add pipeline-with-callback

based on the proposal in diffusers#521
https://github.com/huggingface/diffusers/pull/521/files#diff-ab952f41078da66b9fcbbd913b419f8c334badceefac03a5f7edcd6dd986a8ef

reset diffusers requirement to main repo; specify versions for other
various dependencies

app.py

@@ -1,8 +1,7 @@
 import gradio as gr
 
 import torch
-from torch import autocast
-from diffusers import StableDiffusionPipeline
+from pipeline_with_callback import StableDiffusionPipelineWithCallback
 from datasets import load_dataset
 from PIL import Image  
 import re
@@ -11,7 +10,7 @@ model_id = "CompVis/stable-diffusion-v1-4"
 device = "cuda"
 
 #If you are running this code locally, you need to either do a 'huggingface-cli login` or paste your User Access Token from here https://huggingface.co/settings/tokens into the use_auth_token field below. 
-pipe = StableDiffusionPipeline.from_pretrained(model_id, use_auth_token=True, revision="fp16", torch_dtype=torch.float16)
+pipe = StableDiffusionPipelineWithCallback.from_pretrained(model_id, use_auth_token=True, revision="fp16", torch_dtype=torch.float16)
 pipe = pipe.to(device)
 torch.backends.cudnn.benchmark = True
 
@@ -299,4 +298,4 @@ Despite how impressive being able to turn text into image is, beware to the fact
            """
         )
 
-block.queue(max_size=25).launch()
+block.queue(max_size=25).launch()

pipeline_with_callback.py

@@ -0,0 +1,335 @@
+import inspect
+import warnings
+from typing import Callable, List, Optional, Tuple, Union
+
+import numpy as np
+import torch
+
+from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
+
+from diffusers.models import AutoencoderKL, UNet2DConditionModel
+from diffusers.pipeline_utils import DiffusionPipeline
+from diffusers.schedulers import DDIMScheduler, LMSDiscreteScheduler, PNDMScheduler
+from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput
+from diffusers.pipelines.stable_diffusion.safety_checker import StableDiffusionSafetyChecker
+
+
+class StableDiffusionPipelineWithCallback(DiffusionPipeline):
+    r"""
+    Pipeline for text-to-image generation using Stable Diffusion.
+
+    ** based on https://github.com/huggingface/diffusers/pull/521/files#diff-ab952f41078da66b9fcbbd913b419f8c334badceefac03a5f7edcd6dd986a8ef **
+
+    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
+    library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
+
+    Args:
+        vae ([`AutoencoderKL`]):
+            Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
+        text_encoder ([`CLIPTextModel`]):
+            Frozen text-encoder. Stable Diffusion uses the text portion of
+            [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
+            the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
+        tokenizer (`CLIPTokenizer`):
+            Tokenizer of class
+            [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
+        unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents.
+        scheduler ([`SchedulerMixin`]):
+            A scheduler to be used in combination with `unet` to denoise the encoded image latens. Can be one of
+            [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
+        safety_checker ([`StableDiffusionSafetyChecker`]):
+            Classification module that estimates whether generated images could be considered offsensive or harmful.
+            Please, refer to the [model card](https://huggingface.co/CompVis/stable-diffusion-v1-4) for details.
+        feature_extractor ([`CLIPFeatureExtractor`]):
+            Model that extracts features from generated images to be used as inputs for the `safety_checker`.
+    """
+
+    def __init__(
+        self,
+        vae: AutoencoderKL,
+        text_encoder: CLIPTextModel,
+        tokenizer: CLIPTokenizer,
+        unet: UNet2DConditionModel,
+        scheduler: Union[DDIMScheduler, PNDMScheduler, LMSDiscreteScheduler],
+        safety_checker: StableDiffusionSafetyChecker,
+        feature_extractor: CLIPFeatureExtractor,
+    ):
+        super().__init__()
+        scheduler = scheduler.set_format("pt")
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            unet=unet,
+            scheduler=scheduler,
+            safety_checker=safety_checker,
+            feature_extractor=feature_extractor,
+        )
+
+    def enable_attention_slicing(self, slice_size: Optional[Union[str, int]] = "auto"):
+        r"""
+        Enable sliced attention computation.
+
+        When this option is enabled, the attention module will split the input tensor in slices, to compute attention
+        in several steps. This is useful to save some memory in exchange for a small speed decrease.
+
+        Args:
+            slice_size (`str` or `int`, *optional*, defaults to `"auto"`):
+                When `"auto"`, halves the input to the attention heads, so attention will be computed in two steps. If
+                a number is provided, uses as many slices as `attention_head_dim // slice_size`. In this case,
+                `attention_head_dim` must be a multiple of `slice_size`.
+        """
+        if slice_size == "auto":
+            # half the attention head size is usually a good trade-off between
+            # speed and memory
+            slice_size = self.unet.config.attention_head_dim // 2
+        self.unet.set_attention_slice(slice_size)
+
+    def disable_attention_slicing(self):
+        r"""
+        Disable sliced attention computation. If `enable_attention_slicing` was previously invoked, this method will go
+        back to computing attention in one step.
+        """
+        # set slice_size = `None` to disable `attention slicing`
+        self.enable_attention_slicing(None)
+
+    @torch.no_grad()
+    def decode_latents(self, latents: torch.FloatTensor) -> np.ndarray:
+        r"""
+        Scale and decode the latent representations into images using the VAE.
+
+        Args:
+            latents (`torch.FloatTensor`):
+                Latent representations to decode into images.
+
+        Returns:
+            `np.ndarray`: Decoded images.
+        """
+        latents = 1 / 0.18215 * latents
+        image = self.vae.decode(latents).sample
+
+        image = (image / 2 + 0.5).clamp(0, 1)
+        image = image.cpu().permute(0, 2, 3, 1).numpy()
+        return image
+
+    @torch.no_grad()
+    def run_safety_checker(self, image: np.ndarray) -> Tuple[np.ndarray, List[bool]]:
+        r"""
+        Run the safety checker on the generated images. If potential NSFW content was detected, a warning will be
+        raised and a black image will be returned instead.
+
+        Args:
+            image (`np.ndarray`):
+                Images to run the safety checker on.
+
+        Returns:
+            `Tuple[np.ndarray, List[bool]]`: The first element contains the images that has been processed by the
+            safety checker. The second element is a boolean array indicating whether the images contain NSFW content.
+        """
+        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)
+        return image, has_nsfw_concept
+
+    @torch.no_grad()
+    def __call__(
+        self,
+        prompt: Union[str, List[str]],
+        height: Optional[int] = 512,
+        width: Optional[int] = 512,
+        num_inference_steps: Optional[int] = 50,
+        guidance_scale: Optional[float] = 7.5,
+        eta: Optional[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, np.ndarray, torch.FloatTensor], None]
+        ] = None,
+        callback_frequency: 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.
+            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 `nd.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_frequency` steps during inference. The function will be
+                called with the following arguments: `callback(step: int, timestep: np.ndarray, latents:
+                torch.FloatTensor, image: Union[List[PIL.Image.Image], np.ndarray])`.
+            callback_frequency (`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 "torch_device" in kwargs:
+            device = kwargs.pop("torch_device")
+            warnings.warn(
+                "`torch_device` is deprecated as an input argument to `__call__` and will be removed in v0.3.0."
+                " Consider using `pipe.to(torch_device)` instead."
+            )
+
+            # Set device as before (to be removed in 0.3.0)
+            if device is None:
+                device = "cuda" if torch.cuda.is_available() else "cpu"
+            self.to(device)
+
+        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_frequency is None) or (
+            callback_frequency is not None and (not isinstance(callback_frequency, int) or callback_frequency <= 0)
+        ):
+            raise ValueError(
+                f"`callback_frequency` has to be a positive integer but is {callback_frequency} of type"
+                f" {type(callback_frequency)}."
+            )
+
+        # get prompt text embeddings
+        text_input = self.tokenizer(
+            prompt,
+            padding="max_length",
+            max_length=self.tokenizer.model_max_length,
+            truncation=True,
+            return_tensors="pt",
+        )
+        text_embeddings = self.text_encoder(text_input.input_ids.to(self.device))[0]
+
+        # 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:
+            max_length = text_input.input_ids.shape[-1]
+            uncond_input = self.tokenizer(
+                [""] * batch_size, padding="max_length", max_length=max_length, return_tensors="pt"
+            )
+            uncond_embeddings = self.text_encoder(uncond_input.input_ids.to(self.device))[0]
+
+            # 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])
+
+        # 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_device = "cpu" if self.device.type == "mps" else self.device
+        latents_shape = (batch_size, self.unet.in_channels, height // 8, width // 8)
+        if latents is None:
+            latents = torch.randn(
+                latents_shape,
+                generator=generator,
+                device=latents_device,
+            )
+        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
+        accepts_offset = "offset" in set(inspect.signature(self.scheduler.set_timesteps).parameters.keys())
+        extra_set_kwargs = {}
+        if accepts_offset:
+            extra_set_kwargs["offset"] = 1
+
+        self.scheduler.set_timesteps(num_inference_steps, **extra_set_kwargs)
+
+        # if we use LMSDiscreteScheduler, let's make sure latents are mulitplied by sigmas
+        if isinstance(self.scheduler, LMSDiscreteScheduler):
+            latents = latents * self.scheduler.sigmas[0]
+
+        # 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
+
+        for i, t in enumerate(self.progress_bar(self.scheduler.timesteps)):
+            # expand the latents if we are doing classifier free guidance
+            latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+            if isinstance(self.scheduler, LMSDiscreteScheduler):
+                sigma = self.scheduler.sigmas[i]
+                # the model input needs to be scaled to match the continuous ODE formulation in K-LMS
+                latent_model_input = latent_model_input / ((sigma**2 + 1) ** 0.5)
+
+            # 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
+            if isinstance(self.scheduler, LMSDiscreteScheduler):
+                latents = self.scheduler.step(noise_pred, i, latents, **extra_step_kwargs).prev_sample
+            else:
+                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_frequency == 0:
+                callback(i, t, latents)
+
+        image = self.decode_latents(latents)
+
+        image, has_nsfw_concept = self.run_safety_checker(image)
+
+        if output_type == "pil":
+            image = self.numpy_to_pil(image)
+
+        if not return_dict:
+            return (image, has_nsfw_concept)
+
+        return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)

requirements.txt

@@ -1,6 +1,10 @@
--e git+https://github.com/Narsil/diffusers.git@5ad7ed8d3eb2216b8f8c45d3d1fca929882441be#egg=diffusers
-transformers
-nvidia-ml-py3
+-e git+https://github.com/huggingface/diffusers.git@429dace10a356a776f935fc11e16d5b321b496f3#egg=diffusers
+datasets~=2.4.0
 ftfy
---extra-index-url https://download.pytorch.org/whl/cu113 
-torch
+gradio~=3.3.1
+numpy~=1.23.2
+nvidia-ml-py3
+Pillow~=9.2.0
+transformers~=4.21.3
+--extra-index-url https://download.pytorch.org/whl/cu113
+torch~=1.12.1