Upload 24 files

.gitattributes

@@ -0,0 +1,35 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,84 @@
+---
+license: mit
+language:
+- en
+pipeline_tag: text-to-image
+tags:
+- openvino
+- text-to-image
+---
+
+Model Descriptions:
+
+This repo contains OpenVino model files for [SimianLuo's LCM_Dreamshaper_v7](https://huggingface.co/SimianLuo/LCM_Dreamshaper_v7).
+
+Hugging Face Demo on CPU:
+
+[![Hugging Face Spaces](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Spaces-blue)](https://huggingface.co/spaces/deinferno/Latent_Consistency_Model_OpenVino_CPU)
+
+Generation Results:
+
+By converting model to OpenVino format and using Intel(R) Xeon(R) Gold 5220R CPU @ 2.20GHz 24C/48T x 2 we can achieve following results compared to original PyTorch LCM.
+
+Results time includes first compile and reshape phases and should be taken with grain of salt because benchmark was run using 2 socketed server which can underperform in those types of workload. 
+
+Number of images per batch is set to 1
+
+|Run No.|Pytorch|OpenVino|Openvino w/reshape|
+|-------|-------|--------|------------------|
+|1      |15.5841|18.0010 |13.4928           |
+|2      |12.4634|5.0208  |3.6855            |
+|3      |12.1551|4.9462  |3.7228            |
+
+Number of images per batch is set to 4
+
+|Run No.|Pytorch|OpenVino|Openvino w/reshape|
+|-------|-------|--------|------------------|
+|1      |31.3666|33.1488 |25.7044           |
+|2      |33.4797|17.7456 |12.8295           |
+|3      |28.6561|17.9216 |12.7198           |
+
+
+To run the model yourself, you can leverage the 🧨 Diffusers/🤗 Optimum library:
+1. Install the library:
+```
+pip install diffusers transformers accelerate optimum 
+pip install --upgrade-strategy eager optimum[openvino]
+```
+
+2. Clone inference code:
+```
+git clone https://huggingface.co/deinferno/LCM_Dreamshaper_v7-openvino
+cd LCM_Dreamshaper_v7-openvino
+```
+
+2. Run the model:
+```py
+from lcm_ov_pipeline import OVLatentConsistencyModelPipeline
+from lcm_scheduler import LCMScheduler
+
+model_id = "deinferno/LCM_Dreamshaper_v7-openvino"
+
+scheduler = LCMScheduler.from_pretrained(model_id, subfolder = "scheduler")
+
+# Use "compile = True" if you don't plan to reshape and recompile model after loading
+# Don't forget to disable OpenVino cache via "ov_config = {"CACHE_DIR":""}" because optimum won't use it anyway and it will stay as dead weight in your RAM when loading pipeline again
+pipe = OVLatentConsistencyModelPipeline.from_pretrained(model_id, scheduler = scheduler, compile = False, ov_config = {"CACHE_DIR":""})
+
+prompt = "Self-portrait oil painting, a beautiful cyborg with golden hair, 8k"
+
+# Can be set to 1~50 steps. LCM support fast inference even <= 4 steps. Recommend: 1~8 steps.
+
+width = 512
+height = 512
+num_images = 1
+batch_size = 1
+num_inference_steps = 4
+
+# Reshape and recompile for inference speed
+
+pipe.reshape(batch_size=batch_size, height=height, width=width, num_images_per_prompt=num_images)
+pipe.compile()
+
+images = pipe(prompt=prompt, width=width, height=height, num_inference_steps=num_inference_steps, guidance_scale=8.0, output_type="pil").images
+```

convert_to_openvino.py

@@ -0,0 +1,135 @@
+from typing import Dict, Optional, Tuple, OrderedDict
+from transformers import CLIPTextConfig
+from diffusers import UNet2DConditionModel
+
+import torch
+
+from optimum.exporters.onnx.model_configs import VisionOnnxConfig, NormalizedConfig, DummyVisionInputGenerator, DummyTimestepInputGenerator, DummySeq2SeqDecoderTextInputGenerator, DummySeq2SeqDecoderTextInputGenerator
+from optimum.exporters.openvino import main_export
+from optimum.utils.input_generators import DummyInputGenerator, DEFAULT_DUMMY_SHAPES
+from optimum.utils.normalized_config import NormalizedTextConfig
+
+# IMPORTANT: You need to specify some scheduler in downloaded model cache folder to avoid errors
+
+class CustomDummyTimestepInputGenerator(DummyInputGenerator):
+    """
+    Generates dummy time step inputs.
+    """
+
+    SUPPORTED_INPUT_NAMES = (
+        "timestep",
+        "timestep_cond",
+        "text_embeds",
+        "time_ids",
+    )
+
+    def __init__(
+        self,
+        task: str,
+        normalized_config: NormalizedConfig,
+        batch_size: int = DEFAULT_DUMMY_SHAPES["batch_size"],
+        time_cond_proj_dim: int = 256,
+        random_batch_size_range: Optional[Tuple[int, int]] = None,
+        **kwargs,
+    ):
+        self.task = task
+        self.vocab_size = normalized_config.vocab_size
+        self.text_encoder_projection_dim = normalized_config.text_encoder_projection_dim
+        self.time_ids = 5 if normalized_config.requires_aesthetics_score else 6
+        if random_batch_size_range:
+            low, high = random_batch_size_range
+            self.batch_size = random.randint(low, high)
+        else:
+            self.batch_size = batch_size
+        self.time_cond_proj_dim = normalized_config.get("time_cond_proj_dim", time_cond_proj_dim)
+
+    def generate(self, input_name: str, framework: str = "pt", int_dtype: str = "int64", float_dtype: str = "fp32"):
+        shape = [self.batch_size]
+
+        if input_name == "timestep":
+            return self.random_int_tensor(shape, max_value=self.vocab_size, framework=framework, dtype=int_dtype)
+
+        if input_name == "timestep_cond":
+            shape.append(self.time_cond_proj_dim)
+            return self.random_float_tensor(shape, min_value=-1.0, max_value=1.0, framework=framework, dtype=float_dtype)
+
+
+        shape.append(self.text_encoder_projection_dim if input_name == "text_embeds" else self.time_ids)
+        return self.random_float_tensor(shape, max_value=self.vocab_size, framework=framework, dtype=float_dtype)
+
+class LCMUNetOnnxConfig(VisionOnnxConfig):
+    ATOL_FOR_VALIDATION = 1e-3
+    # The ONNX export of a CLIPText architecture, an other Stable Diffusion component, needs the Trilu
+    # operator support, available since opset 14
+    DEFAULT_ONNX_OPSET = 14
+
+    NORMALIZED_CONFIG_CLASS = NormalizedConfig.with_args(
+        image_size="sample_size",
+        num_channels="in_channels",
+        hidden_size="cross_attention_dim",
+        vocab_size="norm_num_groups",
+        allow_new=True,
+    )
+
+    DUMMY_INPUT_GENERATOR_CLASSES = (
+        DummyVisionInputGenerator,
+        CustomDummyTimestepInputGenerator,
+        DummySeq2SeqDecoderTextInputGenerator,
+    )
+
+    @property
+    def inputs(self) -> Dict[str, Dict[int, str]]:
+        common_inputs = OrderedDict({
+            "sample": {0: "batch_size", 1: "num_channels", 2: "height", 3: "width"},
+            "timestep": {0: "steps"},
+            "encoder_hidden_states": {0: "batch_size", 1: "sequence_length"},
+            "timestep_cond": {0: "batch_size"},
+        })
+
+        # TODO : add text_image, image and image_embeds
+        if getattr(self._normalized_config, "addition_embed_type", None) == "text_time":
+            common_inputs["text_embeds"] = {0: "batch_size"}
+            common_inputs["time_ids"] = {0: "batch_size"}
+
+        return common_inputs
+
+    @property
+    def outputs(self) -> Dict[str, Dict[int, str]]:
+        return {
+            "out_sample": {0: "batch_size", 1: "num_channels", 2: "height", 3: "width"},
+        }
+
+    @property
+    def torch_to_onnx_output_map(self) -> Dict[str, str]:
+        return {
+            "sample": "out_sample",
+        }
+
+    def generate_dummy_inputs(self, framework: str = "pt", **kwargs):
+        dummy_inputs = super().generate_dummy_inputs(framework=framework, **kwargs)
+        dummy_inputs["encoder_hidden_states"] = dummy_inputs["encoder_hidden_states"][0]
+
+        if getattr(self._normalized_config, "addition_embed_type", None) == "text_time":
+            dummy_inputs["added_cond_kwargs"] = {
+                "text_embeds": dummy_inputs.pop("text_embeds"),
+                "time_ids": dummy_inputs.pop("time_ids"),
+            }
+
+        return dummy_inputs
+
+    def ordered_inputs(self, model) -> Dict[str, Dict[int, str]]:
+        return self.inputs # Breaks order if timestep_cond involved ( so just copy original one )
+
+model_id = "SimianLuo/LCM_Dreamshaper_v7"
+
+text_encoder_config = CLIPTextConfig.from_pretrained(model_id, subfolder = "text_encoder")
+unet_config = UNet2DConditionModel.from_pretrained(model_id, subfolder = "unet").config
+
+unet_config.text_encoder_projection_dim = text_encoder_config.projection_dim
+unet_config.requires_aesthetics_score = False
+
+custom_onnx_configs = {
+    "unet": LCMUNetOnnxConfig(config = unet_config, task = "semantic-segmentation")
+}
+
+main_export(model_name_or_path = model_id, output = "./", task = "stable-diffusion", fp16 = False, int8 = False, custom_onnx_configs = custom_onnx_configs)

feature_extractor/preprocessor_config.json

@@ -0,0 +1,28 @@
+{
+  "crop_size": {
+    "height": 224,
+    "width": 224
+  },
+  "do_center_crop": true,
+  "do_convert_rgb": true,
+  "do_normalize": true,
+  "do_rescale": true,
+  "do_resize": true,
+  "feature_extractor_type": "CLIPFeatureExtractor",
+  "image_mean": [
+    0.48145466,
+    0.4578275,
+    0.40821073
+  ],
+  "image_processor_type": "CLIPImageProcessor",
+  "image_std": [
+    0.26862954,
+    0.26130258,
+    0.27577711
+  ],
+  "resample": 3,
+  "rescale_factor": 0.00392156862745098,
+  "size": {
+    "shortest_edge": 224
+  }
+}

lcm_ov_pipeline.py

@@ -0,0 +1,388 @@
+import inspect
+
+from pathlib import Path
+from tempfile import TemporaryDirectory
+from typing import List, Optional, Tuple, Union, Dict, Any, Callable, OrderedDict
+
+import numpy as np
+import openvino
+import torch
+
+from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput
+from optimum.intel.openvino.modeling_diffusion import OVStableDiffusionPipeline, OVModelUnet, OVModelVaeDecoder, OVModelTextEncoder, OVModelVaeEncoder, VaeImageProcessor
+from optimum.utils import (
+    DIFFUSION_MODEL_TEXT_ENCODER_2_SUBFOLDER,
+    DIFFUSION_MODEL_TEXT_ENCODER_SUBFOLDER,
+    DIFFUSION_MODEL_UNET_SUBFOLDER,
+    DIFFUSION_MODEL_VAE_DECODER_SUBFOLDER,
+    DIFFUSION_MODEL_VAE_ENCODER_SUBFOLDER,
+)
+
+
+from diffusers import logging
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+class LCMOVModelUnet(OVModelUnet):
+    def __call__(
+        self,
+        sample: np.ndarray,
+        timestep: np.ndarray,
+        encoder_hidden_states: np.ndarray,
+        timestep_cond: Optional[np.ndarray] = None,
+        text_embeds: Optional[np.ndarray] = None,
+        time_ids: Optional[np.ndarray] = None,
+    ):
+        self._compile()
+
+        inputs = {
+            "sample": sample,
+            "timestep": timestep,
+            "encoder_hidden_states": encoder_hidden_states,
+        }
+
+        if timestep_cond is not None:
+            inputs["timestep_cond"] = timestep_cond
+        if text_embeds is not None:
+            inputs["text_embeds"] = text_embeds
+        if time_ids is not None:
+            inputs["time_ids"] = time_ids
+
+        outputs = self.request(inputs, shared_memory=True)
+        return list(outputs.values())
+
+class OVLatentConsistencyModelPipeline(OVStableDiffusionPipeline):
+
+    def __init__(
+        self,
+        vae_decoder: openvino.runtime.Model,
+        text_encoder: openvino.runtime.Model,
+        unet: openvino.runtime.Model,
+        config: Dict[str, Any],
+        tokenizer: "CLIPTokenizer",
+        scheduler: Union["DDIMScheduler", "PNDMScheduler", "LMSDiscreteScheduler"],
+        feature_extractor: Optional["CLIPFeatureExtractor"] = None,
+        vae_encoder: Optional[openvino.runtime.Model] = None,
+        text_encoder_2: Optional[openvino.runtime.Model] = None,
+        tokenizer_2: Optional["CLIPTokenizer"] = None,
+        device: str = "CPU",
+        dynamic_shapes: bool = True,
+        compile: bool = True,
+        ov_config: Optional[Dict[str, str]] = None,
+        model_save_dir: Optional[Union[str, Path, TemporaryDirectory]] = None,
+        **kwargs,
+    ):
+        self._internal_dict = config
+        self._device = device.upper()
+        self.is_dynamic = dynamic_shapes
+        self.ov_config = ov_config if ov_config is not None else {}
+        self._model_save_dir = (
+            Path(model_save_dir.name) if isinstance(model_save_dir, TemporaryDirectory) else model_save_dir
+        )
+        self.vae_decoder = OVModelVaeDecoder(vae_decoder, self)
+        self.unet = LCMOVModelUnet(unet, self)
+        self.text_encoder = OVModelTextEncoder(text_encoder, self) if text_encoder is not None else None
+        self.text_encoder_2 = (
+            OVModelTextEncoder(text_encoder_2, self, model_name=DIFFUSION_MODEL_TEXT_ENCODER_2_SUBFOLDER)
+            if text_encoder_2 is not None
+            else None
+        )
+        self.vae_encoder = OVModelVaeEncoder(vae_encoder, self) if vae_encoder is not None else None
+
+        if "block_out_channels" in self.vae_decoder.config:
+            self.vae_scale_factor = 2 ** (len(self.vae_decoder.config["block_out_channels"]) - 1)
+        else:
+            self.vae_scale_factor = 8
+
+        self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
+
+        self.tokenizer = tokenizer
+        self.tokenizer_2 = tokenizer_2
+        self.scheduler = scheduler
+        self.feature_extractor = feature_extractor
+        self.safety_checker = None
+        self.preprocessors = []
+
+        if self.is_dynamic:
+            self.reshape(batch_size=-1, height=-1, width=-1, num_images_per_prompt=-1)
+
+        if compile:
+            self.compile()
+
+        sub_models = {
+            DIFFUSION_MODEL_TEXT_ENCODER_SUBFOLDER: self.text_encoder,
+            DIFFUSION_MODEL_UNET_SUBFOLDER: self.unet,
+            DIFFUSION_MODEL_VAE_DECODER_SUBFOLDER: self.vae_decoder,
+            DIFFUSION_MODEL_VAE_ENCODER_SUBFOLDER: self.vae_encoder,
+            DIFFUSION_MODEL_TEXT_ENCODER_2_SUBFOLDER: self.text_encoder_2,
+        }
+        for name in sub_models.keys():
+            self._internal_dict[name] = (
+                ("optimum", sub_models[name].__class__.__name__) if sub_models[name] is not None else (None, None)
+            )
+
+        self._internal_dict.pop("vae", None)
+
+    def _reshape_unet(
+        self,
+        model: openvino.runtime.Model,
+        batch_size: int = -1,
+        height: int = -1,
+        width: int = -1,
+        num_images_per_prompt: int = -1,
+        tokenizer_max_length: int = -1,
+    ):  
+        if batch_size == -1 or num_images_per_prompt == -1:
+            batch_size = -1
+        else:
+            batch_size = batch_size * num_images_per_prompt
+
+        height = height // self.vae_scale_factor if height > 0 else height
+        width = width // self.vae_scale_factor if width > 0 else width
+        shapes = {}
+        for inputs in model.inputs:
+            shapes[inputs] = inputs.get_partial_shape()
+            if inputs.get_any_name() == "timestep":
+                shapes[inputs][0] = 1
+            elif inputs.get_any_name() == "sample":
+                in_channels = self.unet.config.get("in_channels", None)
+                if in_channels is None:
+                    in_channels = shapes[inputs][1]
+                    if in_channels.is_dynamic:
+                        logger.warning(
+                            "Could not identify `in_channels` from the unet configuration, to statically reshape the unet please provide a configuration."
+                        ) 
+                        self.is_dynamic = True
+    
+                shapes[inputs] = [batch_size, in_channels, height, width]
+            elif inputs.get_any_name() == "timestep_cond":
+                shapes[inputs] = [batch_size, inputs.get_partial_shape()[1]]
+            elif inputs.get_any_name() == "text_embeds":
+                shapes[inputs] = [batch_size, self.text_encoder_2.config["projection_dim"]]
+            elif inputs.get_any_name() == "time_ids":
+                shapes[inputs] = [batch_size, inputs.get_partial_shape()[1]]
+            else:
+                shapes[inputs][0] = batch_size
+                shapes[inputs][1] = tokenizer_max_length
+        model.reshape(shapes)
+        return model
+
+    def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=np.float32):
+        """
+        see https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
+        Args:
+        timesteps: np.array: generate embedding vectors at these timesteps
+        embedding_dim: int: dimension of the embeddings to generate
+        dtype: data type of the generated embeddings
+
+        Returns:
+        embedding vectors with shape `(len(timesteps), embedding_dim)`
+        """
+        assert len(w.shape) == 1
+        w = w * 1000.
+
+        half_dim = embedding_dim // 2
+        emb = np.log(np.array(10000.)) / (half_dim - 1)
+        emb = np.exp(np.arange(half_dim, dtype=dtype) * -emb)
+        emb = w.astype(dtype)[:, None] * emb[None, :]
+        emb = np.concatenate([np.sin(emb), np.cos(emb)], axis=1)
+        if embedding_dim % 2 == 1:  # zero pad
+            emb = np.pad(emb, (0, 1))
+        assert emb.shape == (w.shape[0], embedding_dim)
+        return emb
+
+    # Adapted from https://github.com/huggingface/optimum/blob/15b8d1eed4d83c5004d3b60f6b6f13744b358f01/optimum/pipelines/diffusers/pipeline_stable_diffusion.py#L201
+    def __call__(
+        self,
+        prompt: Optional[Union[str, List[str]]] = None,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        num_inference_steps: int = 4,
+        original_inference_steps: int = None,
+        guidance_scale: float = 7.5,
+        num_images_per_prompt: int = 1,
+        eta: float = 0.0,
+        generator: Optional[np.random.RandomState] = None,
+        latents: Optional[np.ndarray] = None,
+        prompt_embeds: Optional[np.ndarray] = None,
+        output_type: str = "pil",
+        return_dict: bool = True,
+        callback: Optional[Callable[[int, int, np.ndarray], None]] = None,
+        callback_steps: int = 1,
+        guidance_rescale: float = 0.0,
+    ):
+        r"""
+        Function invoked when calling the pipeline for generation.
+
+        Args:
+            prompt (`Optional[Union[str, List[str]]]`, defaults to None):
+                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+                instead.
+            height (`Optional[int]`, defaults to None):
+                The height in pixels of the generated image.
+            width (`Optional[int]`, defaults to None):
+                The width in pixels of the generated image.
+            num_inference_steps (`int`, defaults to 4):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            original_inference_steps (`int`, *optional*):
+                The original number of inference steps use to generate a linearly-spaced timestep schedule, from which
+                we will draw `num_inference_steps` evenly spaced timesteps from as our final timestep schedule,
+                following the Skipping-Step method in the paper (see Section 4.3). If not set this will default to the
+                scheduler's `original_inference_steps` attribute.
+            guidance_scale (`float`, 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.
+            num_images_per_prompt (`int`, defaults to 1):
+                The number of images to generate per prompt.
+            eta (`float`, 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 (`Optional[np.random.RandomState]`, defaults to `None`)::
+                A np.random.RandomState to make generation deterministic.
+            latents (`Optional[np.ndarray]`, defaults to `None`):
+                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`.
+            prompt_embeds (`Optional[np.ndarray]`, defaults to `None`):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+                provided, text embeddings will be generated from `prompt` input argument.
+            output_type (`str`, 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`, defaults to `True`):
+                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+                plain tuple.
+            callback (Optional[Callable], defaults to `None`):
+                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`, defaults to 1):
+                The frequency at which the `callback` function will be called. If not specified, the callback will be
+                called at every step.
+            guidance_rescale (`float`, defaults to 0.0):
+                Guidance rescale factor proposed by [Common Diffusion Noise Schedules and Sample Steps are
+                Flawed](https://arxiv.org/pdf/2305.08891.pdf) `guidance_scale` is defined as `φ` in equation 16. of
+                [Common Diffusion Noise Schedules and Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+                Guidance rescale factor should fix overexposure when using zero terminal SNR.
+
+        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`.
+        """
+        height = height or self.unet.config.get("sample_size", 64) * self.vae_scale_factor
+        width = width or self.unet.config.get("sample_size", 64) * self.vae_scale_factor
+
+        # check inputs. Raise error if not correct
+        self.check_inputs(
+            prompt, height, width, callback_steps, None, prompt_embeds, None
+        )
+
+        # define call parameters
+        if isinstance(prompt, str):
+            batch_size = 1
+        elif isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
+
+        if generator is None:
+            generator = np.random
+
+        # Create torch.Generator instance with same state as np.random.RandomState
+        torch_generator = torch.Generator().manual_seed(int(generator.get_state()[1][0]))
+
+        #do_classifier_free_guidance = guidance_scale > 1.0
+
+        # NOTE: when a LCM is distilled from an LDM via latent consistency distillation (Algorithm 1) with guided
+        # distillation, the forward pass of the LCM learns to approximate sampling from the LDM using CFG with the
+        # unconditional prompt "" (the empty string). Due to this, LCMs currently do not support negative prompts.
+        prompt_embeds = self._encode_prompt(
+            prompt,
+            num_images_per_prompt,
+            False,
+            negative_prompt=None,
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=None,
+        )
+
+        # set timesteps
+        self.scheduler.set_timesteps(num_inference_steps, "cpu", original_inference_steps=original_inference_steps)
+        timesteps = self.scheduler.timesteps
+
+        latents = self.prepare_latents(
+            batch_size * num_images_per_prompt,
+            self.unet.config.get("in_channels", 4),
+            height,
+            width,
+            prompt_embeds.dtype,
+            generator,
+            latents,
+        )
+
+        # Get Guidance Scale Embedding
+        w = np.tile(guidance_scale - 1, batch_size * num_images_per_prompt)
+        w_embedding = self.get_guidance_scale_embedding(w, embedding_dim=self.unet.config.get("time_cond_proj_dim", 256))
+
+        # Adapted from diffusers to extend it for other runtimes than ORT
+        timestep_dtype = self.unet.input_dtype.get("timestep", np.float32)
+
+        # 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
+
+        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        if accepts_generator:
+            extra_step_kwargs["generator"] = torch_generator
+
+        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+        for i, t in enumerate(self.progress_bar(timesteps)):
+
+            # predict the noise residual
+            timestep = np.array([t], dtype=timestep_dtype)
+            
+            noise_pred = self.unet(sample=latents, timestep=timestep, timestep_cond = w_embedding, encoder_hidden_states=prompt_embeds)[0]
+
+            # compute the previous noisy sample x_t -> x_t-1
+            latents, denoised = self.scheduler.step(
+                torch.from_numpy(noise_pred), t, torch.from_numpy(latents), **extra_step_kwargs, return_dict = False
+            )
+
+            latents, denoised = latents.numpy(), denoised.numpy()
+
+            # call the callback, if provided
+            if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+                if callback is not None and i % callback_steps == 0:
+                    callback(i, t, latents)
+
+        if output_type == "latent":
+            image = latents
+            has_nsfw_concept = None
+        else:
+            denoised /= self.vae_decoder.config.get("scaling_factor", 0.18215)
+            # it seems likes there is a strange result for using half-precision vae decoder if batchsize>1
+            image = np.concatenate(
+                [self.vae_decoder(latent_sample=denoised[i : i + 1])[0] for i in range(latents.shape[0])]
+            )
+            image, has_nsfw_concept = self.run_safety_checker(image)
+
+        if has_nsfw_concept is None:
+            do_denormalize = [True] * image.shape[0]
+        else:
+            do_denormalize = [not has_nsfw for has_nsfw in has_nsfw_concept]
+
+        image = self.image_processor.postprocess(image, output_type=output_type, do_denormalize=do_denormalize)
+
+        if not return_dict:
+            return (image, has_nsfw_concept)
+
+        return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)

lcm_scheduler.py

@@ -0,0 +1,529 @@
+# Copyright 2023 Stanford University Team and 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.
+
+# DISCLAIMER: This code is strongly influenced by https://github.com/pesser/pytorch_diffusion
+# and https://github.com/hojonathanho/diffusion
+
+import math
+from dataclasses import dataclass
+from typing import List, Optional, Tuple, Union
+
+import numpy as np
+import torch
+
+from diffusers.configuration_utils import ConfigMixin, register_to_config
+from diffusers.utils import BaseOutput, logging
+from diffusers.utils.torch_utils import randn_tensor
+from diffusers.schedulers.scheduling_utils import SchedulerMixin
+
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+@dataclass
+class LCMSchedulerOutput(BaseOutput):
+    """
+    Output class for the scheduler's `step` function output.
+
+    Args:
+        prev_sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)` for images):
+            Computed sample `(x_{t-1})` of previous timestep. `prev_sample` should be used as next model input in the
+            denoising loop.
+        pred_original_sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)` for images):
+            The predicted denoised sample `(x_{0})` based on the model output from the current timestep.
+            `pred_original_sample` can be used to preview progress or for guidance.
+    """
+
+    prev_sample: torch.FloatTensor
+    denoised: Optional[torch.FloatTensor] = None
+
+
+# Copied from diffusers.schedulers.scheduling_ddpm.betas_for_alpha_bar
+def betas_for_alpha_bar(
+    num_diffusion_timesteps,
+    max_beta=0.999,
+    alpha_transform_type="cosine",
+):
+    """
+    Create a beta schedule that discretizes the given alpha_t_bar function, which defines the cumulative product of
+    (1-beta) over time from t = [0,1].
+
+    Contains a function alpha_bar that takes an argument t and transforms it to the cumulative product of (1-beta) up
+    to that part of the diffusion process.
+
+
+    Args:
+        num_diffusion_timesteps (`int`): the number of betas to produce.
+        max_beta (`float`): the maximum beta to use; use values lower than 1 to
+                     prevent singularities.
+        alpha_transform_type (`str`, *optional*, default to `cosine`): the type of noise schedule for alpha_bar.
+                     Choose from `cosine` or `exp`
+
+    Returns:
+        betas (`np.ndarray`): the betas used by the scheduler to step the model outputs
+    """
+    if alpha_transform_type == "cosine":
+
+        def alpha_bar_fn(t):
+            return math.cos((t + 0.008) / 1.008 * math.pi / 2) ** 2
+
+    elif alpha_transform_type == "exp":
+
+        def alpha_bar_fn(t):
+            return math.exp(t * -12.0)
+
+    else:
+        raise ValueError(f"Unsupported alpha_tranform_type: {alpha_transform_type}")
+
+    betas = []
+    for i in range(num_diffusion_timesteps):
+        t1 = i / num_diffusion_timesteps
+        t2 = (i + 1) / num_diffusion_timesteps
+        betas.append(min(1 - alpha_bar_fn(t2) / alpha_bar_fn(t1), max_beta))
+    return torch.tensor(betas, dtype=torch.float32)
+
+
+# Copied from diffusers.schedulers.scheduling_ddim.rescale_zero_terminal_snr
+def rescale_zero_terminal_snr(betas: torch.FloatTensor) -> torch.FloatTensor:
+    """
+    Rescales betas to have zero terminal SNR Based on https://arxiv.org/pdf/2305.08891.pdf (Algorithm 1)
+
+
+    Args:
+        betas (`torch.FloatTensor`):
+            the betas that the scheduler is being initialized with.
+
+    Returns:
+        `torch.FloatTensor`: rescaled betas with zero terminal SNR
+    """
+    # Convert betas to alphas_bar_sqrt
+    alphas = 1.0 - betas
+    alphas_cumprod = torch.cumprod(alphas, dim=0)
+    alphas_bar_sqrt = alphas_cumprod.sqrt()
+
+    # Store old values.
+    alphas_bar_sqrt_0 = alphas_bar_sqrt[0].clone()
+    alphas_bar_sqrt_T = alphas_bar_sqrt[-1].clone()
+
+    # Shift so the last timestep is zero.
+    alphas_bar_sqrt -= alphas_bar_sqrt_T
+
+    # Scale so the first timestep is back to the old value.
+    alphas_bar_sqrt *= alphas_bar_sqrt_0 / (alphas_bar_sqrt_0 - alphas_bar_sqrt_T)
+
+    # Convert alphas_bar_sqrt to betas
+    alphas_bar = alphas_bar_sqrt**2  # Revert sqrt
+    alphas = alphas_bar[1:] / alphas_bar[:-1]  # Revert cumprod
+    alphas = torch.cat([alphas_bar[0:1], alphas])
+    betas = 1 - alphas
+
+    return betas
+
+
+class LCMScheduler(SchedulerMixin, ConfigMixin):
+    """
+    `LCMScheduler` extends the denoising procedure introduced in denoising diffusion probabilistic models (DDPMs) with
+    non-Markovian guidance.
+
+    This model inherits from [`SchedulerMixin`] and [`ConfigMixin`]. [`~ConfigMixin`] takes care of storing all config
+    attributes that are passed in the scheduler's `__init__` function, such as `num_train_timesteps`. They can be
+    accessed via `scheduler.config.num_train_timesteps`. [`SchedulerMixin`] provides general loading and saving
+    functionality via the [`SchedulerMixin.save_pretrained`] and [`~SchedulerMixin.from_pretrained`] functions.
+
+    Args:
+        num_train_timesteps (`int`, defaults to 1000):
+            The number of diffusion steps to train the model.
+        beta_start (`float`, defaults to 0.0001):
+            The starting `beta` value of inference.
+        beta_end (`float`, defaults to 0.02):
+            The final `beta` value.
+        beta_schedule (`str`, defaults to `"linear"`):
+            The beta schedule, a mapping from a beta range to a sequence of betas for stepping the model. Choose from
+            `linear`, `scaled_linear`, or `squaredcos_cap_v2`.
+        trained_betas (`np.ndarray`, *optional*):
+            Pass an array of betas directly to the constructor to bypass `beta_start` and `beta_end`.
+        original_inference_steps (`int`, *optional*, defaults to 50):
+            The default number of inference steps used to generate a linearly-spaced timestep schedule, from which we
+            will ultimately take `num_inference_steps` evenly spaced timesteps to form the final timestep schedule.
+        clip_sample (`bool`, defaults to `True`):
+            Clip the predicted sample for numerical stability.
+        clip_sample_range (`float`, defaults to 1.0):
+            The maximum magnitude for sample clipping. Valid only when `clip_sample=True`.
+        set_alpha_to_one (`bool`, defaults to `True`):
+            Each diffusion step uses the alphas product value at that step and at the previous one. For the final step
+            there is no previous alpha. When this option is `True` the previous alpha product is fixed to `1`,
+            otherwise it uses the alpha value at step 0.
+        steps_offset (`int`, defaults to 0):
+            An offset added to the inference steps. You can use a combination of `offset=1` and
+            `set_alpha_to_one=False` to make the last step use step 0 for the previous alpha product like in Stable
+            Diffusion.
+        prediction_type (`str`, defaults to `epsilon`, *optional*):
+            Prediction type of the scheduler function; can be `epsilon` (predicts the noise of the diffusion process),
+            `sample` (directly predicts the noisy sample`) or `v_prediction` (see section 2.4 of [Imagen
+            Video](https://imagen.research.google/video/paper.pdf) paper).
+        thresholding (`bool`, defaults to `False`):
+            Whether to use the "dynamic thresholding" method. This is unsuitable for latent-space diffusion models such
+            as Stable Diffusion.
+        dynamic_thresholding_ratio (`float`, defaults to 0.995):
+            The ratio for the dynamic thresholding method. Valid only when `thresholding=True`.
+        sample_max_value (`float`, defaults to 1.0):
+            The threshold value for dynamic thresholding. Valid only when `thresholding=True`.
+        timestep_spacing (`str`, defaults to `"leading"`):
+            The way the timesteps should be scaled. Refer to Table 2 of the [Common Diffusion Noise Schedules and
+            Sample Steps are Flawed](https://huggingface.co/papers/2305.08891) for more information.
+        rescale_betas_zero_snr (`bool`, defaults to `False`):
+            Whether to rescale the betas to have zero terminal SNR. This enables the model to generate very bright and
+            dark samples instead of limiting it to samples with medium brightness. Loosely related to
+            [`--offset_noise`](https://github.com/huggingface/diffusers/blob/74fd735eb073eb1d774b1ab4154a0876eb82f055/examples/dreambooth/train_dreambooth.py#L506).
+    """
+
+    order = 1
+
+    @register_to_config
+    def __init__(
+        self,
+        num_train_timesteps: int = 1000,
+        beta_start: float = 0.00085,
+        beta_end: float = 0.012,
+        beta_schedule: str = "scaled_linear",
+        trained_betas: Optional[Union[np.ndarray, List[float]]] = None,
+        original_inference_steps: int = 50,
+        clip_sample: bool = False,
+        clip_sample_range: float = 1.0,
+        set_alpha_to_one: bool = True,
+        steps_offset: int = 0,
+        prediction_type: str = "epsilon",
+        thresholding: bool = False,
+        dynamic_thresholding_ratio: float = 0.995,
+        sample_max_value: float = 1.0,
+        timestep_spacing: str = "leading",
+        rescale_betas_zero_snr: bool = False,
+    ):
+        if trained_betas is not None:
+            self.betas = torch.tensor(trained_betas, dtype=torch.float32)
+        elif beta_schedule == "linear":
+            self.betas = torch.linspace(beta_start, beta_end, num_train_timesteps, dtype=torch.float32)
+        elif beta_schedule == "scaled_linear":
+            # this schedule is very specific to the latent diffusion model.
+            self.betas = (
+                torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
+            )
+        elif beta_schedule == "squaredcos_cap_v2":
+            # Glide cosine schedule
+            self.betas = betas_for_alpha_bar(num_train_timesteps)
+        else:
+            raise NotImplementedError(f"{beta_schedule} does is not implemented for {self.__class__}")
+
+        # Rescale for zero SNR
+        if rescale_betas_zero_snr:
+            self.betas = rescale_zero_terminal_snr(self.betas)
+
+        self.alphas = 1.0 - self.betas
+        self.alphas_cumprod = torch.cumprod(self.alphas, dim=0)
+
+        # At every step in ddim, we are looking into the previous alphas_cumprod
+        # For the final step, there is no previous alphas_cumprod because we are already at 0
+        # `set_alpha_to_one` decides whether we set this parameter simply to one or
+        # whether we use the final alpha of the "non-previous" one.
+        self.final_alpha_cumprod = torch.tensor(1.0) if set_alpha_to_one else self.alphas_cumprod[0]
+
+        # standard deviation of the initial noise distribution
+        self.init_noise_sigma = 1.0
+
+        # setable values
+        self.num_inference_steps = None
+        self.timesteps = torch.from_numpy(np.arange(0, num_train_timesteps)[::-1].copy().astype(np.int64))
+
+        self._step_index = None
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._init_step_index
+    def _init_step_index(self, timestep):
+        if isinstance(timestep, torch.Tensor):
+            timestep = timestep.to(self.timesteps.device)
+
+        index_candidates = (self.timesteps == timestep).nonzero()
+
+        # The sigma index that is taken for the **very** first `step`
+        # is always the second index (or the last index if there is only 1)
+        # This way we can ensure we don't accidentally skip a sigma in
+        # case we start in the middle of the denoising schedule (e.g. for image-to-image)
+        if len(index_candidates) > 1:
+            step_index = index_candidates[1]
+        else:
+            step_index = index_candidates[0]
+
+        self._step_index = step_index.item()
+
+    @property
+    def step_index(self):
+        return self._step_index
+
+    def scale_model_input(self, sample: torch.FloatTensor, timestep: Optional[int] = None) -> torch.FloatTensor:
+        """
+        Ensures interchangeability with schedulers that need to scale the denoising model input depending on the
+        current timestep.
+
+        Args:
+            sample (`torch.FloatTensor`):
+                The input sample.
+            timestep (`int`, *optional*):
+                The current timestep in the diffusion chain.
+        Returns:
+            `torch.FloatTensor`:
+                A scaled input sample.
+        """
+        return sample
+
+    # Copied from diffusers.schedulers.scheduling_ddpm.DDPMScheduler._threshold_sample
+    def _threshold_sample(self, sample: torch.FloatTensor) -> torch.FloatTensor:
+        """
+        "Dynamic thresholding: At each sampling step we set s to a certain percentile absolute pixel value in xt0 (the
+        prediction of x_0 at timestep t), and if s > 1, then we threshold xt0 to the range [-s, s] and then divide by
+        s. Dynamic thresholding pushes saturated pixels (those near -1 and 1) inwards, thereby actively preventing
+        pixels from saturation at each step. We find that dynamic thresholding results in significantly better
+        photorealism as well as better image-text alignment, especially when using very large guidance weights."
+
+        https://arxiv.org/abs/2205.11487
+        """
+        dtype = sample.dtype
+        batch_size, channels, *remaining_dims = sample.shape
+
+        if dtype not in (torch.float32, torch.float64):
+            sample = sample.float()  # upcast for quantile calculation, and clamp not implemented for cpu half
+
+        # Flatten sample for doing quantile calculation along each image
+        sample = sample.reshape(batch_size, channels * np.prod(remaining_dims))
+
+        abs_sample = sample.abs()  # "a certain percentile absolute pixel value"
+
+        s = torch.quantile(abs_sample, self.config.dynamic_thresholding_ratio, dim=1)
+        s = torch.clamp(
+            s, min=1, max=self.config.sample_max_value
+        )  # When clamped to min=1, equivalent to standard clipping to [-1, 1]
+        s = s.unsqueeze(1)  # (batch_size, 1) because clamp will broadcast along dim=0
+        sample = torch.clamp(sample, -s, s) / s  # "we threshold xt0 to the range [-s, s] and then divide by s"
+
+        sample = sample.reshape(batch_size, channels, *remaining_dims)
+        sample = sample.to(dtype)
+
+        return sample
+
+    def set_timesteps(
+        self,
+        num_inference_steps: int,
+        device: Union[str, torch.device] = None,
+        original_inference_steps: Optional[int] = None,
+    ):
+        """
+        Sets the discrete timesteps used for the diffusion chain (to be run before inference).
+
+        Args:
+            num_inference_steps (`int`):
+                The number of diffusion steps used when generating samples with a pre-trained model.
+            device (`str` or `torch.device`, *optional*):
+                The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+            original_inference_steps (`int`, *optional*):
+                The original number of inference steps, which will be used to generate a linearly-spaced timestep
+                schedule (which is different from the standard `diffusers` implementation). We will then take
+                `num_inference_steps` timesteps from this schedule, evenly spaced in terms of indices, and use that as
+                our final timestep schedule. If not set, this will default to the `original_inference_steps` attribute.
+        """
+
+        if num_inference_steps > self.config.num_train_timesteps:
+            raise ValueError(
+                f"`num_inference_steps`: {num_inference_steps} cannot be larger than `self.config.train_timesteps`:"
+                f" {self.config.num_train_timesteps} as the unet model trained with this scheduler can only handle"
+                f" maximal {self.config.num_train_timesteps} timesteps."
+            )
+
+        self.num_inference_steps = num_inference_steps
+        original_steps = (
+            original_inference_steps if original_inference_steps is not None else self.original_inference_steps
+        )
+
+        if original_steps > self.config.num_train_timesteps:
+            raise ValueError(
+                f"`original_steps`: {original_steps} cannot be larger than `self.config.train_timesteps`:"
+                f" {self.config.num_train_timesteps} as the unet model trained with this scheduler can only handle"
+                f" maximal {self.config.num_train_timesteps} timesteps."
+            )
+
+        if num_inference_steps > original_steps:
+            raise ValueError(
+                f"`num_inference_steps`: {num_inference_steps} cannot be larger than `original_inference_steps`:"
+                f" {original_steps} because the final timestep schedule will be a subset of the"
+                f" `original_inference_steps`-sized initial timestep schedule."
+            )
+
+        # LCM Timesteps Setting
+        # Currently, only linear spacing is supported.
+        c = self.config.num_train_timesteps // original_steps
+        # LCM Training Steps Schedule
+        lcm_origin_timesteps = np.asarray(list(range(1, original_steps + 1))) * c - 1
+        skipping_step = len(lcm_origin_timesteps) // num_inference_steps
+        # LCM Inference Steps Schedule
+        timesteps = lcm_origin_timesteps[::-skipping_step][:num_inference_steps]
+
+        self.timesteps = torch.from_numpy(timesteps.copy()).to(device=device, dtype=torch.long)
+
+        self._step_index = None
+
+    def get_scalings_for_boundary_condition_discrete(self, t):
+        self.sigma_data = 0.5  # Default: 0.5
+
+        # By dividing 0.1: This is almost a delta function at t=0.
+        c_skip = self.sigma_data**2 / ((t / 0.1) ** 2 + self.sigma_data**2)
+        c_out = (t / 0.1) / ((t / 0.1) ** 2 + self.sigma_data**2) ** 0.5
+        return c_skip, c_out
+
+    def step(
+        self,
+        model_output: torch.FloatTensor,
+        timestep: int,
+        sample: torch.FloatTensor,
+        generator: Optional[torch.Generator] = None,
+        return_dict: bool = True,
+    ) -> Union[LCMSchedulerOutput, Tuple]:
+        """
+        Predict the sample from the previous timestep by reversing the SDE. This function propagates the diffusion
+        process from the learned model outputs (most often the predicted noise).
+
+        Args:
+            model_output (`torch.FloatTensor`):
+                The direct output from learned diffusion model.
+            timestep (`float`):
+                The current discrete timestep in the diffusion chain.
+            sample (`torch.FloatTensor`):
+                A current instance of a sample created by the diffusion process.
+            generator (`torch.Generator`, *optional*):
+                A random number generator.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~schedulers.scheduling_lcm.LCMSchedulerOutput`] or `tuple`.
+        Returns:
+            [`~schedulers.scheduling_utils.LCMSchedulerOutput`] or `tuple`:
+                If return_dict is `True`, [`~schedulers.scheduling_lcm.LCMSchedulerOutput`] is returned, otherwise a
+                tuple is returned where the first element is the sample tensor.
+        """
+        if self.num_inference_steps is None:
+            raise ValueError(
+                "Number of inference steps is 'None', you need to run 'set_timesteps' after creating the scheduler"
+            )
+
+        if self.step_index is None:
+            self._init_step_index(timestep)
+
+        # 1. get previous step value
+        prev_step_index = self.step_index + 1
+        if prev_step_index < len(self.timesteps):
+            prev_timestep = self.timesteps[prev_step_index]
+        else:
+            prev_timestep = timestep
+
+        # 2. compute alphas, betas
+        alpha_prod_t = self.alphas_cumprod[timestep]
+        alpha_prod_t_prev = self.alphas_cumprod[prev_timestep] if prev_timestep >= 0 else self.final_alpha_cumprod
+
+        beta_prod_t = 1 - alpha_prod_t
+        beta_prod_t_prev = 1 - alpha_prod_t_prev
+
+        # 3. Get scalings for boundary conditions
+        c_skip, c_out = self.get_scalings_for_boundary_condition_discrete(timestep)
+
+        # 4. Compute the predicted original sample x_0 based on the model parameterization
+        if self.config.prediction_type == "epsilon":  # noise-prediction
+            predicted_original_sample = (sample - beta_prod_t.sqrt() * model_output) / alpha_prod_t.sqrt()
+        elif self.config.prediction_type == "sample":  # x-prediction
+            predicted_original_sample = model_output
+        elif self.config.prediction_type == "v_prediction":  # v-prediction
+            predicted_original_sample = alpha_prod_t.sqrt() * sample - beta_prod_t.sqrt() * model_output
+        else:
+            raise ValueError(
+                f"prediction_type given as {self.config.prediction_type} must be one of `epsilon`, `sample` or"
+                " `v_prediction` for `LCMScheduler`."
+            )
+
+        # 5. Clip or threshold "predicted x_0"
+        if self.config.thresholding:
+            predicted_original_sample = self._threshold_sample(predicted_original_sample)
+        elif self.config.clip_sample:
+            predicted_original_sample = predicted_original_sample.clamp(
+                -self.config.clip_sample_range, self.config.clip_sample_range
+            )
+
+        # 6. Denoise model output using boundary conditions
+        denoised = c_out * predicted_original_sample + c_skip * sample
+
+        # 7. Sample and inject noise z ~ N(0, I) for MultiStep Inference
+        # Noise is not used for one-step sampling.
+        if len(self.timesteps) > 1:
+            noise = randn_tensor(model_output.shape, generator=generator, device=model_output.device)
+            prev_sample = alpha_prod_t_prev.sqrt() * denoised + beta_prod_t_prev.sqrt() * noise
+        else:
+            prev_sample = denoised
+
+        # upon completion increase step index by one
+        self._step_index += 1
+
+        if not return_dict:
+            return (prev_sample, denoised)
+
+        return LCMSchedulerOutput(prev_sample=prev_sample, denoised=denoised)
+
+    # Copied from diffusers.schedulers.scheduling_ddpm.DDPMScheduler.add_noise
+    def add_noise(
+        self,
+        original_samples: torch.FloatTensor,
+        noise: torch.FloatTensor,
+        timesteps: torch.IntTensor,
+    ) -> torch.FloatTensor:
+        # Make sure alphas_cumprod and timestep have same device and dtype as original_samples
+        alphas_cumprod = self.alphas_cumprod.to(device=original_samples.device, dtype=original_samples.dtype)
+        timesteps = timesteps.to(original_samples.device)
+
+        sqrt_alpha_prod = alphas_cumprod[timesteps] ** 0.5
+        sqrt_alpha_prod = sqrt_alpha_prod.flatten()
+        while len(sqrt_alpha_prod.shape) < len(original_samples.shape):
+            sqrt_alpha_prod = sqrt_alpha_prod.unsqueeze(-1)
+
+        sqrt_one_minus_alpha_prod = (1 - alphas_cumprod[timesteps]) ** 0.5
+        sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.flatten()
+        while len(sqrt_one_minus_alpha_prod.shape) < len(original_samples.shape):
+            sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.unsqueeze(-1)
+
+        noisy_samples = sqrt_alpha_prod * original_samples + sqrt_one_minus_alpha_prod * noise
+        return noisy_samples
+
+    # Copied from diffusers.schedulers.scheduling_ddpm.DDPMScheduler.get_velocity
+    def get_velocity(
+        self, sample: torch.FloatTensor, noise: torch.FloatTensor, timesteps: torch.IntTensor
+    ) -> torch.FloatTensor:
+        # Make sure alphas_cumprod and timestep have same device and dtype as sample
+        alphas_cumprod = self.alphas_cumprod.to(device=sample.device, dtype=sample.dtype)
+        timesteps = timesteps.to(sample.device)
+
+        sqrt_alpha_prod = alphas_cumprod[timesteps] ** 0.5
+        sqrt_alpha_prod = sqrt_alpha_prod.flatten()
+        while len(sqrt_alpha_prod.shape) < len(sample.shape):
+            sqrt_alpha_prod = sqrt_alpha_prod.unsqueeze(-1)
+
+        sqrt_one_minus_alpha_prod = (1 - alphas_cumprod[timesteps]) ** 0.5
+        sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.flatten()
+        while len(sqrt_one_minus_alpha_prod.shape) < len(sample.shape):
+            sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.unsqueeze(-1)
+
+        velocity = sqrt_alpha_prod * noise - sqrt_one_minus_alpha_prod * sample
+        return velocity
+
+    def __len__(self):
+        return self.config.num_train_timesteps

model_index.json

@@ -0,0 +1,34 @@
+{
+  "_class_name": "StableDiffusionPipeline",
+  "_diffusers_version": "0.20.2",
+  "_name_or_path": "SimianLuo/LCM_Dreamshaper_v7",
+  "feature_extractor": [
+    "transformers",
+    "CLIPImageProcessor"
+  ],
+  "requires_safety_checker": true,
+  "safety_checker": [
+    "stable_diffusion",
+    "StableDiffusionSafetyChecker"
+  ],
+  "scheduler": [
+    null,
+    null
+  ],
+  "text_encoder": [
+    "transformers",
+    "CLIPTextModel"
+  ],
+  "tokenizer": [
+    "transformers",
+    "CLIPTokenizer"
+  ],
+  "unet": [
+    "diffusers",
+    "UNet2DConditionModel"
+  ],
+  "vae": [
+    "diffusers",
+    "AutoencoderKL"
+  ]
+}

scheduler/scheduler_config.json

@@ -0,0 +1,20 @@
+{
+  "_class_name": "LCMScheduler",
+  "_diffusers_version": "0.22.0.dev0",
+  "beta_end": 0.012,
+  "beta_schedule": "scaled_linear",
+  "beta_start": 0.00085,
+  "clip_sample": false,
+  "clip_sample_range": 1.0,
+  "dynamic_thresholding_ratio": 0.995,
+  "num_train_timesteps": 1000,
+  "original_inference_steps": 50,
+  "prediction_type": "epsilon",
+  "rescale_betas_zero_snr": false,
+  "sample_max_value": 1.0,
+  "set_alpha_to_one": true,
+  "steps_offset": 0,
+  "thresholding": false,
+  "timestep_spacing": "leading",
+  "trained_betas": null
+}

text_encoder/config.json

@@ -0,0 +1,25 @@
+{
+  "_name_or_path": "/home/user/.cache/huggingface/hub/models--SimianLuo--LCM_Dreamshaper_v7/snapshots/c7f9b672c65a664af57d1de926819fd79cb26eb8/text_encoder",
+  "architectures": [
+    "CLIPTextModel"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 0,
+  "dropout": 0.0,
+  "eos_token_id": 2,
+  "hidden_act": "quick_gelu",
+  "hidden_size": 768,
+  "initializer_factor": 1.0,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-05,
+  "max_position_embeddings": 77,
+  "model_type": "clip_text_model",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 1,
+  "projection_dim": 768,
+  "torch_dtype": "float32",
+  "transformers_version": "4.31.0",
+  "vocab_size": 49408
+}

text_encoder/openvino_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e9e2783c544f9432ab465bc1222eae65b5fa51ebbbb5fb981633c30892dec3a4
+size 492242764

tokenizer/special_tokens_map.json

@@ -0,0 +1,6 @@
+{
+  "bos_token": "<|startoftext|>",
+  "eos_token": "<|endoftext|>",
+  "pad_token": "<|endoftext|>",
+  "unk_token": "<|endoftext|>"
+}

tokenizer/tokenizer_config.json

@@ -0,0 +1,53 @@
+{
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "49406": {
+      "content": "<|startoftext|>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "49407": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [],
+  "bos_token": {
+    "__type": "AddedToken",
+    "content": "<|startoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "clean_up_tokenization_spaces": true,
+  "do_lower_case": true,
+  "eos_token": {
+    "__type": "AddedToken",
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "errors": "replace",
+  "model_max_length": 77,
+  "pad_token": "<|endoftext|>",
+  "tokenizer_class": "CLIPTokenizer",
+  "tokenizer_file": null,
+  "unk_token": {
+    "__type": "AddedToken",
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  }
+}

unet/config.json

@@ -0,0 +1,67 @@
+{
+  "_class_name": "UNet2DConditionModel",
+  "_diffusers_version": "0.20.2",
+  "_name_or_path": "/home/user/.cache/huggingface/hub/models--SimianLuo--LCM_Dreamshaper_v7/snapshots/c7f9b672c65a664af57d1de926819fd79cb26eb8/unet",
+  "act_fn": "silu",
+  "addition_embed_type": null,
+  "addition_embed_type_num_heads": 64,
+  "addition_time_embed_dim": null,
+  "attention_head_dim": 8,
+  "attention_type": "default",
+  "block_out_channels": [
+    320,
+    640,
+    1280,
+    1280
+  ],
+  "center_input_sample": false,
+  "class_embed_type": null,
+  "class_embeddings_concat": false,
+  "conv_in_kernel": 3,
+  "conv_out_kernel": 3,
+  "cross_attention_dim": 768,
+  "cross_attention_norm": null,
+  "down_block_types": [
+    "CrossAttnDownBlock2D",
+    "CrossAttnDownBlock2D",
+    "CrossAttnDownBlock2D",
+    "DownBlock2D"
+  ],
+  "downsample_padding": 1,
+  "dropout": 0.0,
+  "dual_cross_attention": false,
+  "encoder_hid_dim": null,
+  "encoder_hid_dim_type": null,
+  "flip_sin_to_cos": true,
+  "freq_shift": 0,
+  "in_channels": 4,
+  "layers_per_block": 2,
+  "mid_block_only_cross_attention": null,
+  "mid_block_scale_factor": 1,
+  "mid_block_type": "UNetMidBlock2DCrossAttn",
+  "norm_eps": 1e-05,
+  "norm_num_groups": 32,
+  "num_attention_heads": null,
+  "num_class_embeds": null,
+  "only_cross_attention": false,
+  "out_channels": 4,
+  "projection_class_embeddings_input_dim": null,
+  "resnet_out_scale_factor": 1.0,
+  "resnet_skip_time_act": false,
+  "resnet_time_scale_shift": "default",
+  "sample_size": 96,
+  "time_cond_proj_dim": 256,
+  "time_embedding_act_fn": null,
+  "time_embedding_dim": null,
+  "time_embedding_type": "positional",
+  "timestep_post_act": null,
+  "transformer_layers_per_block": 1,
+  "up_block_types": [
+    "UpBlock2D",
+    "CrossAttnUpBlock2D",
+    "CrossAttnUpBlock2D",
+    "CrossAttnUpBlock2D"
+  ],
+  "upcast_attention": null,
+  "use_linear_projection": false
+}

unet/openvino_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc40724121308b865c3ffbae8c39991a2f3510e9251f3b0c59db37833628bc38
+size 3438412388

vae_decoder/config.json

@@ -0,0 +1,32 @@
+{
+  "_class_name": "AutoencoderKL",
+  "_diffusers_version": "0.20.2",
+  "_name_or_path": "/home/user/.cache/huggingface/hub/models--SimianLuo--LCM_Dreamshaper_v7/snapshots/c7f9b672c65a664af57d1de926819fd79cb26eb8/vae",
+  "act_fn": "silu",
+  "block_out_channels": [
+    128,
+    256,
+    512,
+    512
+  ],
+  "down_block_types": [
+    "DownEncoderBlock2D",
+    "DownEncoderBlock2D",
+    "DownEncoderBlock2D",
+    "DownEncoderBlock2D"
+  ],
+  "force_upcast": true,
+  "in_channels": 3,
+  "latent_channels": 4,
+  "layers_per_block": 2,
+  "norm_num_groups": 32,
+  "out_channels": 3,
+  "sample_size": 768,
+  "scaling_factor": 0.18215,
+  "up_block_types": [
+    "UpDecoderBlock2D",
+    "UpDecoderBlock2D",
+    "UpDecoderBlock2D",
+    "UpDecoderBlock2D"
+  ]
+}

vae_decoder/openvino_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:81a8010713a2554061d107967f6a6b301386f661684da510f6035eb379782679
+size 197960932

vae_encoder/config.json

@@ -0,0 +1,32 @@
+{
+  "_class_name": "AutoencoderKL",
+  "_diffusers_version": "0.20.2",
+  "_name_or_path": "/home/user/.cache/huggingface/hub/models--SimianLuo--LCM_Dreamshaper_v7/snapshots/c7f9b672c65a664af57d1de926819fd79cb26eb8/vae",
+  "act_fn": "silu",
+  "block_out_channels": [
+    128,
+    256,
+    512,
+    512
+  ],
+  "down_block_types": [
+    "DownEncoderBlock2D",
+    "DownEncoderBlock2D",
+    "DownEncoderBlock2D",
+    "DownEncoderBlock2D"
+  ],
+  "force_upcast": true,
+  "in_channels": 3,
+  "latent_channels": 4,
+  "layers_per_block": 2,
+  "norm_num_groups": 32,
+  "out_channels": 3,
+  "sample_size": 768,
+  "scaling_factor": 0.18215,
+  "up_block_types": [
+    "UpDecoderBlock2D",
+    "UpDecoderBlock2D",
+    "UpDecoderBlock2D",
+    "UpDecoderBlock2D"
+  ]
+}

vae_encoder/openvino_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f0198e0db626c241c992904f9a34be172e895028c7dcbb63c66ced4e8e496a8a
+size 136654796