diffusers/pipelines/stable_diffusion/pipeline_flax_stable_diffusion.py

# Copyright 2023 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import warnings
from functools import partial
from typing import Dict, List, Optional, Union

import jax
import jax.numpy as jnp
import numpy as np
from flax.core.frozen_dict import FrozenDict
from flax.jax_utils import unreplicate
from flax.training.common_utils import shard
from packaging import version
from PIL import Image
from transformers import CLIPFeatureExtractor, CLIPTokenizer, FlaxCLIPTextModel

from ...models import FlaxAutoencoderKL, FlaxUNet2DConditionModel
from ...schedulers import (
    FlaxDDIMScheduler,
    FlaxDPMSolverMultistepScheduler,
    FlaxLMSDiscreteScheduler,
    FlaxPNDMScheduler,
)
from ...utils import deprecate, logging
from ..pipeline_flax_utils import FlaxDiffusionPipeline
from . import FlaxStableDiffusionPipelineOutput
from .safety_checker_flax import FlaxStableDiffusionSafetyChecker


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

# Set to True to use python for loop instead of jax.fori_loop for easier debugging
DEBUG = False


class FlaxStableDiffusionPipeline(FlaxDiffusionPipeline):
    r"""
    Pipeline for text-to-image generation using Stable Diffusion.

    This model inherits from [`FlaxDiffusionPipeline`]. 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 ([`FlaxAutoencoderKL`]):
            Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
        text_encoder ([`FlaxCLIPTextModel`]):
            Frozen text-encoder. Stable Diffusion uses the text portion of
            [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.FlaxCLIPTextModel),
            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 ([`FlaxUNet2DConditionModel`]): 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 latents. Can be one of
            [`FlaxDDIMScheduler`], [`FlaxLMSDiscreteScheduler`], [`FlaxPNDMScheduler`], or
            [`FlaxDPMSolverMultistepScheduler`].
        safety_checker ([`FlaxStableDiffusionSafetyChecker`]):
            Classification module that estimates whether generated images could be considered offensive or harmful.
            Please, refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) 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: FlaxAutoencoderKL,
        text_encoder: FlaxCLIPTextModel,
        tokenizer: CLIPTokenizer,
        unet: FlaxUNet2DConditionModel,
        scheduler: Union[
            FlaxDDIMScheduler, FlaxPNDMScheduler, FlaxLMSDiscreteScheduler, FlaxDPMSolverMultistepScheduler
        ],
        safety_checker: FlaxStableDiffusionSafetyChecker,
        feature_extractor: CLIPFeatureExtractor,
        dtype: jnp.dtype = jnp.float32,
    ):
        super().__init__()
        self.dtype = dtype

        if safety_checker is None:
            logger.warning(
                f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
                " results in services or applications open to the public. Both the diffusers team and Hugging Face"
                " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling"
                " it only for use-cases that involve analyzing network behavior or auditing its results. For more"
                " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ."
            )

        is_unet_version_less_0_9_0 = hasattr(unet.config, "_diffusers_version") and version.parse(
            version.parse(unet.config._diffusers_version).base_version
        ) < version.parse("0.9.0.dev0")
        is_unet_sample_size_less_64 = hasattr(unet.config, "sample_size") and unet.config.sample_size < 64
        if is_unet_version_less_0_9_0 and is_unet_sample_size_less_64:
            deprecation_message = (
                "The configuration file of the unet has set the default `sample_size` to smaller than"
                " 64 which seems highly unlikely .If you're checkpoint is a fine-tuned version of any of the"
                " following: \n- CompVis/stable-diffusion-v1-4 \n- CompVis/stable-diffusion-v1-3 \n-"
                " CompVis/stable-diffusion-v1-2 \n- CompVis/stable-diffusion-v1-1 \n- runwayml/stable-diffusion-v1-5"
                " \n- runwayml/stable-diffusion-inpainting \n you should change 'sample_size' to 64 in the"
                " configuration file. Please make sure to update the config accordingly as leaving `sample_size=32`"
                " in the config might lead to incorrect results in future versions. If you have downloaded this"
                " checkpoint from the Hugging Face Hub, it would be very nice if you could open a Pull request for"
                " the `unet/config.json` file"
            )
            deprecate("sample_size<64", "1.0.0", deprecation_message, standard_warn=False)
            new_config = dict(unet.config)
            new_config["sample_size"] = 64
            unet._internal_dict = FrozenDict(new_config)

        self.register_modules(
            vae=vae,
            text_encoder=text_encoder,
            tokenizer=tokenizer,
            unet=unet,
            scheduler=scheduler,
            safety_checker=safety_checker,
            feature_extractor=feature_extractor,
        )
        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)

    def prepare_inputs(self, prompt: Union[str, List[str]]):
        if not isinstance(prompt, (str, list)):
            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")

        text_input = self.tokenizer(
            prompt,
            padding="max_length",
            max_length=self.tokenizer.model_max_length,
            truncation=True,
            return_tensors="np",
        )
        return text_input.input_ids

    def _get_has_nsfw_concepts(self, features, params):
        has_nsfw_concepts = self.safety_checker(features, params)
        return has_nsfw_concepts

    def _run_safety_checker(self, images, safety_model_params, jit=False):
        # safety_model_params should already be replicated when jit is True
        pil_images = [Image.fromarray(image) for image in images]
        features = self.feature_extractor(pil_images, return_tensors="np").pixel_values

        if jit:
            features = shard(features)
            has_nsfw_concepts = _p_get_has_nsfw_concepts(self, features, safety_model_params)
            has_nsfw_concepts = unshard(has_nsfw_concepts)
            safety_model_params = unreplicate(safety_model_params)
        else:
            has_nsfw_concepts = self._get_has_nsfw_concepts(features, safety_model_params)

        images_was_copied = False
        for idx, has_nsfw_concept in enumerate(has_nsfw_concepts):
            if has_nsfw_concept:
                if not images_was_copied:
                    images_was_copied = True
                    images = images.copy()

                #images[idx] = np.zeros(images[idx].shape, dtype=np.uint8)  # black image

            if any(has_nsfw_concepts):
                warnings.warn(
                    "Potential NSFW content was detected in one or more images. A black image 3 will be returned"
                    " instead. Try again with a different prompt and/or seed."
                )

        return images, has_nsfw_concepts

    def _generate(
        self,
        prompt_ids: jnp.array,
        params: Union[Dict, FrozenDict],
        prng_seed: jax.random.KeyArray,
        num_inference_steps: int,
        height: int,
        width: int,
        guidance_scale: float,
        latents: Optional[jnp.array] = None,
        neg_prompt_ids: Optional[jnp.array] = None,
    ):
        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}.")

        # get prompt text embeddings
        prompt_embeds = self.text_encoder(prompt_ids, params=params["text_encoder"])[0]

        # TODO: currently it is assumed `do_classifier_free_guidance = guidance_scale > 1.0`
        # implement this conditional `do_classifier_free_guidance = guidance_scale > 1.0`
        batch_size = prompt_ids.shape[0]

        max_length = prompt_ids.shape[-1]

        if neg_prompt_ids is None:
            uncond_input = self.tokenizer(
                [""] * batch_size, padding="max_length", max_length=max_length, return_tensors="np"
            ).input_ids
        else:
            uncond_input = neg_prompt_ids
        negative_prompt_embeds = self.text_encoder(uncond_input, params=params["text_encoder"])[0]
        context = jnp.concatenate([negative_prompt_embeds, prompt_embeds])

        latents_shape = (
            batch_size,
            self.unet.in_channels,
            height // self.vae_scale_factor,
            width // self.vae_scale_factor,
        )
        if latents is None:
            latents = jax.random.normal(prng_seed, shape=latents_shape, dtype=jnp.float32)
        else:
            if latents.shape != latents_shape:
                raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {latents_shape}")

        def loop_body(step, args):
            latents, scheduler_state = args
            # 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
            latents_input = jnp.concatenate([latents] * 2)

            t = jnp.array(scheduler_state.timesteps, dtype=jnp.int32)[step]
            timestep = jnp.broadcast_to(t, latents_input.shape[0])

            latents_input = self.scheduler.scale_model_input(scheduler_state, latents_input, t)

            # predict the noise residual
            noise_pred = self.unet.apply(
                {"params": params["unet"]},
                jnp.array(latents_input),
                jnp.array(timestep, dtype=jnp.int32),
                encoder_hidden_states=context,
            ).sample
            # perform guidance
            noise_pred_uncond, noise_prediction_text = jnp.split(noise_pred, 2, axis=0)
            noise_pred = noise_pred_uncond + guidance_scale * (noise_prediction_text - noise_pred_uncond)

            # compute the previous noisy sample x_t -> x_t-1
            latents, scheduler_state = self.scheduler.step(scheduler_state, noise_pred, t, latents).to_tuple()
            return latents, scheduler_state

        scheduler_state = self.scheduler.set_timesteps(
            params["scheduler"], num_inference_steps=num_inference_steps, shape=latents.shape
        )

        # scale the initial noise by the standard deviation required by the scheduler
        latents = latents * params["scheduler"].init_noise_sigma

        if DEBUG:
            # run with python for loop
            for i in range(num_inference_steps):
                latents, scheduler_state = loop_body(i, (latents, scheduler_state))
        else:
            latents, _ = jax.lax.fori_loop(0, num_inference_steps, loop_body, (latents, scheduler_state))

        # scale and decode the image latents with vae
        latents = 1 / self.vae.config.scaling_factor * latents
        image = self.vae.apply({"params": params["vae"]}, latents, method=self.vae.decode).sample

        image = (image / 2 + 0.5).clip(0, 1).transpose(0, 2, 3, 1)
        return image

    def __call__(
        self,
        prompt_ids: jnp.array,
        params: Union[Dict, FrozenDict],
        prng_seed: jax.random.KeyArray,
        num_inference_steps: int = 50,
        height: Optional[int] = None,
        width: Optional[int] = None,
        guidance_scale: Union[float, jnp.array] = 7.5,
        latents: jnp.array = None,
        neg_prompt_ids: jnp.array = None,
        return_dict: bool = True,
        jit: bool = False,
    ):
        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 self.unet.config.sample_size * self.vae_scale_factor):
                The height in pixels of the generated image.
            width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
                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.
            latents (`jnp.array`, *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. tensor will ge generated
                by sampling using the supplied random `generator`.
            jit (`bool`, defaults to `False`):
                Whether to run `pmap` versions of the generation and safety scoring functions. NOTE: This argument
                exists because `__call__` is not yet end-to-end pmap-able. It will be removed in a future release.
            return_dict (`bool`, *optional*, defaults to `True`):
                Whether or not to return a [`~pipelines.stable_diffusion.FlaxStableDiffusionPipelineOutput`] instead of
                a plain tuple.

        Returns:
            [`~pipelines.stable_diffusion.FlaxStableDiffusionPipelineOutput`] or `tuple`:
            [`~pipelines.stable_diffusion.FlaxStableDiffusionPipelineOutput`] 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`.
        """
        # 0. Default height and width to unet
        height = height or self.unet.config.sample_size * self.vae_scale_factor
        width = width or self.unet.config.sample_size * self.vae_scale_factor

        if isinstance(guidance_scale, float):
            # Convert to a tensor so each device gets a copy. Follow the prompt_ids for
            # shape information, as they may be sharded (when `jit` is `True`), or not.
            guidance_scale = jnp.array([guidance_scale] * prompt_ids.shape[0])
            if len(prompt_ids.shape) > 2:
                # Assume sharded
                guidance_scale = guidance_scale[:, None]

        if jit:
            images = _p_generate(
                self,
                prompt_ids,
                params,
                prng_seed,
                num_inference_steps,
                height,
                width,
                guidance_scale,
                latents,
                neg_prompt_ids,
            )
        else:
            images = self._generate(
                prompt_ids,
                params,
                prng_seed,
                num_inference_steps,
                height,
                width,
                guidance_scale,
                latents,
                neg_prompt_ids,
            )

        if self.safety_checker is not None:
            safety_params = params["safety_checker"]
            images_uint8_casted = (images * 255).round().astype("uint8")
            num_devices, batch_size = images.shape[:2]

            images_uint8_casted = np.asarray(images_uint8_casted).reshape(num_devices * batch_size, height, width, 3)
            images_uint8_casted, has_nsfw_concept = self._run_safety_checker(images_uint8_casted, safety_params, jit)
            images = np.asarray(images)

            # block images
            if any(has_nsfw_concept):
                for i, is_nsfw in enumerate(has_nsfw_concept):
                    if is_nsfw:
                        images[i] = np.asarray(images_uint8_casted[i])

            images = images.reshape(num_devices, batch_size, height, width, 3)
        else:
            images = np.asarray(images)
            has_nsfw_concept = False

        if not return_dict:
            return (images, has_nsfw_concept)

        return FlaxStableDiffusionPipelineOutput(images=images, nsfw_content_detected=has_nsfw_concept)


# Static argnums are pipe, num_inference_steps, height, width. A change would trigger recompilation.
# Non-static args are (sharded) input tensors mapped over their first dimension (hence, `0`).
@partial(
    jax.pmap,
    in_axes=(None, 0, 0, 0, None, None, None, 0, 0, 0),
    static_broadcasted_argnums=(0, 4, 5, 6),
)
def _p_generate(
    pipe,
    prompt_ids,
    params,
    prng_seed,
    num_inference_steps,
    height,
    width,
    guidance_scale,
    latents,
    neg_prompt_ids,
):
    return pipe._generate(
        prompt_ids,
        params,
        prng_seed,
        num_inference_steps,
        height,
        width,
        guidance_scale,
        latents,
        neg_prompt_ids,
    )


@partial(jax.pmap, static_broadcasted_argnums=(0,))
def _p_get_has_nsfw_concepts(pipe, features, params):
    return pipe._get_has_nsfw_concepts(features, params)


def unshard(x: jnp.ndarray):
    # einops.rearrange(x, 'd b ... -> (d b) ...')
    num_devices, batch_size = x.shape[:2]
    rest = x.shape[2:]
    return x.reshape(num_devices * batch_size, *rest)