dtype=latents_dtype).to(
                    self.device
                )
            else:
                latents = torch.randn(latents_shape, generator=generator, device=self.device, dtype=latents_dtype)
        else:
            if latents.shape != latents_shape:
                raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {latents_shape}")
            latents = latents.to(self.device)

        # set timesteps
        self.scheduler.set_timesteps(num_inference_steps)

        # Some schedulers like PNDM have timesteps as arrays
        # It's more optimized to move all timesteps to correct device beforehand
        timesteps_tensor = self.scheduler.timesteps.to(self.device)

        # scale the initial noise by the standard deviation required by the scheduler
        latents = latents * self.scheduler.init_noise_sigma

        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
        # and should be between [0, 1]
        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
        extra_step_kwargs = {}
        if accepts_eta:
            extra_step_kwargs["eta"] = eta

        # check if the scheduler accepts generator
        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
        if accepts_generator:
            extra_step_kwargs["generator"] = generator

        for i, t in enumerate(self.progress_bar(timesteps_tensor)):
            normalized_time = t / self.scheduler.num_train_timesteps
            
            # use the right embeddings based on time
            self.update_embeddings_based_on_time(normalized_time, string_to_param_dict)
            text_embeddings = self.get_embeddings(text_input_ids, batch_size, prompt, negative_prompt, guidance_scale, num_images_per_prompt)
            # expand the latents if we are doing classifier free guidance
            latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
            latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)

            # predict the noise residual
            noise_pred = self.unet(latent_model_input, t, encoder_hidden_states=text_embeddings).sample

            # perform guidance
            if do_classifier_free_guidance:
                noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
                noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)

            # compute the previous noisy sample x_t -> x_t-1
            latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample

            # call the callback, if provided
            if callback is not None and i % callback_steps == 0:
                callback(i, t, latents)

        latents = 1 / 0.18215 * latents
        image = self.vae.decode(latents).sample

        image = (image / 2 + 0.5).clamp(0, 1)

        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloa16
        image = image.cpu().permute(0, 2, 3, 1).float().numpy()

        if self.safety_checker is not None:
            safety_checker_input = self.feature_extractor(self.numpy_to_pil(image), return_tensors="pt").to(
                self.device
            )
            image, has_nsfw_concept = self.safety_checker(
                images=image, clip_input=safety_checker_input.pixel_values.to(text_embeddings.dtype)
            )
        else:
            has_nsfw_concept = None

        if output_type == "pil":
            image = self.numpy_to_pil(image)

        if not return_dict:
            return (image, has_nsfw_concept)

        return image



def find_all(regex, text):
    match_list = []
    while True:
        match  = re.search(regex, text)
        if match:
            match_list.append(match)
            text = text[match.end():]
        else:
            return match_list


class DreamBoothMultiResPipeline(MultiResPipeline):
    @torch.no_grad()
    def __call__(
        self,
        prompt: Union[str, List[str]],
        instance_prompt: str = "S",
        num_scales: int = 10,
        height: int = 512,
        width: int = 512,
        num_inference_steps: int = 50,
        guidance_scale: float = 7.5,
        negative_prompt: Optional[Union[str, List[str]]] = None,
        num_images_per_prompt: Optional[int] = 1,
        eta: float = 0.0,
        generator: Optional[torch.Generator] = None,
        latents: Optional[torch.FloatTensor] = None,
        output_type: Optional[str] = "pil",
        return_dict: bool = True,
        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
        callback_steps: Optional[int] = 1,
        **kwargs,
    ):
        r"""
        Function invoked when calling the pipeline for generation.
        Args:
            prompt (`str` or `List[str]`):
                The prompt or prompts to guide the image generation.
            height (`int`, *optional*, defaults to 512):
                The height in pixels of the generated image.
            width (`int`, *optional*, defaults to 512):
                The width in pixels of the generated image.
            num_inference_steps (`int`, *optional*, defaults to 50):
                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
                expense of slower inference.
            guidance_scale (`float`, *optional*, defaults to 7.5):
                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
                `guidance_scale` is defined as `w` of equation 2. of [Imagen
                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
                usually at the expense of lower image quality.
            negative_prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts not to guide the image generation. Ignored when not using guidance (i.e., ignored
                if `guidance_scale` is less than `1`).
            num_images_per_prompt (`int`, *optional*, defaults to 1):
                The number of images to generate per prompt.
            eta (`float`, *optional*, defaults to 0.0):
                Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
                [`schedulers.DDIMScheduler`], will be ignored for others.
            generator (`torch.Generator`, *optional*):
                A [torch generator](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make generation
                deterministic.
            latents (`torch.FloatTensor`, *optional*):
                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
                tensor will ge generated by sampling using the supplied random `generator`.
            output_type (`str`, *optional*, defaults to `"pil"`):
                The output format of the generate image. Choose between
                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
            return_dict (`bool`, *optional*, defaults to `True`):
                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
                plain tuple.
            callback (`Callable`, *optional*):
                A function that will be called every `callback_steps` steps during inference. The function will be
                called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
            callback_steps (`int`, *optional*, defaults to 1):
                The frequency at which the `callback` function will be called. If not specified, the callback will be
                called at every step.
        Returns:
            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple.
            When returning a tuple, the first element is a list with the generated images, and the second element is a
            list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
            (nsfw) content, according to the `safety_checker`.
        """

        # make sure that tokenizer knows about the special tokens.
        tokenizer = self.tokenizer
        # Add the placeholder token in tokenizer
        for scale_index in range(num_scales):
            placeholder_string = f"<{instance_prompt}|{scale_index}|>"
            increment = tokenizer.add_special_tokens({"additional_special_tokens": [placeholder_string]})



        if isinstance(prompt, str):
            batch_size = 1
        elif isinstance(prompt, list):
            batch_size = len(prompt)
        else:
            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")

        if height % 8 != 0 or width % 8 != 0:
            raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")

        if (callback_steps is None) or (
            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
        ):
            raise ValueError(
                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
                f" {type(callback_steps)}."
            )


        
        do_classifier_free_guidance = guidance_scale > 1.0
        # get the initial random noise unless the user supplied it

        # Unlike in other pipelines, latents need to be generated in the target device
        # for 1-to-1 results reproducibility with the CompVis implementation.
        # However this currently doesn't work in `mps`.
        latents_shape = (batch_size * num_images_per_prompt, self.unet.in_channels, height // 8, width // 8)
        latents_dtype = self.text_encoder.dtype
        if latents is None:
            if self.device.type == "mps":
                # randn does not work reproducibly on mps
                latents = torch.randn(latents_shape, generator=generator, device="cpu", dtype=latents_dtype).to(
                    self.device
                )
            else:
                latents = torch.randn(latents_shape, generator=generator, device=self.device, dtype=latents_dtype)
        else:
            if latents.shape != latents_shape:
                raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {latents_shape}")
            latents = latents.to(self.device)

        # set timesteps
        self.scheduler.set_timesteps(num_inference_steps)

        # Some schedulers like PNDM have timesteps as arrays
        # It's more optimized to move all timesteps to correct device beforehand
        timesteps_tensor = self.scheduler.timesteps.to(self.device)

        # scale the initial noise by the standard deviation required by the scheduler
        latents = latents * self.scheduler.init_noise_sigma

        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
        # and should be between [0, 1]
        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
        extra_step_kwargs = {}
        if accepts_eta:
            extra_step_kwargs["eta"] = eta

        # check if the scheduler accepts generator
        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
        if accepts_generator:
            extra_step_kwargs["generator"] = generator

        for i, t in enumerate(self.progress_bar(timesteps_tensor)):
            normalized_time = t / self.scheduler.num_train_timesteps

            # semi resolution-dependent sampling
            matches = find_all(r"<S\(\d+\)>", prompt)
            new_prompt = copy.deepcopy(prompt)
            for match in matches:

                # extract digits
                number = int(re.search(r"\d+", match.group(0)).group(0))
                if number / num_scales <= normalized_time:
                    new_number = int(normalized_time * num_scales)
                    new_prompt = new_prompt.replace(match.group(0), f"<S|{new_number}|>")
                else:
                    new_prompt = new_prompt.replace(match.group(0), "")

            # fully resolution-dependent sampling
            matches = find_all(r"<S\[\d+\]>", prompt)
            for match in matches:

                # extract digits
                number = int(re.search(r"\d+", match.group(0)).group(0))
                if number / num_scales <= normalized_time:
                    new_number = int(normalized_time * num_scales)
                    new_prompt = new_prompt.replace(match.group(0), f"<S|{new_number}|>")
                else:
                    new_prompt = new_prompt.replace(match.group(0), f"<S|{number}|>")
            
            print(new_prompt)

            # get prompt text embeddings
            text_inputs = self.tokenizer(
                new_prompt,
                padding="max_length",
                max_length=self.tokenizer.model_max_length,
                return_tensors="pt",
            )
            text_input_ids = text_inputs.input_ids

            if text_input_ids.shape[-1] > self.tokenizer.model_max_length:
                removed_text = self.tokenizer.batch_decode(text_input_ids[:, self.tokenizer.model_max_length :])
                print(
                    "The following part of your input was truncated because CLIP can only handle sequences up to"
                    f" {self.tokenizer.model_max_length} tokens: {removed_text}"
                )
                text_input_ids = text_input_ids[:, : self.tokenizer.model_max_length]

            text_embeddings = self.get_embeddings(text_input_ids, batch_size, prompt, negative_prompt, guidance_scale, num_images_per_prompt)
            # expand the latents if we are doing classifier free guidance
            latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
            latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)

            # predict the noise residual
            noise_pred = self.unet(latent_model_input, t, encoder_hidden_states=text_embeddings).sample

            # perform guidance
            if do_classifier_free_guidance:
                noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
                noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)

            # compute the previous noisy sample x_t -> x_t-1
            latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample

            # call the callback, if provided
            if callback is not None and i % callback_steps == 0:
                callback(i, t, latents)

        latents = 1 / 0.18215 * latents
        image = self.vae.decode(latents).sample

        image = (image / 2 + 0.5).clamp(0, 1)

        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloa16
        image = image.cpu().permute(0, 2, 3, 1).float().numpy()

        if self.safety_checker is not None:
            safety_checker_input = self.feature_extractor(self.numpy_to_pil(image), return_tensors="pt").to(
                self.device
            )
            image, has_nsfw_concept = self.safety_checker(
                images=image, clip_input=safety_checker_input.pixel_values.to(text_embeddings.dtype)
            )
        else:
            has_nsfw_concept = None

        if output_type == "pil":
            image = self.numpy_to_pil(image)

        if not return_dict:
            return (image, has_nsfw_concept)

        return image