Add files using upload-large-folder tool

pythonProject/.venv/Lib/site-packages/diffusers/schedulers/scheduling_dpm_cogvideox.py

@@ -0,0 +1,489 @@
+# Copyright 2025 The CogVideoX team, Tsinghua University & ZhipuAI 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 ..configuration_utils import ConfigMixin, register_to_config
+from ..utils import BaseOutput
+from ..utils.torch_utils import randn_tensor
+from .scheduling_utils import KarrasDiffusionSchedulers, SchedulerMixin
+
+
+@dataclass
+# Copied from diffusers.schedulers.scheduling_ddpm.DDPMSchedulerOutput with DDPM->DDIM
+class DDIMSchedulerOutput(BaseOutput):
+    """
+    Output class for the scheduler's `step` function output.
+
+    Args:
+        prev_sample (`torch.Tensor` 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.Tensor` 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.Tensor
+    pred_original_sample: Optional[torch.Tensor] = 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_transform_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)
+
+
+def rescale_zero_terminal_snr(alphas_cumprod):
+    """
+    Rescales betas to have zero terminal SNR Based on https://huggingface.co/papers/2305.08891 (Algorithm 1)
+
+
+    Args:
+        betas (`torch.Tensor`):
+            the betas that the scheduler is being initialized with.
+
+    Returns:
+        `torch.Tensor`: rescaled betas with zero terminal SNR
+    """
+
+    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
+
+    return alphas_bar
+
+
+class CogVideoXDPMScheduler(SchedulerMixin, ConfigMixin):
+    """
+    `DDIMScheduler` extends the denoising procedure introduced in denoising diffusion probabilistic models (DDPMs) with
+    non-Markovian guidance.
+
+    This model inherits from [`SchedulerMixin`] and [`ConfigMixin`]. Check the superclass documentation for the generic
+    methods the library implements for all schedulers such as loading and saving.
+
+    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`.
+        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, as required by some model families.
+        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).
+    """
+
+    _compatibles = [e.name for e in KarrasDiffusionSchedulers]
+    order = 1
+
+    @register_to_config
+    def __init__(
+        self,
+        num_train_timesteps: int = 1000,
+        beta_start: float = 0.00085,
+        beta_end: float = 0.0120,
+        beta_schedule: str = "scaled_linear",
+        trained_betas: Optional[Union[np.ndarray, List[float]]] = None,
+        clip_sample: bool = True,
+        set_alpha_to_one: bool = True,
+        steps_offset: int = 0,
+        prediction_type: str = "epsilon",
+        clip_sample_range: float = 1.0,
+        sample_max_value: float = 1.0,
+        timestep_spacing: str = "leading",
+        rescale_betas_zero_snr: bool = False,
+        snr_shift_scale: float = 3.0,
+    ):
+        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.float64) ** 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} is not implemented for {self.__class__}")
+
+        self.alphas = 1.0 - self.betas
+        self.alphas_cumprod = torch.cumprod(self.alphas, dim=0)
+
+        # Modify: SNR shift following SD3
+        self.alphas_cumprod = self.alphas_cumprod / (snr_shift_scale + (1 - snr_shift_scale) * self.alphas_cumprod)
+
+        # Rescale for zero SNR
+        if rescale_betas_zero_snr:
+            self.alphas_cumprod = rescale_zero_terminal_snr(self.alphas_cumprod)
+
+        # 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))
+
+    def _get_variance(self, timestep, prev_timestep):
+        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
+
+        variance = (beta_prod_t_prev / beta_prod_t) * (1 - alpha_prod_t / alpha_prod_t_prev)
+
+        return variance
+
+    def scale_model_input(self, sample: torch.Tensor, timestep: Optional[int] = None) -> torch.Tensor:
+        """
+        Ensures interchangeability with schedulers that need to scale the denoising model input depending on the
+        current timestep.
+
+        Args:
+            sample (`torch.Tensor`):
+                The input sample.
+            timestep (`int`, *optional*):
+                The current timestep in the diffusion chain.
+
+        Returns:
+            `torch.Tensor`:
+                A scaled input sample.
+        """
+        return sample
+
+    def set_timesteps(self, num_inference_steps: int, device: Union[str, torch.device] = 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.
+        """
+
+        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
+
+        # "linspace", "leading", "trailing" corresponds to annotation of Table 2. of https://huggingface.co/papers/2305.08891
+        if self.config.timestep_spacing == "linspace":
+            timesteps = (
+                np.linspace(0, self.config.num_train_timesteps - 1, num_inference_steps)
+                .round()[::-1]
+                .copy()
+                .astype(np.int64)
+            )
+        elif self.config.timestep_spacing == "leading":
+            step_ratio = self.config.num_train_timesteps // self.num_inference_steps
+            # creates integer timesteps by multiplying by ratio
+            # casting to int to avoid issues when num_inference_step is power of 3
+            timesteps = (np.arange(0, num_inference_steps) * step_ratio).round()[::-1].copy().astype(np.int64)
+            timesteps += self.config.steps_offset
+        elif self.config.timestep_spacing == "trailing":
+            step_ratio = self.config.num_train_timesteps / self.num_inference_steps
+            # creates integer timesteps by multiplying by ratio
+            # casting to int to avoid issues when num_inference_step is power of 3
+            timesteps = np.round(np.arange(self.config.num_train_timesteps, 0, -step_ratio)).astype(np.int64)
+            timesteps -= 1
+        else:
+            raise ValueError(
+                f"{self.config.timestep_spacing} is not supported. Please make sure to choose one of 'leading' or 'trailing'."
+            )
+
+        self.timesteps = torch.from_numpy(timesteps).to(device)
+
+    def get_variables(self, alpha_prod_t, alpha_prod_t_prev, alpha_prod_t_back=None):
+        lamb = ((alpha_prod_t / (1 - alpha_prod_t)) ** 0.5).log()
+        lamb_next = ((alpha_prod_t_prev / (1 - alpha_prod_t_prev)) ** 0.5).log()
+        h = lamb_next - lamb
+
+        if alpha_prod_t_back is not None:
+            lamb_previous = ((alpha_prod_t_back / (1 - alpha_prod_t_back)) ** 0.5).log()
+            h_last = lamb - lamb_previous
+            r = h_last / h
+            return h, r, lamb, lamb_next
+        else:
+            return h, None, lamb, lamb_next
+
+    def get_mult(self, h, r, alpha_prod_t, alpha_prod_t_prev, alpha_prod_t_back):
+        mult1 = ((1 - alpha_prod_t_prev) / (1 - alpha_prod_t)) ** 0.5 * (-h).exp()
+        mult2 = (-2 * h).expm1() * alpha_prod_t_prev**0.5
+
+        if alpha_prod_t_back is not None:
+            mult3 = 1 + 1 / (2 * r)
+            mult4 = 1 / (2 * r)
+            return mult1, mult2, mult3, mult4
+        else:
+            return mult1, mult2
+
+    def step(
+        self,
+        model_output: torch.Tensor,
+        old_pred_original_sample: torch.Tensor,
+        timestep: int,
+        timestep_back: int,
+        sample: torch.Tensor,
+        eta: float = 0.0,
+        use_clipped_model_output: bool = False,
+        generator=None,
+        variance_noise: Optional[torch.Tensor] = None,
+        return_dict: bool = False,
+    ) -> Union[DDIMSchedulerOutput, 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.Tensor`):
+                The direct output from learned diffusion model.
+            timestep (`float`):
+                The current discrete timestep in the diffusion chain.
+            sample (`torch.Tensor`):
+                A current instance of a sample created by the diffusion process.
+            eta (`float`):
+                The weight of noise for added noise in diffusion step.
+            use_clipped_model_output (`bool`, defaults to `False`):
+                If `True`, computes "corrected" `model_output` from the clipped predicted original sample. Necessary
+                because predicted original sample is clipped to [-1, 1] when `self.config.clip_sample` is `True`. If no
+                clipping has happened, "corrected" `model_output` would coincide with the one provided as input and
+                `use_clipped_model_output` has no effect.
+            generator (`torch.Generator`, *optional*):
+                A random number generator.
+            variance_noise (`torch.Tensor`):
+                Alternative to generating noise with `generator` by directly providing the noise for the variance
+                itself. Useful for methods such as [`CycleDiffusion`].
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~schedulers.scheduling_ddim.DDIMSchedulerOutput`] or `tuple`.
+
+        Returns:
+            [`~schedulers.scheduling_ddim.DDIMSchedulerOutput`] or `tuple`:
+                If return_dict is `True`, [`~schedulers.scheduling_ddim.DDIMSchedulerOutput`] 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"
+            )
+
+        # See formulas (12) and (16) of DDIM paper https://huggingface.co/papers/2010.02502
+        # Ideally, read DDIM paper in-detail understanding
+
+        # Notation (<variable name> -> <name in paper>
+        # - pred_noise_t -> e_theta(x_t, t)
+        # - pred_original_sample -> f_theta(x_t, t) or x_0
+        # - std_dev_t -> sigma_t
+        # - eta -> η
+        # - pred_sample_direction -> "direction pointing to x_t"
+        # - pred_prev_sample -> "x_t-1"
+
+        # 1. get previous step value (=t-1)
+        prev_timestep = timestep - self.config.num_train_timesteps // self.num_inference_steps
+
+        # 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
+        alpha_prod_t_back = self.alphas_cumprod[timestep_back] if timestep_back is not None else None
+
+        beta_prod_t = 1 - alpha_prod_t
+
+        # 3. compute predicted original sample from predicted noise also called
+        # "predicted x_0" of formula (12) from https://huggingface.co/papers/2010.02502
+        # To make style tests pass, commented out `pred_epsilon` as it is an unused variable
+        if self.config.prediction_type == "epsilon":
+            pred_original_sample = (sample - beta_prod_t ** (0.5) * model_output) / alpha_prod_t ** (0.5)
+            # pred_epsilon = model_output
+        elif self.config.prediction_type == "sample":
+            pred_original_sample = model_output
+            # pred_epsilon = (sample - alpha_prod_t ** (0.5) * pred_original_sample) / beta_prod_t ** (0.5)
+        elif self.config.prediction_type == "v_prediction":
+            pred_original_sample = (alpha_prod_t**0.5) * sample - (beta_prod_t**0.5) * model_output
+            # pred_epsilon = (alpha_prod_t**0.5) * model_output + (beta_prod_t**0.5) * sample
+        else:
+            raise ValueError(
+                f"prediction_type given as {self.config.prediction_type} must be one of `epsilon`, `sample`, or"
+                " `v_prediction`"
+            )
+
+        h, r, lamb, lamb_next = self.get_variables(alpha_prod_t, alpha_prod_t_prev, alpha_prod_t_back)
+        mult = list(self.get_mult(h, r, alpha_prod_t, alpha_prod_t_prev, alpha_prod_t_back))
+        mult_noise = (1 - alpha_prod_t_prev) ** 0.5 * (1 - (-2 * h).exp()) ** 0.5
+
+        noise = randn_tensor(sample.shape, generator=generator, device=sample.device, dtype=sample.dtype)
+        prev_sample = mult[0] * sample - mult[1] * pred_original_sample + mult_noise * noise
+
+        if old_pred_original_sample is None or prev_timestep < 0:
+            # Save a network evaluation if all noise levels are 0 or on the first step
+            return prev_sample, pred_original_sample
+        else:
+            denoised_d = mult[2] * pred_original_sample - mult[3] * old_pred_original_sample
+            noise = randn_tensor(sample.shape, generator=generator, device=sample.device, dtype=sample.dtype)
+            x_advanced = mult[0] * sample - mult[1] * denoised_d + mult_noise * noise
+
+            prev_sample = x_advanced
+
+        if not return_dict:
+            return (prev_sample, pred_original_sample)
+
+        return DDIMSchedulerOutput(prev_sample=prev_sample, pred_original_sample=pred_original_sample)
+
+    # Copied from diffusers.schedulers.scheduling_ddpm.DDPMScheduler.add_noise
+    def add_noise(
+        self,
+        original_samples: torch.Tensor,
+        noise: torch.Tensor,
+        timesteps: torch.IntTensor,
+    ) -> torch.Tensor:
+        # Make sure alphas_cumprod and timestep have same device and dtype as original_samples
+        # Move the self.alphas_cumprod to device to avoid redundant CPU to GPU data movement
+        # for the subsequent add_noise calls
+        self.alphas_cumprod = self.alphas_cumprod.to(device=original_samples.device)
+        alphas_cumprod = self.alphas_cumprod.to(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.Tensor, noise: torch.Tensor, timesteps: torch.IntTensor) -> torch.Tensor:
+        # Make sure alphas_cumprod and timestep have same device and dtype as sample
+        self.alphas_cumprod = self.alphas_cumprod.to(device=sample.device)
+        alphas_cumprod = self.alphas_cumprod.to(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

pythonProject/.venv/Lib/site-packages/diffusers/schedulers/scheduling_edm_dpmsolver_multistep.py

@@ -0,0 +1,712 @@
+# Copyright 2025 TSAIL 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 file is strongly influenced by https://github.com/LuChengTHU/dpm-solver and https://github.com/NVlabs/edm
+
+import math
+from typing import List, Optional, Tuple, Union
+
+import numpy as np
+import torch
+
+from ..configuration_utils import ConfigMixin, register_to_config
+from ..utils.torch_utils import randn_tensor
+from .scheduling_utils import SchedulerMixin, SchedulerOutput
+
+
+class EDMDPMSolverMultistepScheduler(SchedulerMixin, ConfigMixin):
+    """
+    Implements DPMSolverMultistepScheduler in EDM formulation as presented in Karras et al. 2022 [1].
+    `EDMDPMSolverMultistepScheduler` is a fast dedicated high-order solver for diffusion ODEs.
+
+    [1] Karras, Tero, et al. "Elucidating the Design Space of Diffusion-Based Generative Models."
+    https://huggingface.co/papers/2206.00364
+
+    This model inherits from [`SchedulerMixin`] and [`ConfigMixin`]. Check the superclass documentation for the generic
+    methods the library implements for all schedulers such as loading and saving.
+
+    Args:
+        sigma_min (`float`, *optional*, defaults to 0.002):
+            Minimum noise magnitude in the sigma schedule. This was set to 0.002 in the EDM paper [1]; a reasonable
+            range is [0, 10].
+        sigma_max (`float`, *optional*, defaults to 80.0):
+            Maximum noise magnitude in the sigma schedule. This was set to 80.0 in the EDM paper [1]; a reasonable
+            range is [0.2, 80.0].
+        sigma_data (`float`, *optional*, defaults to 0.5):
+            The standard deviation of the data distribution. This is set to 0.5 in the EDM paper [1].
+        sigma_schedule (`str`, *optional*, defaults to `karras`):
+            Sigma schedule to compute the `sigmas`. By default, we the schedule introduced in the EDM paper
+            (https://huggingface.co/papers/2206.00364). Other acceptable value is "exponential". The exponential
+            schedule was incorporated in this model: https://huggingface.co/stabilityai/cosxl.
+        num_train_timesteps (`int`, defaults to 1000):
+            The number of diffusion steps to train the model.
+        solver_order (`int`, defaults to 2):
+            The DPMSolver order which can be `1` or `2` or `3`. It is recommended to use `solver_order=2` for guided
+            sampling, and `solver_order=3` for unconditional sampling.
+        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` and
+            `algorithm_type="dpmsolver++"`.
+        algorithm_type (`str`, defaults to `dpmsolver++`):
+            Algorithm type for the solver; can be `dpmsolver++` or `sde-dpmsolver++`. The `dpmsolver++` type implements
+            the algorithms in the [DPMSolver++](https://huggingface.co/papers/2211.01095) paper. It is recommended to
+            use `dpmsolver++` or `sde-dpmsolver++` with `solver_order=2` for guided sampling like in Stable Diffusion.
+        solver_type (`str`, defaults to `midpoint`):
+            Solver type for the second-order solver; can be `midpoint` or `heun`. The solver type slightly affects the
+            sample quality, especially for a small number of steps. It is recommended to use `midpoint` solvers.
+        lower_order_final (`bool`, defaults to `True`):
+            Whether to use lower-order solvers in the final steps. Only valid for < 15 inference steps. This can
+            stabilize the sampling of DPMSolver for steps < 15, especially for steps <= 10.
+        euler_at_final (`bool`, defaults to `False`):
+            Whether to use Euler's method in the final step. It is a trade-off between numerical stability and detail
+            richness. This can stabilize the sampling of the SDE variant of DPMSolver for small number of inference
+            steps, but sometimes may result in blurring.
+        final_sigmas_type (`str`, defaults to `"zero"`):
+            The final `sigma` value for the noise schedule during the sampling process. If `"sigma_min"`, the final
+            sigma is the same as the last sigma in the training schedule. If `zero`, the final sigma is set to 0.
+    """
+
+    _compatibles = []
+    order = 1
+
+    @register_to_config
+    def __init__(
+        self,
+        sigma_min: float = 0.002,
+        sigma_max: float = 80.0,
+        sigma_data: float = 0.5,
+        sigma_schedule: str = "karras",
+        num_train_timesteps: int = 1000,
+        prediction_type: str = "epsilon",
+        rho: float = 7.0,
+        solver_order: int = 2,
+        thresholding: bool = False,
+        dynamic_thresholding_ratio: float = 0.995,
+        sample_max_value: float = 1.0,
+        algorithm_type: str = "dpmsolver++",
+        solver_type: str = "midpoint",
+        lower_order_final: bool = True,
+        euler_at_final: bool = False,
+        final_sigmas_type: Optional[str] = "zero",  # "zero", "sigma_min"
+    ):
+        # settings for DPM-Solver
+        if algorithm_type not in ["dpmsolver++", "sde-dpmsolver++"]:
+            if algorithm_type == "deis":
+                self.register_to_config(algorithm_type="dpmsolver++")
+            else:
+                raise NotImplementedError(f"{algorithm_type} is not implemented for {self.__class__}")
+
+        if solver_type not in ["midpoint", "heun"]:
+            if solver_type in ["logrho", "bh1", "bh2"]:
+                self.register_to_config(solver_type="midpoint")
+            else:
+                raise NotImplementedError(f"{solver_type} is not implemented for {self.__class__}")
+
+        if algorithm_type not in ["dpmsolver++", "sde-dpmsolver++"] and final_sigmas_type == "zero":
+            raise ValueError(
+                f"`final_sigmas_type` {final_sigmas_type} is not supported for `algorithm_type` {algorithm_type}. Please choose `sigma_min` instead."
+            )
+
+        ramp = torch.linspace(0, 1, num_train_timesteps)
+        if sigma_schedule == "karras":
+            sigmas = self._compute_karras_sigmas(ramp)
+        elif sigma_schedule == "exponential":
+            sigmas = self._compute_exponential_sigmas(ramp)
+
+        self.timesteps = self.precondition_noise(sigmas)
+
+        self.sigmas = torch.cat([sigmas, torch.zeros(1, device=sigmas.device)])
+
+        # setable values
+        self.num_inference_steps = None
+        self.model_outputs = [None] * solver_order
+        self.lower_order_nums = 0
+        self._step_index = None
+        self._begin_index = None
+        self.sigmas = self.sigmas.to("cpu")  # to avoid too much CPU/GPU communication
+
+    @property
+    def init_noise_sigma(self):
+        # standard deviation of the initial noise distribution
+        return (self.config.sigma_max**2 + 1) ** 0.5
+
+    @property
+    def step_index(self):
+        """
+        The index counter for current timestep. It will increase 1 after each scheduler step.
+        """
+        return self._step_index
+
+    @property
+    def begin_index(self):
+        """
+        The index for the first timestep. It should be set from pipeline with `set_begin_index` method.
+        """
+        return self._begin_index
+
+    # Copied from diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler.set_begin_index
+    def set_begin_index(self, begin_index: int = 0):
+        """
+        Sets the begin index for the scheduler. This function should be run from pipeline before the inference.
+
+        Args:
+            begin_index (`int`):
+                The begin index for the scheduler.
+        """
+        self._begin_index = begin_index
+
+    # Copied from diffusers.schedulers.scheduling_edm_euler.EDMEulerScheduler.precondition_inputs
+    def precondition_inputs(self, sample, sigma):
+        c_in = self._get_conditioning_c_in(sigma)
+        scaled_sample = sample * c_in
+        return scaled_sample
+
+    # Copied from diffusers.schedulers.scheduling_edm_euler.EDMEulerScheduler.precondition_noise
+    def precondition_noise(self, sigma):
+        if not isinstance(sigma, torch.Tensor):
+            sigma = torch.tensor([sigma])
+
+        c_noise = 0.25 * torch.log(sigma)
+
+        return c_noise
+
+    # Copied from diffusers.schedulers.scheduling_edm_euler.EDMEulerScheduler.precondition_outputs
+    def precondition_outputs(self, sample, model_output, sigma):
+        sigma_data = self.config.sigma_data
+        c_skip = sigma_data**2 / (sigma**2 + sigma_data**2)
+
+        if self.config.prediction_type == "epsilon":
+            c_out = sigma * sigma_data / (sigma**2 + sigma_data**2) ** 0.5
+        elif self.config.prediction_type == "v_prediction":
+            c_out = -sigma * sigma_data / (sigma**2 + sigma_data**2) ** 0.5
+        else:
+            raise ValueError(f"Prediction type {self.config.prediction_type} is not supported.")
+
+        denoised = c_skip * sample + c_out * model_output
+
+        return denoised
+
+    # Copied from diffusers.schedulers.scheduling_edm_euler.EDMEulerScheduler.scale_model_input
+    def scale_model_input(self, sample: torch.Tensor, timestep: Union[float, torch.Tensor]) -> torch.Tensor:
+        """
+        Ensures interchangeability with schedulers that need to scale the denoising model input depending on the
+        current timestep. Scales the denoising model input by `(sigma**2 + 1) ** 0.5` to match the Euler algorithm.
+
+        Args:
+            sample (`torch.Tensor`):
+                The input sample.
+            timestep (`int`, *optional*):
+                The current timestep in the diffusion chain.
+
+        Returns:
+            `torch.Tensor`:
+                A scaled input sample.
+        """
+        if self.step_index is None:
+            self._init_step_index(timestep)
+
+        sigma = self.sigmas[self.step_index]
+        sample = self.precondition_inputs(sample, sigma)
+
+        self.is_scale_input_called = True
+        return sample
+
+    def set_timesteps(self, num_inference_steps: int = None, device: Union[str, torch.device] = 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.
+        """
+
+        self.num_inference_steps = num_inference_steps
+
+        ramp = torch.linspace(0, 1, self.num_inference_steps)
+        if self.config.sigma_schedule == "karras":
+            sigmas = self._compute_karras_sigmas(ramp)
+        elif self.config.sigma_schedule == "exponential":
+            sigmas = self._compute_exponential_sigmas(ramp)
+
+        sigmas = sigmas.to(dtype=torch.float32, device=device)
+        self.timesteps = self.precondition_noise(sigmas)
+
+        if self.config.final_sigmas_type == "sigma_min":
+            sigma_last = self.config.sigma_min
+        elif self.config.final_sigmas_type == "zero":
+            sigma_last = 0
+        else:
+            raise ValueError(
+                f"`final_sigmas_type` must be one of 'zero', or 'sigma_min', but got {self.config.final_sigmas_type}"
+            )
+
+        self.sigmas = torch.cat([sigmas, torch.tensor([sigma_last], dtype=torch.float32, device=device)])
+
+        self.model_outputs = [
+            None,
+        ] * self.config.solver_order
+        self.lower_order_nums = 0
+
+        # add an index counter for schedulers that allow duplicated timesteps
+        self._step_index = None
+        self._begin_index = None
+        self.sigmas = self.sigmas.to("cpu")  # to avoid too much CPU/GPU communication
+
+    # Copied from diffusers.schedulers.scheduling_edm_euler.EDMEulerScheduler._compute_karras_sigmas
+    def _compute_karras_sigmas(self, ramp, sigma_min=None, sigma_max=None) -> torch.Tensor:
+        """Constructs the noise schedule of Karras et al. (2022)."""
+        sigma_min = sigma_min or self.config.sigma_min
+        sigma_max = sigma_max or self.config.sigma_max
+
+        rho = self.config.rho
+        min_inv_rho = sigma_min ** (1 / rho)
+        max_inv_rho = sigma_max ** (1 / rho)
+        sigmas = (max_inv_rho + ramp * (min_inv_rho - max_inv_rho)) ** rho
+        return sigmas
+
+    # Copied from diffusers.schedulers.scheduling_edm_euler.EDMEulerScheduler._compute_exponential_sigmas
+    def _compute_exponential_sigmas(self, ramp, sigma_min=None, sigma_max=None) -> torch.Tensor:
+        """Implementation closely follows k-diffusion.
+
+        https://github.com/crowsonkb/k-diffusion/blob/6ab5146d4a5ef63901326489f31f1d8e7dd36b48/k_diffusion/sampling.py#L26
+        """
+        sigma_min = sigma_min or self.config.sigma_min
+        sigma_max = sigma_max or self.config.sigma_max
+        sigmas = torch.linspace(math.log(sigma_min), math.log(sigma_max), len(ramp)).exp().flip(0)
+        return sigmas
+
+    # Copied from diffusers.schedulers.scheduling_ddpm.DDPMScheduler._threshold_sample
+    def _threshold_sample(self, sample: torch.Tensor) -> torch.Tensor:
+        """
+        "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://huggingface.co/papers/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
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._sigma_to_t
+    def _sigma_to_t(self, sigma, log_sigmas):
+        # get log sigma
+        log_sigma = np.log(np.maximum(sigma, 1e-10))
+
+        # get distribution
+        dists = log_sigma - log_sigmas[:, np.newaxis]
+
+        # get sigmas range
+        low_idx = np.cumsum((dists >= 0), axis=0).argmax(axis=0).clip(max=log_sigmas.shape[0] - 2)
+        high_idx = low_idx + 1
+
+        low = log_sigmas[low_idx]
+        high = log_sigmas[high_idx]
+
+        # interpolate sigmas
+        w = (low - log_sigma) / (low - high)
+        w = np.clip(w, 0, 1)
+
+        # transform interpolation to time range
+        t = (1 - w) * low_idx + w * high_idx
+        t = t.reshape(sigma.shape)
+        return t
+
+    def _sigma_to_alpha_sigma_t(self, sigma):
+        alpha_t = torch.tensor(1)  # Inputs are pre-scaled before going into unet, so alpha_t = 1
+        sigma_t = sigma
+
+        return alpha_t, sigma_t
+
+    def convert_model_output(
+        self,
+        model_output: torch.Tensor,
+        sample: torch.Tensor = None,
+    ) -> torch.Tensor:
+        """
+        Convert the model output to the corresponding type the DPMSolver/DPMSolver++ algorithm needs. DPM-Solver is
+        designed to discretize an integral of the noise prediction model, and DPM-Solver++ is designed to discretize an
+        integral of the data prediction model.
+
+        <Tip>
+
+        The algorithm and model type are decoupled. You can use either DPMSolver or DPMSolver++ for both noise
+        prediction and data prediction models.
+
+        </Tip>
+
+        Args:
+            model_output (`torch.Tensor`):
+                The direct output from the learned diffusion model.
+            sample (`torch.Tensor`):
+                A current instance of a sample created by the diffusion process.
+
+        Returns:
+            `torch.Tensor`:
+                The converted model output.
+        """
+        sigma = self.sigmas[self.step_index]
+        x0_pred = self.precondition_outputs(sample, model_output, sigma)
+
+        if self.config.thresholding:
+            x0_pred = self._threshold_sample(x0_pred)
+
+        return x0_pred
+
+    def dpm_solver_first_order_update(
+        self,
+        model_output: torch.Tensor,
+        sample: torch.Tensor = None,
+        noise: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
+        """
+        One step for the first-order DPMSolver (equivalent to DDIM).
+
+        Args:
+            model_output (`torch.Tensor`):
+                The direct output from the learned diffusion model.
+            sample (`torch.Tensor`):
+                A current instance of a sample created by the diffusion process.
+
+        Returns:
+            `torch.Tensor`:
+                The sample tensor at the previous timestep.
+        """
+        sigma_t, sigma_s = self.sigmas[self.step_index + 1], self.sigmas[self.step_index]
+        alpha_t, sigma_t = self._sigma_to_alpha_sigma_t(sigma_t)
+        alpha_s, sigma_s = self._sigma_to_alpha_sigma_t(sigma_s)
+        lambda_t = torch.log(alpha_t) - torch.log(sigma_t)
+        lambda_s = torch.log(alpha_s) - torch.log(sigma_s)
+
+        h = lambda_t - lambda_s
+        if self.config.algorithm_type == "dpmsolver++":
+            x_t = (sigma_t / sigma_s) * sample - (alpha_t * (torch.exp(-h) - 1.0)) * model_output
+        elif self.config.algorithm_type == "sde-dpmsolver++":
+            assert noise is not None
+            x_t = (
+                (sigma_t / sigma_s * torch.exp(-h)) * sample
+                + (alpha_t * (1 - torch.exp(-2.0 * h))) * model_output
+                + sigma_t * torch.sqrt(1.0 - torch.exp(-2 * h)) * noise
+            )
+
+        return x_t
+
+    def multistep_dpm_solver_second_order_update(
+        self,
+        model_output_list: List[torch.Tensor],
+        sample: torch.Tensor = None,
+        noise: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
+        """
+        One step for the second-order multistep DPMSolver.
+
+        Args:
+            model_output_list (`List[torch.Tensor]`):
+                The direct outputs from learned diffusion model at current and latter timesteps.
+            sample (`torch.Tensor`):
+                A current instance of a sample created by the diffusion process.
+
+        Returns:
+            `torch.Tensor`:
+                The sample tensor at the previous timestep.
+        """
+        sigma_t, sigma_s0, sigma_s1 = (
+            self.sigmas[self.step_index + 1],
+            self.sigmas[self.step_index],
+            self.sigmas[self.step_index - 1],
+        )
+
+        alpha_t, sigma_t = self._sigma_to_alpha_sigma_t(sigma_t)
+        alpha_s0, sigma_s0 = self._sigma_to_alpha_sigma_t(sigma_s0)
+        alpha_s1, sigma_s1 = self._sigma_to_alpha_sigma_t(sigma_s1)
+
+        lambda_t = torch.log(alpha_t) - torch.log(sigma_t)
+        lambda_s0 = torch.log(alpha_s0) - torch.log(sigma_s0)
+        lambda_s1 = torch.log(alpha_s1) - torch.log(sigma_s1)
+
+        m0, m1 = model_output_list[-1], model_output_list[-2]
+
+        h, h_0 = lambda_t - lambda_s0, lambda_s0 - lambda_s1
+        r0 = h_0 / h
+        D0, D1 = m0, (1.0 / r0) * (m0 - m1)
+        if self.config.algorithm_type == "dpmsolver++":
+            # See https://huggingface.co/papers/2211.01095 for detailed derivations
+            if self.config.solver_type == "midpoint":
+                x_t = (
+                    (sigma_t / sigma_s0) * sample
+                    - (alpha_t * (torch.exp(-h) - 1.0)) * D0
+                    - 0.5 * (alpha_t * (torch.exp(-h) - 1.0)) * D1
+                )
+            elif self.config.solver_type == "heun":
+                x_t = (
+                    (sigma_t / sigma_s0) * sample
+                    - (alpha_t * (torch.exp(-h) - 1.0)) * D0
+                    + (alpha_t * ((torch.exp(-h) - 1.0) / h + 1.0)) * D1
+                )
+        elif self.config.algorithm_type == "sde-dpmsolver++":
+            assert noise is not None
+            if self.config.solver_type == "midpoint":
+                x_t = (
+                    (sigma_t / sigma_s0 * torch.exp(-h)) * sample
+                    + (alpha_t * (1 - torch.exp(-2.0 * h))) * D0
+                    + 0.5 * (alpha_t * (1 - torch.exp(-2.0 * h))) * D1
+                    + sigma_t * torch.sqrt(1.0 - torch.exp(-2 * h)) * noise
+                )
+            elif self.config.solver_type == "heun":
+                x_t = (
+                    (sigma_t / sigma_s0 * torch.exp(-h)) * sample
+                    + (alpha_t * (1 - torch.exp(-2.0 * h))) * D0
+                    + (alpha_t * ((1.0 - torch.exp(-2.0 * h)) / (-2.0 * h) + 1.0)) * D1
+                    + sigma_t * torch.sqrt(1.0 - torch.exp(-2 * h)) * noise
+                )
+
+        return x_t
+
+    def multistep_dpm_solver_third_order_update(
+        self,
+        model_output_list: List[torch.Tensor],
+        sample: torch.Tensor = None,
+    ) -> torch.Tensor:
+        """
+        One step for the third-order multistep DPMSolver.
+
+        Args:
+            model_output_list (`List[torch.Tensor]`):
+                The direct outputs from learned diffusion model at current and latter timesteps.
+            sample (`torch.Tensor`):
+                A current instance of a sample created by diffusion process.
+
+        Returns:
+            `torch.Tensor`:
+                The sample tensor at the previous timestep.
+        """
+        sigma_t, sigma_s0, sigma_s1, sigma_s2 = (
+            self.sigmas[self.step_index + 1],
+            self.sigmas[self.step_index],
+            self.sigmas[self.step_index - 1],
+            self.sigmas[self.step_index - 2],
+        )
+
+        alpha_t, sigma_t = self._sigma_to_alpha_sigma_t(sigma_t)
+        alpha_s0, sigma_s0 = self._sigma_to_alpha_sigma_t(sigma_s0)
+        alpha_s1, sigma_s1 = self._sigma_to_alpha_sigma_t(sigma_s1)
+        alpha_s2, sigma_s2 = self._sigma_to_alpha_sigma_t(sigma_s2)
+
+        lambda_t = torch.log(alpha_t) - torch.log(sigma_t)
+        lambda_s0 = torch.log(alpha_s0) - torch.log(sigma_s0)
+        lambda_s1 = torch.log(alpha_s1) - torch.log(sigma_s1)
+        lambda_s2 = torch.log(alpha_s2) - torch.log(sigma_s2)
+
+        m0, m1, m2 = model_output_list[-1], model_output_list[-2], model_output_list[-3]
+
+        h, h_0, h_1 = lambda_t - lambda_s0, lambda_s0 - lambda_s1, lambda_s1 - lambda_s2
+        r0, r1 = h_0 / h, h_1 / h
+        D0 = m0
+        D1_0, D1_1 = (1.0 / r0) * (m0 - m1), (1.0 / r1) * (m1 - m2)
+        D1 = D1_0 + (r0 / (r0 + r1)) * (D1_0 - D1_1)
+        D2 = (1.0 / (r0 + r1)) * (D1_0 - D1_1)
+        if self.config.algorithm_type == "dpmsolver++":
+            # See https://huggingface.co/papers/2206.00927 for detailed derivations
+            x_t = (
+                (sigma_t / sigma_s0) * sample
+                - (alpha_t * (torch.exp(-h) - 1.0)) * D0
+                + (alpha_t * ((torch.exp(-h) - 1.0) / h + 1.0)) * D1
+                - (alpha_t * ((torch.exp(-h) - 1.0 + h) / h**2 - 0.5)) * D2
+            )
+
+        return x_t
+
+    # Copied from diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler.index_for_timestep
+    def index_for_timestep(self, timestep, schedule_timesteps=None):
+        if schedule_timesteps is None:
+            schedule_timesteps = self.timesteps
+
+        index_candidates = (schedule_timesteps == timestep).nonzero()
+
+        if len(index_candidates) == 0:
+            step_index = len(self.timesteps) - 1
+        # 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)
+        elif len(index_candidates) > 1:
+            step_index = index_candidates[1].item()
+        else:
+            step_index = index_candidates[0].item()
+
+        return step_index
+
+    # Copied from diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler._init_step_index
+    def _init_step_index(self, timestep):
+        """
+        Initialize the step_index counter for the scheduler.
+        """
+
+        if self.begin_index is None:
+            if isinstance(timestep, torch.Tensor):
+                timestep = timestep.to(self.timesteps.device)
+            self._step_index = self.index_for_timestep(timestep)
+        else:
+            self._step_index = self._begin_index
+
+    def step(
+        self,
+        model_output: torch.Tensor,
+        timestep: Union[int, torch.Tensor],
+        sample: torch.Tensor,
+        generator=None,
+        return_dict: bool = True,
+    ) -> Union[SchedulerOutput, Tuple]:
+        """
+        Predict the sample from the previous timestep by reversing the SDE. This function propagates the sample with
+        the multistep DPMSolver.
+
+        Args:
+            model_output (`torch.Tensor`):
+                The direct output from learned diffusion model.
+            timestep (`int`):
+                The current discrete timestep in the diffusion chain.
+            sample (`torch.Tensor`):
+                A current instance of a sample created by the diffusion process.
+            generator (`torch.Generator`, *optional*):
+                A random number generator.
+            return_dict (`bool`):
+                Whether or not to return a [`~schedulers.scheduling_utils.SchedulerOutput`] or `tuple`.
+
+        Returns:
+            [`~schedulers.scheduling_utils.SchedulerOutput`] or `tuple`:
+                If return_dict is `True`, [`~schedulers.scheduling_utils.SchedulerOutput`] 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)
+
+        # Improve numerical stability for small number of steps
+        lower_order_final = (self.step_index == len(self.timesteps) - 1) and (
+            self.config.euler_at_final
+            or (self.config.lower_order_final and len(self.timesteps) < 15)
+            or self.config.final_sigmas_type == "zero"
+        )
+        lower_order_second = (
+            (self.step_index == len(self.timesteps) - 2) and self.config.lower_order_final and len(self.timesteps) < 15
+        )
+
+        model_output = self.convert_model_output(model_output, sample=sample)
+        for i in range(self.config.solver_order - 1):
+            self.model_outputs[i] = self.model_outputs[i + 1]
+        self.model_outputs[-1] = model_output
+
+        if self.config.algorithm_type == "sde-dpmsolver++":
+            noise = randn_tensor(
+                model_output.shape, generator=generator, device=model_output.device, dtype=model_output.dtype
+            )
+        else:
+            noise = None
+
+        if self.config.solver_order == 1 or self.lower_order_nums < 1 or lower_order_final:
+            prev_sample = self.dpm_solver_first_order_update(model_output, sample=sample, noise=noise)
+        elif self.config.solver_order == 2 or self.lower_order_nums < 2 or lower_order_second:
+            prev_sample = self.multistep_dpm_solver_second_order_update(self.model_outputs, sample=sample, noise=noise)
+        else:
+            prev_sample = self.multistep_dpm_solver_third_order_update(self.model_outputs, sample=sample)
+
+        if self.lower_order_nums < self.config.solver_order:
+            self.lower_order_nums += 1
+
+        # upon completion increase step index by one
+        self._step_index += 1
+
+        if not return_dict:
+            return (prev_sample,)
+
+        return SchedulerOutput(prev_sample=prev_sample)
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler.add_noise
+    def add_noise(
+        self,
+        original_samples: torch.Tensor,
+        noise: torch.Tensor,
+        timesteps: torch.Tensor,
+    ) -> torch.Tensor:
+        # Make sure sigmas and timesteps have the same device and dtype as original_samples
+        sigmas = self.sigmas.to(device=original_samples.device, dtype=original_samples.dtype)
+        if original_samples.device.type == "mps" and torch.is_floating_point(timesteps):
+            # mps does not support float64
+            schedule_timesteps = self.timesteps.to(original_samples.device, dtype=torch.float32)
+            timesteps = timesteps.to(original_samples.device, dtype=torch.float32)
+        else:
+            schedule_timesteps = self.timesteps.to(original_samples.device)
+            timesteps = timesteps.to(original_samples.device)
+
+        # self.begin_index is None when scheduler is used for training, or pipeline does not implement set_begin_index
+        if self.begin_index is None:
+            step_indices = [self.index_for_timestep(t, schedule_timesteps) for t in timesteps]
+        elif self.step_index is not None:
+            # add_noise is called after first denoising step (for inpainting)
+            step_indices = [self.step_index] * timesteps.shape[0]
+        else:
+            # add noise is called before first denoising step to create initial latent(img2img)
+            step_indices = [self.begin_index] * timesteps.shape[0]
+
+        sigma = sigmas[step_indices].flatten()
+        while len(sigma.shape) < len(original_samples.shape):
+            sigma = sigma.unsqueeze(-1)
+
+        noisy_samples = original_samples + noise * sigma
+        return noisy_samples
+
+    # Copied from diffusers.schedulers.scheduling_edm_euler.EDMEulerScheduler._get_conditioning_c_in
+    def _get_conditioning_c_in(self, sigma):
+        c_in = 1 / ((sigma**2 + self.config.sigma_data**2) ** 0.5)
+        return c_in
+
+    def __len__(self):
+        return self.config.num_train_timesteps

pythonProject/.venv/Lib/site-packages/diffusers/schedulers/scheduling_edm_euler.py

@@ -0,0 +1,448 @@
+# Copyright 2025 Katherine Crowson 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.
+
+import math
+from dataclasses import dataclass
+from typing import List, Optional, Tuple, Union
+
+import torch
+
+from ..configuration_utils import ConfigMixin, register_to_config
+from ..utils import BaseOutput, logging
+from ..utils.torch_utils import randn_tensor
+from .scheduling_utils import SchedulerMixin
+
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+@dataclass
+# Copied from diffusers.schedulers.scheduling_ddpm.DDPMSchedulerOutput with DDPM->EulerDiscrete
+class EDMEulerSchedulerOutput(BaseOutput):
+    """
+    Output class for the scheduler's `step` function output.
+
+    Args:
+        prev_sample (`torch.Tensor` 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.Tensor` 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.Tensor
+    pred_original_sample: Optional[torch.Tensor] = None
+
+
+class EDMEulerScheduler(SchedulerMixin, ConfigMixin):
+    """
+    Implements the Euler scheduler in EDM formulation as presented in Karras et al. 2022 [1].
+
+    [1] Karras, Tero, et al. "Elucidating the Design Space of Diffusion-Based Generative Models."
+    https://huggingface.co/papers/2206.00364
+
+    This model inherits from [`SchedulerMixin`] and [`ConfigMixin`]. Check the superclass documentation for the generic
+    methods the library implements for all schedulers such as loading and saving.
+
+    Args:
+        sigma_min (`float`, *optional*, defaults to 0.002):
+            Minimum noise magnitude in the sigma schedule. This was set to 0.002 in the EDM paper [1]; a reasonable
+            range is [0, 10].
+        sigma_max (`float`, *optional*, defaults to 80.0):
+            Maximum noise magnitude in the sigma schedule. This was set to 80.0 in the EDM paper [1]; a reasonable
+            range is [0.2, 80.0].
+        sigma_data (`float`, *optional*, defaults to 0.5):
+            The standard deviation of the data distribution. This is set to 0.5 in the EDM paper [1].
+        sigma_schedule (`str`, *optional*, defaults to `karras`):
+            Sigma schedule to compute the `sigmas`. By default, we the schedule introduced in the EDM paper
+            (https://huggingface.co/papers/2206.00364). Other acceptable value is "exponential". The exponential
+            schedule was incorporated in this model: https://huggingface.co/stabilityai/cosxl.
+        num_train_timesteps (`int`, defaults to 1000):
+            The number of diffusion steps to train the model.
+        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).
+        rho (`float`, *optional*, defaults to 7.0):
+            The rho parameter used for calculating the Karras sigma schedule, which is set to 7.0 in the EDM paper [1].
+        final_sigmas_type (`str`, defaults to `"zero"`):
+            The final `sigma` value for the noise schedule during the sampling process. If `"sigma_min"`, the final
+            sigma is the same as the last sigma in the training schedule. If `zero`, the final sigma is set to 0.
+    """
+
+    _compatibles = []
+    order = 1
+
+    @register_to_config
+    def __init__(
+        self,
+        sigma_min: float = 0.002,
+        sigma_max: float = 80.0,
+        sigma_data: float = 0.5,
+        sigma_schedule: str = "karras",
+        num_train_timesteps: int = 1000,
+        prediction_type: str = "epsilon",
+        rho: float = 7.0,
+        final_sigmas_type: str = "zero",  # can be "zero" or "sigma_min"
+    ):
+        if sigma_schedule not in ["karras", "exponential"]:
+            raise ValueError(f"Wrong value for provided for `{sigma_schedule=}`.`")
+
+        # setable values
+        self.num_inference_steps = None
+
+        sigmas_dtype = torch.float32 if torch.backends.mps.is_available() else torch.float64
+        sigmas = torch.arange(num_train_timesteps + 1, dtype=sigmas_dtype) / num_train_timesteps
+        if sigma_schedule == "karras":
+            sigmas = self._compute_karras_sigmas(sigmas)
+        elif sigma_schedule == "exponential":
+            sigmas = self._compute_exponential_sigmas(sigmas)
+        sigmas = sigmas.to(torch.float32)
+
+        self.timesteps = self.precondition_noise(sigmas)
+
+        if self.config.final_sigmas_type == "sigma_min":
+            sigma_last = sigmas[-1]
+        elif self.config.final_sigmas_type == "zero":
+            sigma_last = 0
+        else:
+            raise ValueError(
+                f"`final_sigmas_type` must be one of 'zero', or 'sigma_min', but got {self.config.final_sigmas_type}"
+            )
+
+        self.sigmas = torch.cat([sigmas, torch.full((1,), fill_value=sigma_last, device=sigmas.device)])
+
+        self.is_scale_input_called = False
+
+        self._step_index = None
+        self._begin_index = None
+        self.sigmas = self.sigmas.to("cpu")  # to avoid too much CPU/GPU communication
+
+    @property
+    def init_noise_sigma(self):
+        # standard deviation of the initial noise distribution
+        return (self.config.sigma_max**2 + 1) ** 0.5
+
+    @property
+    def step_index(self):
+        """
+        The index counter for current timestep. It will increase 1 after each scheduler step.
+        """
+        return self._step_index
+
+    @property
+    def begin_index(self):
+        """
+        The index for the first timestep. It should be set from pipeline with `set_begin_index` method.
+        """
+        return self._begin_index
+
+    # Copied from diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler.set_begin_index
+    def set_begin_index(self, begin_index: int = 0):
+        """
+        Sets the begin index for the scheduler. This function should be run from pipeline before the inference.
+
+        Args:
+            begin_index (`int`):
+                The begin index for the scheduler.
+        """
+        self._begin_index = begin_index
+
+    def precondition_inputs(self, sample, sigma):
+        c_in = self._get_conditioning_c_in(sigma)
+        scaled_sample = sample * c_in
+        return scaled_sample
+
+    def precondition_noise(self, sigma):
+        if not isinstance(sigma, torch.Tensor):
+            sigma = torch.tensor([sigma])
+
+        c_noise = 0.25 * torch.log(sigma)
+
+        return c_noise
+
+    def precondition_outputs(self, sample, model_output, sigma):
+        sigma_data = self.config.sigma_data
+        c_skip = sigma_data**2 / (sigma**2 + sigma_data**2)
+
+        if self.config.prediction_type == "epsilon":
+            c_out = sigma * sigma_data / (sigma**2 + sigma_data**2) ** 0.5
+        elif self.config.prediction_type == "v_prediction":
+            c_out = -sigma * sigma_data / (sigma**2 + sigma_data**2) ** 0.5
+        else:
+            raise ValueError(f"Prediction type {self.config.prediction_type} is not supported.")
+
+        denoised = c_skip * sample + c_out * model_output
+
+        return denoised
+
+    def scale_model_input(self, sample: torch.Tensor, timestep: Union[float, torch.Tensor]) -> torch.Tensor:
+        """
+        Ensures interchangeability with schedulers that need to scale the denoising model input depending on the
+        current timestep. Scales the denoising model input by `(sigma**2 + 1) ** 0.5` to match the Euler algorithm.
+
+        Args:
+            sample (`torch.Tensor`):
+                The input sample.
+            timestep (`int`, *optional*):
+                The current timestep in the diffusion chain.
+
+        Returns:
+            `torch.Tensor`:
+                A scaled input sample.
+        """
+        if self.step_index is None:
+            self._init_step_index(timestep)
+
+        sigma = self.sigmas[self.step_index]
+        sample = self.precondition_inputs(sample, sigma)
+
+        self.is_scale_input_called = True
+        return sample
+
+    def set_timesteps(
+        self,
+        num_inference_steps: int = None,
+        device: Union[str, torch.device] = None,
+        sigmas: Optional[Union[torch.Tensor, List[float]]] = 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.
+            sigmas (`Union[torch.Tensor, List[float]]`, *optional*):
+                Custom sigmas to use for the denoising process. If not defined, the default behavior when
+                `num_inference_steps` is passed will be used.
+        """
+        self.num_inference_steps = num_inference_steps
+
+        sigmas_dtype = torch.float32 if torch.backends.mps.is_available() else torch.float64
+        if sigmas is None:
+            sigmas = torch.linspace(0, 1, self.num_inference_steps, dtype=sigmas_dtype)
+        elif isinstance(sigmas, float):
+            sigmas = torch.tensor(sigmas, dtype=sigmas_dtype)
+        else:
+            sigmas = sigmas.to(sigmas_dtype)
+        if self.config.sigma_schedule == "karras":
+            sigmas = self._compute_karras_sigmas(sigmas)
+        elif self.config.sigma_schedule == "exponential":
+            sigmas = self._compute_exponential_sigmas(sigmas)
+        sigmas = sigmas.to(dtype=torch.float32, device=device)
+
+        self.timesteps = self.precondition_noise(sigmas)
+
+        if self.config.final_sigmas_type == "sigma_min":
+            sigma_last = sigmas[-1]
+        elif self.config.final_sigmas_type == "zero":
+            sigma_last = 0
+        else:
+            raise ValueError(
+                f"`final_sigmas_type` must be one of 'zero', or 'sigma_min', but got {self.config.final_sigmas_type}"
+            )
+
+        self.sigmas = torch.cat([sigmas, torch.full((1,), fill_value=sigma_last, device=sigmas.device)])
+        self._step_index = None
+        self._begin_index = None
+        self.sigmas = self.sigmas.to("cpu")  # to avoid too much CPU/GPU communication
+
+    # Taken from https://github.com/crowsonkb/k-diffusion/blob/686dbad0f39640ea25c8a8c6a6e56bb40eacefa2/k_diffusion/sampling.py#L17
+    def _compute_karras_sigmas(self, ramp, sigma_min=None, sigma_max=None) -> torch.Tensor:
+        """Constructs the noise schedule of Karras et al. (2022)."""
+        sigma_min = sigma_min or self.config.sigma_min
+        sigma_max = sigma_max or self.config.sigma_max
+
+        rho = self.config.rho
+        min_inv_rho = sigma_min ** (1 / rho)
+        max_inv_rho = sigma_max ** (1 / rho)
+        sigmas = (max_inv_rho + ramp * (min_inv_rho - max_inv_rho)) ** rho
+        return sigmas
+
+    def _compute_exponential_sigmas(self, ramp, sigma_min=None, sigma_max=None) -> torch.Tensor:
+        """Implementation closely follows k-diffusion.
+
+        https://github.com/crowsonkb/k-diffusion/blob/6ab5146d4a5ef63901326489f31f1d8e7dd36b48/k_diffusion/sampling.py#L26
+        """
+        sigma_min = sigma_min or self.config.sigma_min
+        sigma_max = sigma_max or self.config.sigma_max
+        sigmas = torch.linspace(math.log(sigma_min), math.log(sigma_max), len(ramp)).exp().flip(0)
+        return sigmas
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler.index_for_timestep
+    def index_for_timestep(self, timestep, schedule_timesteps=None):
+        if schedule_timesteps is None:
+            schedule_timesteps = self.timesteps
+
+        indices = (schedule_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)
+        pos = 1 if len(indices) > 1 else 0
+
+        return indices[pos].item()
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._init_step_index
+    def _init_step_index(self, timestep):
+        if self.begin_index is None:
+            if isinstance(timestep, torch.Tensor):
+                timestep = timestep.to(self.timesteps.device)
+            self._step_index = self.index_for_timestep(timestep)
+        else:
+            self._step_index = self._begin_index
+
+    def step(
+        self,
+        model_output: torch.Tensor,
+        timestep: Union[float, torch.Tensor],
+        sample: torch.Tensor,
+        s_churn: float = 0.0,
+        s_tmin: float = 0.0,
+        s_tmax: float = float("inf"),
+        s_noise: float = 1.0,
+        generator: Optional[torch.Generator] = None,
+        return_dict: bool = True,
+        pred_original_sample: Optional[torch.Tensor] = None,
+    ) -> Union[EDMEulerSchedulerOutput, 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.Tensor`):
+                The direct output from learned diffusion model.
+            timestep (`float`):
+                The current discrete timestep in the diffusion chain.
+            sample (`torch.Tensor`):
+                A current instance of a sample created by the diffusion process.
+            s_churn (`float`):
+            s_tmin  (`float`):
+            s_tmax  (`float`):
+            s_noise (`float`, defaults to 1.0):
+                Scaling factor for noise added to the sample.
+            generator (`torch.Generator`, *optional*):
+                A random number generator.
+            return_dict (`bool`):
+                Whether or not to return a [`~schedulers.scheduling_euler_discrete.EDMEulerSchedulerOutput`] or tuple.
+
+        Returns:
+            [`~schedulers.scheduling_euler_discrete.EDMEulerSchedulerOutput`] or `tuple`:
+                If return_dict is `True`, [`~schedulers.scheduling_euler_discrete.EDMEulerSchedulerOutput`] is
+                returned, otherwise a tuple is returned where the first element is the sample tensor.
+        """
+
+        if isinstance(timestep, (int, torch.IntTensor, torch.LongTensor)):
+            raise ValueError(
+                (
+                    "Passing integer indices (e.g. from `enumerate(timesteps)`) as timesteps to"
+                    " `EDMEulerScheduler.step()` is not supported. Make sure to pass"
+                    " one of the `scheduler.timesteps` as a timestep."
+                ),
+            )
+
+        if not self.is_scale_input_called:
+            logger.warning(
+                "The `scale_model_input` function should be called before `step` to ensure correct denoising. "
+                "See `StableDiffusionPipeline` for a usage example."
+            )
+
+        if self.step_index is None:
+            self._init_step_index(timestep)
+
+        # Upcast to avoid precision issues when computing prev_sample
+        sample = sample.to(torch.float32)
+
+        sigma = self.sigmas[self.step_index]
+
+        gamma = min(s_churn / (len(self.sigmas) - 1), 2**0.5 - 1) if s_tmin <= sigma <= s_tmax else 0.0
+
+        sigma_hat = sigma * (gamma + 1)
+
+        if gamma > 0:
+            noise = randn_tensor(
+                model_output.shape, dtype=model_output.dtype, device=model_output.device, generator=generator
+            )
+            eps = noise * s_noise
+            sample = sample + eps * (sigma_hat**2 - sigma**2) ** 0.5
+
+        # 1. compute predicted original sample (x_0) from sigma-scaled predicted noise
+        if pred_original_sample is None:
+            pred_original_sample = self.precondition_outputs(sample, model_output, sigma_hat)
+
+        # 2. Convert to an ODE derivative
+        derivative = (sample - pred_original_sample) / sigma_hat
+
+        dt = self.sigmas[self.step_index + 1] - sigma_hat
+
+        prev_sample = sample + derivative * dt
+
+        # Cast sample back to model compatible dtype
+        prev_sample = prev_sample.to(model_output.dtype)
+
+        # upon completion increase step index by one
+        self._step_index += 1
+
+        if not return_dict:
+            return (
+                prev_sample,
+                pred_original_sample,
+            )
+
+        return EDMEulerSchedulerOutput(prev_sample=prev_sample, pred_original_sample=pred_original_sample)
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler.add_noise
+    def add_noise(
+        self,
+        original_samples: torch.Tensor,
+        noise: torch.Tensor,
+        timesteps: torch.Tensor,
+    ) -> torch.Tensor:
+        # Make sure sigmas and timesteps have the same device and dtype as original_samples
+        sigmas = self.sigmas.to(device=original_samples.device, dtype=original_samples.dtype)
+        if original_samples.device.type == "mps" and torch.is_floating_point(timesteps):
+            # mps does not support float64
+            schedule_timesteps = self.timesteps.to(original_samples.device, dtype=torch.float32)
+            timesteps = timesteps.to(original_samples.device, dtype=torch.float32)
+        else:
+            schedule_timesteps = self.timesteps.to(original_samples.device)
+            timesteps = timesteps.to(original_samples.device)
+
+        # self.begin_index is None when scheduler is used for training, or pipeline does not implement set_begin_index
+        if self.begin_index is None:
+            step_indices = [self.index_for_timestep(t, schedule_timesteps) for t in timesteps]
+        elif self.step_index is not None:
+            # add_noise is called after first denoising step (for inpainting)
+            step_indices = [self.step_index] * timesteps.shape[0]
+        else:
+            # add noise is called before first denoising step to create initial latent(img2img)
+            step_indices = [self.begin_index] * timesteps.shape[0]
+
+        sigma = sigmas[step_indices].flatten()
+        while len(sigma.shape) < len(original_samples.shape):
+            sigma = sigma.unsqueeze(-1)
+
+        noisy_samples = original_samples + noise * sigma
+        return noisy_samples
+
+    def _get_conditioning_c_in(self, sigma):
+        c_in = 1 / ((sigma**2 + self.config.sigma_data**2) ** 0.5)
+        return c_in
+
+    def __len__(self):
+        return self.config.num_train_timesteps

pythonProject/.venv/Lib/site-packages/diffusers/schedulers/scheduling_euler_ancestral_discrete.py

@@ -0,0 +1,482 @@
+# Copyright 2025 Katherine Crowson 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.
+
+import math
+from dataclasses import dataclass
+from typing import List, Optional, Tuple, Union
+
+import numpy as np
+import torch
+
+from ..configuration_utils import ConfigMixin, register_to_config
+from ..utils import BaseOutput, logging
+from ..utils.torch_utils import randn_tensor
+from .scheduling_utils import KarrasDiffusionSchedulers, SchedulerMixin
+
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+@dataclass
+# Copied from diffusers.schedulers.scheduling_ddpm.DDPMSchedulerOutput with DDPM->EulerAncestralDiscrete
+class EulerAncestralDiscreteSchedulerOutput(BaseOutput):
+    """
+    Output class for the scheduler's `step` function output.
+
+    Args:
+        prev_sample (`torch.Tensor` 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.Tensor` 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.Tensor
+    pred_original_sample: Optional[torch.Tensor] = 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_transform_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):
+    """
+    Rescales betas to have zero terminal SNR Based on https://huggingface.co/papers/2305.08891 (Algorithm 1)
+
+
+    Args:
+        betas (`torch.Tensor`):
+            the betas that the scheduler is being initialized with.
+
+    Returns:
+        `torch.Tensor`: 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 EulerAncestralDiscreteScheduler(SchedulerMixin, ConfigMixin):
+    """
+    Ancestral sampling with Euler method steps.
+
+    This model inherits from [`SchedulerMixin`] and [`ConfigMixin`]. Check the superclass documentation for the generic
+    methods the library implements for all schedulers such as loading and saving.
+
+    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` or `scaled_linear`.
+        trained_betas (`np.ndarray`, *optional*):
+            Pass an array of betas directly to the constructor to bypass `beta_start` and `beta_end`.
+        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).
+        timestep_spacing (`str`, defaults to `"linspace"`):
+            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.
+        steps_offset (`int`, defaults to 0):
+            An offset added to the inference steps, as required by some model families.
+        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).
+    """
+
+    _compatibles = [e.name for e in KarrasDiffusionSchedulers]
+    order = 1
+
+    @register_to_config
+    def __init__(
+        self,
+        num_train_timesteps: int = 1000,
+        beta_start: float = 0.0001,
+        beta_end: float = 0.02,
+        beta_schedule: str = "linear",
+        trained_betas: Optional[Union[np.ndarray, List[float]]] = None,
+        prediction_type: str = "epsilon",
+        timestep_spacing: str = "linspace",
+        steps_offset: int = 0,
+        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} is not implemented for {self.__class__}")
+
+        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)
+
+        if rescale_betas_zero_snr:
+            # Close to 0 without being 0 so first sigma is not inf
+            # FP16 smallest positive subnormal works well here
+            self.alphas_cumprod[-1] = 2**-24
+
+        sigmas = np.array(((1 - self.alphas_cumprod) / self.alphas_cumprod) ** 0.5)
+        sigmas = np.concatenate([sigmas[::-1], [0.0]]).astype(np.float32)
+        self.sigmas = torch.from_numpy(sigmas)
+
+        # setable values
+        self.num_inference_steps = None
+        timesteps = np.linspace(0, num_train_timesteps - 1, num_train_timesteps, dtype=float)[::-1].copy()
+        self.timesteps = torch.from_numpy(timesteps)
+        self.is_scale_input_called = False
+
+        self._step_index = None
+        self._begin_index = None
+        self.sigmas = self.sigmas.to("cpu")  # to avoid too much CPU/GPU communication
+
+    @property
+    def init_noise_sigma(self):
+        # standard deviation of the initial noise distribution
+        if self.config.timestep_spacing in ["linspace", "trailing"]:
+            return self.sigmas.max()
+
+        return (self.sigmas.max() ** 2 + 1) ** 0.5
+
+    @property
+    def step_index(self):
+        """
+        The index counter for current timestep. It will increase 1 after each scheduler step.
+        """
+        return self._step_index
+
+    @property
+    def begin_index(self):
+        """
+        The index for the first timestep. It should be set from pipeline with `set_begin_index` method.
+        """
+        return self._begin_index
+
+    # Copied from diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler.set_begin_index
+    def set_begin_index(self, begin_index: int = 0):
+        """
+        Sets the begin index for the scheduler. This function should be run from pipeline before the inference.
+
+        Args:
+            begin_index (`int`):
+                The begin index for the scheduler.
+        """
+        self._begin_index = begin_index
+
+    def scale_model_input(self, sample: torch.Tensor, timestep: Union[float, torch.Tensor]) -> torch.Tensor:
+        """
+        Ensures interchangeability with schedulers that need to scale the denoising model input depending on the
+        current timestep. Scales the denoising model input by `(sigma**2 + 1) ** 0.5` to match the Euler algorithm.
+
+        Args:
+            sample (`torch.Tensor`):
+                The input sample.
+            timestep (`int`, *optional*):
+                The current timestep in the diffusion chain.
+
+        Returns:
+            `torch.Tensor`:
+                A scaled input sample.
+        """
+
+        if self.step_index is None:
+            self._init_step_index(timestep)
+
+        sigma = self.sigmas[self.step_index]
+        sample = sample / ((sigma**2 + 1) ** 0.5)
+        self.is_scale_input_called = True
+        return sample
+
+    def set_timesteps(self, num_inference_steps: int, device: Union[str, torch.device] = 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.
+        """
+        self.num_inference_steps = num_inference_steps
+
+        # "linspace", "leading", "trailing" corresponds to annotation of Table 2. of https://huggingface.co/papers/2305.08891
+        if self.config.timestep_spacing == "linspace":
+            timesteps = np.linspace(0, self.config.num_train_timesteps - 1, num_inference_steps, dtype=np.float32)[
+                ::-1
+            ].copy()
+        elif self.config.timestep_spacing == "leading":
+            step_ratio = self.config.num_train_timesteps // self.num_inference_steps
+            # creates integer timesteps by multiplying by ratio
+            # casting to int to avoid issues when num_inference_step is power of 3
+            timesteps = (np.arange(0, num_inference_steps) * step_ratio).round()[::-1].copy().astype(np.float32)
+            timesteps += self.config.steps_offset
+        elif self.config.timestep_spacing == "trailing":
+            step_ratio = self.config.num_train_timesteps / self.num_inference_steps
+            # creates integer timesteps by multiplying by ratio
+            # casting to int to avoid issues when num_inference_step is power of 3
+            timesteps = (np.arange(self.config.num_train_timesteps, 0, -step_ratio)).round().copy().astype(np.float32)
+            timesteps -= 1
+        else:
+            raise ValueError(
+                f"{self.config.timestep_spacing} is not supported. Please make sure to choose one of 'linspace', 'leading' or 'trailing'."
+            )
+
+        sigmas = np.array(((1 - self.alphas_cumprod) / self.alphas_cumprod) ** 0.5)
+        sigmas = np.interp(timesteps, np.arange(0, len(sigmas)), sigmas)
+        sigmas = np.concatenate([sigmas, [0.0]]).astype(np.float32)
+        self.sigmas = torch.from_numpy(sigmas).to(device=device)
+
+        self.timesteps = torch.from_numpy(timesteps).to(device=device)
+        self._step_index = None
+        self._begin_index = None
+        self.sigmas = self.sigmas.to("cpu")  # to avoid too much CPU/GPU communication
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler.index_for_timestep
+    def index_for_timestep(self, timestep, schedule_timesteps=None):
+        if schedule_timesteps is None:
+            schedule_timesteps = self.timesteps
+
+        indices = (schedule_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)
+        pos = 1 if len(indices) > 1 else 0
+
+        return indices[pos].item()
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._init_step_index
+    def _init_step_index(self, timestep):
+        if self.begin_index is None:
+            if isinstance(timestep, torch.Tensor):
+                timestep = timestep.to(self.timesteps.device)
+            self._step_index = self.index_for_timestep(timestep)
+        else:
+            self._step_index = self._begin_index
+
+    def step(
+        self,
+        model_output: torch.Tensor,
+        timestep: Union[float, torch.Tensor],
+        sample: torch.Tensor,
+        generator: Optional[torch.Generator] = None,
+        return_dict: bool = True,
+    ) -> Union[EulerAncestralDiscreteSchedulerOutput, 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.Tensor`):
+                The direct output from learned diffusion model.
+            timestep (`float`):
+                The current discrete timestep in the diffusion chain.
+            sample (`torch.Tensor`):
+                A current instance of a sample created by the diffusion process.
+            generator (`torch.Generator`, *optional*):
+                A random number generator.
+            return_dict (`bool`):
+                Whether or not to return a
+                [`~schedulers.scheduling_euler_ancestral_discrete.EulerAncestralDiscreteSchedulerOutput`] or tuple.
+
+        Returns:
+            [`~schedulers.scheduling_euler_ancestral_discrete.EulerAncestralDiscreteSchedulerOutput`] or `tuple`:
+                If return_dict is `True`,
+                [`~schedulers.scheduling_euler_ancestral_discrete.EulerAncestralDiscreteSchedulerOutput`] is returned,
+                otherwise a tuple is returned where the first element is the sample tensor.
+
+        """
+
+        if isinstance(timestep, (int, torch.IntTensor, torch.LongTensor)):
+            raise ValueError(
+                (
+                    "Passing integer indices (e.g. from `enumerate(timesteps)`) as timesteps to"
+                    " `EulerDiscreteScheduler.step()` is not supported. Make sure to pass"
+                    " one of the `scheduler.timesteps` as a timestep."
+                ),
+            )
+
+        if not self.is_scale_input_called:
+            logger.warning(
+                "The `scale_model_input` function should be called before `step` to ensure correct denoising. "
+                "See `StableDiffusionPipeline` for a usage example."
+            )
+
+        if self.step_index is None:
+            self._init_step_index(timestep)
+
+        sigma = self.sigmas[self.step_index]
+
+        # Upcast to avoid precision issues when computing prev_sample
+        sample = sample.to(torch.float32)
+
+        # 1. compute predicted original sample (x_0) from sigma-scaled predicted noise
+        if self.config.prediction_type == "epsilon":
+            pred_original_sample = sample - sigma * model_output
+        elif self.config.prediction_type == "v_prediction":
+            # * c_out + input * c_skip
+            pred_original_sample = model_output * (-sigma / (sigma**2 + 1) ** 0.5) + (sample / (sigma**2 + 1))
+        elif self.config.prediction_type == "sample":
+            raise NotImplementedError("prediction_type not implemented yet: sample")
+        else:
+            raise ValueError(
+                f"prediction_type given as {self.config.prediction_type} must be one of `epsilon`, or `v_prediction`"
+            )
+
+        sigma_from = self.sigmas[self.step_index]
+        sigma_to = self.sigmas[self.step_index + 1]
+        sigma_up = (sigma_to**2 * (sigma_from**2 - sigma_to**2) / sigma_from**2) ** 0.5
+        sigma_down = (sigma_to**2 - sigma_up**2) ** 0.5
+
+        # 2. Convert to an ODE derivative
+        derivative = (sample - pred_original_sample) / sigma
+
+        dt = sigma_down - sigma
+
+        prev_sample = sample + derivative * dt
+
+        device = model_output.device
+        noise = randn_tensor(model_output.shape, dtype=model_output.dtype, device=device, generator=generator)
+
+        prev_sample = prev_sample + noise * sigma_up
+
+        # Cast sample back to model compatible dtype
+        prev_sample = prev_sample.to(model_output.dtype)
+
+        # upon completion increase step index by one
+        self._step_index += 1
+
+        if not return_dict:
+            return (
+                prev_sample,
+                pred_original_sample,
+            )
+
+        return EulerAncestralDiscreteSchedulerOutput(
+            prev_sample=prev_sample, pred_original_sample=pred_original_sample
+        )
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler.add_noise
+    def add_noise(
+        self,
+        original_samples: torch.Tensor,
+        noise: torch.Tensor,
+        timesteps: torch.Tensor,
+    ) -> torch.Tensor:
+        # Make sure sigmas and timesteps have the same device and dtype as original_samples
+        sigmas = self.sigmas.to(device=original_samples.device, dtype=original_samples.dtype)
+        if original_samples.device.type == "mps" and torch.is_floating_point(timesteps):
+            # mps does not support float64
+            schedule_timesteps = self.timesteps.to(original_samples.device, dtype=torch.float32)
+            timesteps = timesteps.to(original_samples.device, dtype=torch.float32)
+        else:
+            schedule_timesteps = self.timesteps.to(original_samples.device)
+            timesteps = timesteps.to(original_samples.device)
+
+        # self.begin_index is None when scheduler is used for training, or pipeline does not implement set_begin_index
+        if self.begin_index is None:
+            step_indices = [self.index_for_timestep(t, schedule_timesteps) for t in timesteps]
+        elif self.step_index is not None:
+            # add_noise is called after first denoising step (for inpainting)
+            step_indices = [self.step_index] * timesteps.shape[0]
+        else:
+            # add noise is called before first denoising step to create initial latent(img2img)
+            step_indices = [self.begin_index] * timesteps.shape[0]
+
+        sigma = sigmas[step_indices].flatten()
+        while len(sigma.shape) < len(original_samples.shape):
+            sigma = sigma.unsqueeze(-1)
+
+        noisy_samples = original_samples + noise * sigma
+        return noisy_samples
+
+    def __len__(self):
+        return self.config.num_train_timesteps

pythonProject/.venv/Lib/site-packages/diffusers/schedulers/scheduling_euler_discrete.py

@@ -0,0 +1,757 @@
+# Copyright 2025 Katherine Crowson 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.
+
+import math
+from dataclasses import dataclass
+from typing import List, Optional, Tuple, Union
+
+import numpy as np
+import torch
+
+from ..configuration_utils import ConfigMixin, register_to_config
+from ..utils import BaseOutput, is_scipy_available, logging
+from ..utils.torch_utils import randn_tensor
+from .scheduling_utils import KarrasDiffusionSchedulers, SchedulerMixin
+
+
+if is_scipy_available():
+    import scipy.stats
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+@dataclass
+# Copied from diffusers.schedulers.scheduling_ddpm.DDPMSchedulerOutput with DDPM->EulerDiscrete
+class EulerDiscreteSchedulerOutput(BaseOutput):
+    """
+    Output class for the scheduler's `step` function output.
+
+    Args:
+        prev_sample (`torch.Tensor` 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.Tensor` 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.Tensor
+    pred_original_sample: Optional[torch.Tensor] = 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_transform_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):
+    """
+    Rescales betas to have zero terminal SNR Based on https://huggingface.co/papers/2305.08891 (Algorithm 1)
+
+
+    Args:
+        betas (`torch.Tensor`):
+            the betas that the scheduler is being initialized with.
+
+    Returns:
+        `torch.Tensor`: 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 EulerDiscreteScheduler(SchedulerMixin, ConfigMixin):
+    """
+    Euler scheduler.
+
+    This model inherits from [`SchedulerMixin`] and [`ConfigMixin`]. Check the superclass documentation for the generic
+    methods the library implements for all schedulers such as loading and saving.
+
+    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` or `scaled_linear`.
+        trained_betas (`np.ndarray`, *optional*):
+            Pass an array of betas directly to the constructor to bypass `beta_start` and `beta_end`.
+        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).
+        interpolation_type(`str`, defaults to `"linear"`, *optional*):
+            The interpolation type to compute intermediate sigmas for the scheduler denoising steps. Should be on of
+            `"linear"` or `"log_linear"`.
+        use_karras_sigmas (`bool`, *optional*, defaults to `False`):
+            Whether to use Karras sigmas for step sizes in the noise schedule during the sampling process. If `True`,
+            the sigmas are determined according to a sequence of noise levels {σi}.
+        use_exponential_sigmas (`bool`, *optional*, defaults to `False`):
+            Whether to use exponential sigmas for step sizes in the noise schedule during the sampling process.
+        use_beta_sigmas (`bool`, *optional*, defaults to `False`):
+            Whether to use beta sigmas for step sizes in the noise schedule during the sampling process. Refer to [Beta
+            Sampling is All You Need](https://huggingface.co/papers/2407.12173) for more information.
+        timestep_spacing (`str`, defaults to `"linspace"`):
+            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.
+        steps_offset (`int`, defaults to 0):
+            An offset added to the inference steps, as required by some model families.
+        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).
+        final_sigmas_type (`str`, defaults to `"zero"`):
+            The final `sigma` value for the noise schedule during the sampling process. If `"sigma_min"`, the final
+            sigma is the same as the last sigma in the training schedule. If `zero`, the final sigma is set to 0.
+    """
+
+    _compatibles = [e.name for e in KarrasDiffusionSchedulers]
+    order = 1
+
+    @register_to_config
+    def __init__(
+        self,
+        num_train_timesteps: int = 1000,
+        beta_start: float = 0.0001,
+        beta_end: float = 0.02,
+        beta_schedule: str = "linear",
+        trained_betas: Optional[Union[np.ndarray, List[float]]] = None,
+        prediction_type: str = "epsilon",
+        interpolation_type: str = "linear",
+        use_karras_sigmas: Optional[bool] = False,
+        use_exponential_sigmas: Optional[bool] = False,
+        use_beta_sigmas: Optional[bool] = False,
+        sigma_min: Optional[float] = None,
+        sigma_max: Optional[float] = None,
+        timestep_spacing: str = "linspace",
+        timestep_type: str = "discrete",  # can be "discrete" or "continuous"
+        steps_offset: int = 0,
+        rescale_betas_zero_snr: bool = False,
+        final_sigmas_type: str = "zero",  # can be "zero" or "sigma_min"
+    ):
+        if self.config.use_beta_sigmas and not is_scipy_available():
+            raise ImportError("Make sure to install scipy if you want to use beta sigmas.")
+        if sum([self.config.use_beta_sigmas, self.config.use_exponential_sigmas, self.config.use_karras_sigmas]) > 1:
+            raise ValueError(
+                "Only one of `config.use_beta_sigmas`, `config.use_exponential_sigmas`, `config.use_karras_sigmas` can be used."
+            )
+        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} is not implemented for {self.__class__}")
+
+        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)
+
+        if rescale_betas_zero_snr:
+            # Close to 0 without being 0 so first sigma is not inf
+            # FP16 smallest positive subnormal works well here
+            self.alphas_cumprod[-1] = 2**-24
+
+        sigmas = (((1 - self.alphas_cumprod) / self.alphas_cumprod) ** 0.5).flip(0)
+        timesteps = np.linspace(0, num_train_timesteps - 1, num_train_timesteps, dtype=float)[::-1].copy()
+        timesteps = torch.from_numpy(timesteps).to(dtype=torch.float32)
+
+        # setable values
+        self.num_inference_steps = None
+
+        # TODO: Support the full EDM scalings for all prediction types and timestep types
+        if timestep_type == "continuous" and prediction_type == "v_prediction":
+            self.timesteps = torch.Tensor([0.25 * sigma.log() for sigma in sigmas])
+        else:
+            self.timesteps = timesteps
+
+        self.sigmas = torch.cat([sigmas, torch.zeros(1, device=sigmas.device)])
+
+        self.is_scale_input_called = False
+        self.use_karras_sigmas = use_karras_sigmas
+        self.use_exponential_sigmas = use_exponential_sigmas
+        self.use_beta_sigmas = use_beta_sigmas
+
+        self._step_index = None
+        self._begin_index = None
+        self.sigmas = self.sigmas.to("cpu")  # to avoid too much CPU/GPU communication
+
+    @property
+    def init_noise_sigma(self):
+        # standard deviation of the initial noise distribution
+        max_sigma = max(self.sigmas) if isinstance(self.sigmas, list) else self.sigmas.max()
+        if self.config.timestep_spacing in ["linspace", "trailing"]:
+            return max_sigma
+
+        return (max_sigma**2 + 1) ** 0.5
+
+    @property
+    def step_index(self):
+        """
+        The index counter for current timestep. It will increase 1 after each scheduler step.
+        """
+        return self._step_index
+
+    @property
+    def begin_index(self):
+        """
+        The index for the first timestep. It should be set from pipeline with `set_begin_index` method.
+        """
+        return self._begin_index
+
+    # Copied from diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler.set_begin_index
+    def set_begin_index(self, begin_index: int = 0):
+        """
+        Sets the begin index for the scheduler. This function should be run from pipeline before the inference.
+
+        Args:
+            begin_index (`int`):
+                The begin index for the scheduler.
+        """
+        self._begin_index = begin_index
+
+    def scale_model_input(self, sample: torch.Tensor, timestep: Union[float, torch.Tensor]) -> torch.Tensor:
+        """
+        Ensures interchangeability with schedulers that need to scale the denoising model input depending on the
+        current timestep. Scales the denoising model input by `(sigma**2 + 1) ** 0.5` to match the Euler algorithm.
+
+        Args:
+            sample (`torch.Tensor`):
+                The input sample.
+            timestep (`int`, *optional*):
+                The current timestep in the diffusion chain.
+
+        Returns:
+            `torch.Tensor`:
+                A scaled input sample.
+        """
+        if self.step_index is None:
+            self._init_step_index(timestep)
+
+        sigma = self.sigmas[self.step_index]
+        sample = sample / ((sigma**2 + 1) ** 0.5)
+
+        self.is_scale_input_called = True
+        return sample
+
+    def set_timesteps(
+        self,
+        num_inference_steps: int = None,
+        device: Union[str, torch.device] = None,
+        timesteps: Optional[List[int]] = None,
+        sigmas: Optional[List[float]] = 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.
+            timesteps (`List[int]`, *optional*):
+                Custom timesteps used to support arbitrary timesteps schedule. If `None`, timesteps will be generated
+                based on the `timestep_spacing` attribute. If `timesteps` is passed, `num_inference_steps` and `sigmas`
+                must be `None`, and `timestep_spacing` attribute will be ignored.
+            sigmas (`List[float]`, *optional*):
+                Custom sigmas used to support arbitrary timesteps schedule schedule. If `None`, timesteps and sigmas
+                will be generated based on the relevant scheduler attributes. If `sigmas` is passed,
+                `num_inference_steps` and `timesteps` must be `None`, and the timesteps will be generated based on the
+                custom sigmas schedule.
+        """
+
+        if timesteps is not None and sigmas is not None:
+            raise ValueError("Only one of `timesteps` or `sigmas` should be set.")
+        if num_inference_steps is None and timesteps is None and sigmas is None:
+            raise ValueError("Must pass exactly one of `num_inference_steps` or `timesteps` or `sigmas.")
+        if num_inference_steps is not None and (timesteps is not None or sigmas is not None):
+            raise ValueError("Can only pass one of `num_inference_steps` or `timesteps` or `sigmas`.")
+        if timesteps is not None and self.config.use_karras_sigmas:
+            raise ValueError("Cannot set `timesteps` with `config.use_karras_sigmas = True`.")
+        if timesteps is not None and self.config.use_exponential_sigmas:
+            raise ValueError("Cannot set `timesteps` with `config.use_exponential_sigmas = True`.")
+        if timesteps is not None and self.config.use_beta_sigmas:
+            raise ValueError("Cannot set `timesteps` with `config.use_beta_sigmas = True`.")
+        if (
+            timesteps is not None
+            and self.config.timestep_type == "continuous"
+            and self.config.prediction_type == "v_prediction"
+        ):
+            raise ValueError(
+                "Cannot set `timesteps` with `config.timestep_type = 'continuous'` and `config.prediction_type = 'v_prediction'`."
+            )
+
+        if num_inference_steps is None:
+            num_inference_steps = len(timesteps) if timesteps is not None else len(sigmas) - 1
+        self.num_inference_steps = num_inference_steps
+
+        if sigmas is not None:
+            log_sigmas = np.log(np.array(((1 - self.alphas_cumprod) / self.alphas_cumprod) ** 0.5))
+            sigmas = np.array(sigmas).astype(np.float32)
+            timesteps = np.array([self._sigma_to_t(sigma, log_sigmas) for sigma in sigmas[:-1]])
+
+        else:
+            if timesteps is not None:
+                timesteps = np.array(timesteps).astype(np.float32)
+            else:
+                # "linspace", "leading", "trailing" corresponds to annotation of Table 2. of https://huggingface.co/papers/2305.08891
+                if self.config.timestep_spacing == "linspace":
+                    timesteps = np.linspace(
+                        0, self.config.num_train_timesteps - 1, num_inference_steps, dtype=np.float32
+                    )[::-1].copy()
+                elif self.config.timestep_spacing == "leading":
+                    step_ratio = self.config.num_train_timesteps // self.num_inference_steps
+                    # creates integer timesteps by multiplying by ratio
+                    # casting to int to avoid issues when num_inference_step is power of 3
+                    timesteps = (
+                        (np.arange(0, num_inference_steps) * step_ratio).round()[::-1].copy().astype(np.float32)
+                    )
+                    timesteps += self.config.steps_offset
+                elif self.config.timestep_spacing == "trailing":
+                    step_ratio = self.config.num_train_timesteps / self.num_inference_steps
+                    # creates integer timesteps by multiplying by ratio
+                    # casting to int to avoid issues when num_inference_step is power of 3
+                    timesteps = (
+                        (np.arange(self.config.num_train_timesteps, 0, -step_ratio)).round().copy().astype(np.float32)
+                    )
+                    timesteps -= 1
+                else:
+                    raise ValueError(
+                        f"{self.config.timestep_spacing} is not supported. Please make sure to choose one of 'linspace', 'leading' or 'trailing'."
+                    )
+
+            sigmas = np.array(((1 - self.alphas_cumprod) / self.alphas_cumprod) ** 0.5)
+            log_sigmas = np.log(sigmas)
+            if self.config.interpolation_type == "linear":
+                sigmas = np.interp(timesteps, np.arange(0, len(sigmas)), sigmas)
+            elif self.config.interpolation_type == "log_linear":
+                sigmas = torch.linspace(np.log(sigmas[-1]), np.log(sigmas[0]), num_inference_steps + 1).exp().numpy()
+            else:
+                raise ValueError(
+                    f"{self.config.interpolation_type} is not implemented. Please specify interpolation_type to either"
+                    " 'linear' or 'log_linear'"
+                )
+
+            if self.config.use_karras_sigmas:
+                sigmas = self._convert_to_karras(in_sigmas=sigmas, num_inference_steps=self.num_inference_steps)
+                timesteps = np.array([self._sigma_to_t(sigma, log_sigmas) for sigma in sigmas])
+
+            elif self.config.use_exponential_sigmas:
+                sigmas = self._convert_to_exponential(in_sigmas=sigmas, num_inference_steps=num_inference_steps)
+                timesteps = np.array([self._sigma_to_t(sigma, log_sigmas) for sigma in sigmas])
+
+            elif self.config.use_beta_sigmas:
+                sigmas = self._convert_to_beta(in_sigmas=sigmas, num_inference_steps=num_inference_steps)
+                timesteps = np.array([self._sigma_to_t(sigma, log_sigmas) for sigma in sigmas])
+
+            if self.config.final_sigmas_type == "sigma_min":
+                sigma_last = ((1 - self.alphas_cumprod[0]) / self.alphas_cumprod[0]) ** 0.5
+            elif self.config.final_sigmas_type == "zero":
+                sigma_last = 0
+            else:
+                raise ValueError(
+                    f"`final_sigmas_type` must be one of 'zero', or 'sigma_min', but got {self.config.final_sigmas_type}"
+                )
+
+            sigmas = np.concatenate([sigmas, [sigma_last]]).astype(np.float32)
+
+        sigmas = torch.from_numpy(sigmas).to(dtype=torch.float32, device=device)
+
+        # TODO: Support the full EDM scalings for all prediction types and timestep types
+        if self.config.timestep_type == "continuous" and self.config.prediction_type == "v_prediction":
+            self.timesteps = torch.Tensor([0.25 * sigma.log() for sigma in sigmas[:-1]]).to(device=device)
+        else:
+            self.timesteps = torch.from_numpy(timesteps.astype(np.float32)).to(device=device)
+
+        self._step_index = None
+        self._begin_index = None
+        self.sigmas = sigmas.to("cpu")  # to avoid too much CPU/GPU communication
+
+    def _sigma_to_t(self, sigma, log_sigmas):
+        # get log sigma
+        log_sigma = np.log(np.maximum(sigma, 1e-10))
+
+        # get distribution
+        dists = log_sigma - log_sigmas[:, np.newaxis]
+
+        # get sigmas range
+        low_idx = np.cumsum((dists >= 0), axis=0).argmax(axis=0).clip(max=log_sigmas.shape[0] - 2)
+        high_idx = low_idx + 1
+
+        low = log_sigmas[low_idx]
+        high = log_sigmas[high_idx]
+
+        # interpolate sigmas
+        w = (low - log_sigma) / (low - high)
+        w = np.clip(w, 0, 1)
+
+        # transform interpolation to time range
+        t = (1 - w) * low_idx + w * high_idx
+        t = t.reshape(sigma.shape)
+        return t
+
+    # Copied from https://github.com/crowsonkb/k-diffusion/blob/686dbad0f39640ea25c8a8c6a6e56bb40eacefa2/k_diffusion/sampling.py#L17
+    def _convert_to_karras(self, in_sigmas: torch.Tensor, num_inference_steps) -> torch.Tensor:
+        """Constructs the noise schedule of Karras et al. (2022)."""
+
+        # Hack to make sure that other schedulers which copy this function don't break
+        # TODO: Add this logic to the other schedulers
+        if hasattr(self.config, "sigma_min"):
+            sigma_min = self.config.sigma_min
+        else:
+            sigma_min = None
+
+        if hasattr(self.config, "sigma_max"):
+            sigma_max = self.config.sigma_max
+        else:
+            sigma_max = None
+
+        sigma_min = sigma_min if sigma_min is not None else in_sigmas[-1].item()
+        sigma_max = sigma_max if sigma_max is not None else in_sigmas[0].item()
+
+        rho = 7.0  # 7.0 is the value used in the paper
+        ramp = np.linspace(0, 1, num_inference_steps)
+        min_inv_rho = sigma_min ** (1 / rho)
+        max_inv_rho = sigma_max ** (1 / rho)
+        sigmas = (max_inv_rho + ramp * (min_inv_rho - max_inv_rho)) ** rho
+        return sigmas
+
+    # Copied from https://github.com/crowsonkb/k-diffusion/blob/686dbad0f39640ea25c8a8c6a6e56bb40eacefa2/k_diffusion/sampling.py#L26
+    def _convert_to_exponential(self, in_sigmas: torch.Tensor, num_inference_steps: int) -> torch.Tensor:
+        """Constructs an exponential noise schedule."""
+
+        # Hack to make sure that other schedulers which copy this function don't break
+        # TODO: Add this logic to the other schedulers
+        if hasattr(self.config, "sigma_min"):
+            sigma_min = self.config.sigma_min
+        else:
+            sigma_min = None
+
+        if hasattr(self.config, "sigma_max"):
+            sigma_max = self.config.sigma_max
+        else:
+            sigma_max = None
+
+        sigma_min = sigma_min if sigma_min is not None else in_sigmas[-1].item()
+        sigma_max = sigma_max if sigma_max is not None else in_sigmas[0].item()
+
+        sigmas = np.exp(np.linspace(math.log(sigma_max), math.log(sigma_min), num_inference_steps))
+        return sigmas
+
+    def _convert_to_beta(
+        self, in_sigmas: torch.Tensor, num_inference_steps: int, alpha: float = 0.6, beta: float = 0.6
+    ) -> torch.Tensor:
+        """From "Beta Sampling is All You Need" [arXiv:2407.12173] (Lee et. al, 2024)"""
+
+        # Hack to make sure that other schedulers which copy this function don't break
+        # TODO: Add this logic to the other schedulers
+        if hasattr(self.config, "sigma_min"):
+            sigma_min = self.config.sigma_min
+        else:
+            sigma_min = None
+
+        if hasattr(self.config, "sigma_max"):
+            sigma_max = self.config.sigma_max
+        else:
+            sigma_max = None
+
+        sigma_min = sigma_min if sigma_min is not None else in_sigmas[-1].item()
+        sigma_max = sigma_max if sigma_max is not None else in_sigmas[0].item()
+
+        sigmas = np.array(
+            [
+                sigma_min + (ppf * (sigma_max - sigma_min))
+                for ppf in [
+                    scipy.stats.beta.ppf(timestep, alpha, beta)
+                    for timestep in 1 - np.linspace(0, 1, num_inference_steps)
+                ]
+            ]
+        )
+        return sigmas
+
+    def index_for_timestep(self, timestep, schedule_timesteps=None):
+        if schedule_timesteps is None:
+            schedule_timesteps = self.timesteps
+
+        indices = (schedule_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)
+        pos = 1 if len(indices) > 1 else 0
+
+        return indices[pos].item()
+
+    def _init_step_index(self, timestep):
+        if self.begin_index is None:
+            if isinstance(timestep, torch.Tensor):
+                timestep = timestep.to(self.timesteps.device)
+            self._step_index = self.index_for_timestep(timestep)
+        else:
+            self._step_index = self._begin_index
+
+    def step(
+        self,
+        model_output: torch.Tensor,
+        timestep: Union[float, torch.Tensor],
+        sample: torch.Tensor,
+        s_churn: float = 0.0,
+        s_tmin: float = 0.0,
+        s_tmax: float = float("inf"),
+        s_noise: float = 1.0,
+        generator: Optional[torch.Generator] = None,
+        return_dict: bool = True,
+    ) -> Union[EulerDiscreteSchedulerOutput, 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.Tensor`):
+                The direct output from learned diffusion model.
+            timestep (`float`):
+                The current discrete timestep in the diffusion chain.
+            sample (`torch.Tensor`):
+                A current instance of a sample created by the diffusion process.
+            s_churn (`float`):
+            s_tmin  (`float`):
+            s_tmax  (`float`):
+            s_noise (`float`, defaults to 1.0):
+                Scaling factor for noise added to the sample.
+            generator (`torch.Generator`, *optional*):
+                A random number generator.
+            return_dict (`bool`):
+                Whether or not to return a [`~schedulers.scheduling_euler_discrete.EulerDiscreteSchedulerOutput`] or
+                tuple.
+
+        Returns:
+            [`~schedulers.scheduling_euler_discrete.EulerDiscreteSchedulerOutput`] or `tuple`:
+                If return_dict is `True`, [`~schedulers.scheduling_euler_discrete.EulerDiscreteSchedulerOutput`] is
+                returned, otherwise a tuple is returned where the first element is the sample tensor.
+        """
+
+        if isinstance(timestep, (int, torch.IntTensor, torch.LongTensor)):
+            raise ValueError(
+                (
+                    "Passing integer indices (e.g. from `enumerate(timesteps)`) as timesteps to"
+                    " `EulerDiscreteScheduler.step()` is not supported. Make sure to pass"
+                    " one of the `scheduler.timesteps` as a timestep."
+                ),
+            )
+
+        if not self.is_scale_input_called:
+            logger.warning(
+                "The `scale_model_input` function should be called before `step` to ensure correct denoising. "
+                "See `StableDiffusionPipeline` for a usage example."
+            )
+
+        if self.step_index is None:
+            self._init_step_index(timestep)
+
+        # Upcast to avoid precision issues when computing prev_sample
+        sample = sample.to(torch.float32)
+
+        sigma = self.sigmas[self.step_index]
+
+        gamma = min(s_churn / (len(self.sigmas) - 1), 2**0.5 - 1) if s_tmin <= sigma <= s_tmax else 0.0
+
+        sigma_hat = sigma * (gamma + 1)
+
+        if gamma > 0:
+            noise = randn_tensor(
+                model_output.shape, dtype=model_output.dtype, device=model_output.device, generator=generator
+            )
+            eps = noise * s_noise
+            sample = sample + eps * (sigma_hat**2 - sigma**2) ** 0.5
+
+        # 1. compute predicted original sample (x_0) from sigma-scaled predicted noise
+        # NOTE: "original_sample" should not be an expected prediction_type but is left in for
+        # backwards compatibility
+        if self.config.prediction_type == "original_sample" or self.config.prediction_type == "sample":
+            pred_original_sample = model_output
+        elif self.config.prediction_type == "epsilon":
+            pred_original_sample = sample - sigma_hat * model_output
+        elif self.config.prediction_type == "v_prediction":
+            # denoised = model_output * c_out + input * c_skip
+            pred_original_sample = model_output * (-sigma / (sigma**2 + 1) ** 0.5) + (sample / (sigma**2 + 1))
+        else:
+            raise ValueError(
+                f"prediction_type given as {self.config.prediction_type} must be one of `epsilon`, or `v_prediction`"
+            )
+
+        # 2. Convert to an ODE derivative
+        derivative = (sample - pred_original_sample) / sigma_hat
+
+        dt = self.sigmas[self.step_index + 1] - sigma_hat
+
+        prev_sample = sample + derivative * dt
+
+        # Cast sample back to model compatible dtype
+        prev_sample = prev_sample.to(model_output.dtype)
+
+        # upon completion increase step index by one
+        self._step_index += 1
+
+        if not return_dict:
+            return (
+                prev_sample,
+                pred_original_sample,
+            )
+
+        return EulerDiscreteSchedulerOutput(prev_sample=prev_sample, pred_original_sample=pred_original_sample)
+
+    def add_noise(
+        self,
+        original_samples: torch.Tensor,
+        noise: torch.Tensor,
+        timesteps: torch.Tensor,
+    ) -> torch.Tensor:
+        # Make sure sigmas and timesteps have the same device and dtype as original_samples
+        sigmas = self.sigmas.to(device=original_samples.device, dtype=original_samples.dtype)
+        if original_samples.device.type == "mps" and torch.is_floating_point(timesteps):
+            # mps does not support float64
+            schedule_timesteps = self.timesteps.to(original_samples.device, dtype=torch.float32)
+            timesteps = timesteps.to(original_samples.device, dtype=torch.float32)
+        else:
+            schedule_timesteps = self.timesteps.to(original_samples.device)
+            timesteps = timesteps.to(original_samples.device)
+
+        # self.begin_index is None when scheduler is used for training, or pipeline does not implement set_begin_index
+        if self.begin_index is None:
+            step_indices = [self.index_for_timestep(t, schedule_timesteps) for t in timesteps]
+        elif self.step_index is not None:
+            # add_noise is called after first denoising step (for inpainting)
+            step_indices = [self.step_index] * timesteps.shape[0]
+        else:
+            # add noise is called before first denoising step to create initial latent(img2img)
+            step_indices = [self.begin_index] * timesteps.shape[0]
+
+        sigma = sigmas[step_indices].flatten()
+        while len(sigma.shape) < len(original_samples.shape):
+            sigma = sigma.unsqueeze(-1)
+
+        noisy_samples = original_samples + noise * sigma
+        return noisy_samples
+
+    def get_velocity(self, sample: torch.Tensor, noise: torch.Tensor, timesteps: torch.Tensor) -> torch.Tensor:
+        if (
+            isinstance(timesteps, int)
+            or isinstance(timesteps, torch.IntTensor)
+            or isinstance(timesteps, torch.LongTensor)
+        ):
+            raise ValueError(
+                (
+                    "Passing integer indices (e.g. from `enumerate(timesteps)`) as timesteps to"
+                    " `EulerDiscreteScheduler.get_velocity()` is not supported. Make sure to pass"
+                    " one of the `scheduler.timesteps` as a timestep."
+                ),
+            )
+
+        if sample.device.type == "mps" and torch.is_floating_point(timesteps):
+            # mps does not support float64
+            schedule_timesteps = self.timesteps.to(sample.device, dtype=torch.float32)
+            timesteps = timesteps.to(sample.device, dtype=torch.float32)
+        else:
+            schedule_timesteps = self.timesteps.to(sample.device)
+            timesteps = timesteps.to(sample.device)
+
+        step_indices = [self.index_for_timestep(t, schedule_timesteps) for t in timesteps]
+        alphas_cumprod = self.alphas_cumprod.to(sample)
+        sqrt_alpha_prod = alphas_cumprod[step_indices] ** 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[step_indices]) ** 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

pythonProject/.venv/Lib/site-packages/diffusers/schedulers/scheduling_euler_discrete_flax.py

@@ -0,0 +1,265 @@
+# Copyright 2025 Katherine Crowson 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.
+
+from dataclasses import dataclass
+from typing import Optional, Tuple, Union
+
+import flax
+import jax.numpy as jnp
+
+from ..configuration_utils import ConfigMixin, register_to_config
+from .scheduling_utils_flax import (
+    CommonSchedulerState,
+    FlaxKarrasDiffusionSchedulers,
+    FlaxSchedulerMixin,
+    FlaxSchedulerOutput,
+    broadcast_to_shape_from_left,
+)
+
+
+@flax.struct.dataclass
+class EulerDiscreteSchedulerState:
+    common: CommonSchedulerState
+
+    # setable values
+    init_noise_sigma: jnp.ndarray
+    timesteps: jnp.ndarray
+    sigmas: jnp.ndarray
+    num_inference_steps: Optional[int] = None
+
+    @classmethod
+    def create(
+        cls, common: CommonSchedulerState, init_noise_sigma: jnp.ndarray, timesteps: jnp.ndarray, sigmas: jnp.ndarray
+    ):
+        return cls(common=common, init_noise_sigma=init_noise_sigma, timesteps=timesteps, sigmas=sigmas)
+
+
+@dataclass
+class FlaxEulerDiscreteSchedulerOutput(FlaxSchedulerOutput):
+    state: EulerDiscreteSchedulerState
+
+
+class FlaxEulerDiscreteScheduler(FlaxSchedulerMixin, ConfigMixin):
+    """
+    Euler scheduler (Algorithm 2) from Karras et al. (2022) https://huggingface.co/papers/2206.00364. . Based on the
+    original k-diffusion implementation by Katherine Crowson:
+    https://github.com/crowsonkb/k-diffusion/blob/481677d114f6ea445aa009cf5bd7a9cdee909e47/k_diffusion/sampling.py#L51
+
+
+    [`~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`): number of diffusion steps used to train the model.
+        beta_start (`float`): the starting `beta` value of inference.
+        beta_end (`float`): the final `beta` value.
+        beta_schedule (`str`):
+            the beta schedule, a mapping from a beta range to a sequence of betas for stepping the model. Choose from
+            `linear` or `scaled_linear`.
+        trained_betas (`jnp.ndarray`, optional):
+            option to pass an array of betas directly to the constructor to bypass `beta_start`, `beta_end` etc.
+        prediction_type (`str`, default `epsilon`, optional):
+            prediction type of the scheduler function, one of `epsilon` (predicting the noise of the diffusion
+            process), `sample` (directly predicting the noisy sample`) or `v_prediction` (see section 2.4
+            https://imagen.research.google/video/paper.pdf)
+        dtype (`jnp.dtype`, *optional*, defaults to `jnp.float32`):
+            the `dtype` used for params and computation.
+    """
+
+    _compatibles = [e.name for e in FlaxKarrasDiffusionSchedulers]
+
+    dtype: jnp.dtype
+
+    @property
+    def has_state(self):
+        return True
+
+    @register_to_config
+    def __init__(
+        self,
+        num_train_timesteps: int = 1000,
+        beta_start: float = 0.0001,
+        beta_end: float = 0.02,
+        beta_schedule: str = "linear",
+        trained_betas: Optional[jnp.ndarray] = None,
+        prediction_type: str = "epsilon",
+        timestep_spacing: str = "linspace",
+        dtype: jnp.dtype = jnp.float32,
+    ):
+        self.dtype = dtype
+
+    def create_state(self, common: Optional[CommonSchedulerState] = None) -> EulerDiscreteSchedulerState:
+        if common is None:
+            common = CommonSchedulerState.create(self)
+
+        timesteps = jnp.arange(0, self.config.num_train_timesteps).round()[::-1]
+        sigmas = ((1 - common.alphas_cumprod) / common.alphas_cumprod) ** 0.5
+        sigmas = jnp.interp(timesteps, jnp.arange(0, len(sigmas)), sigmas)
+        sigmas = jnp.concatenate([sigmas, jnp.array([0.0], dtype=self.dtype)])
+
+        # standard deviation of the initial noise distribution
+        if self.config.timestep_spacing in ["linspace", "trailing"]:
+            init_noise_sigma = sigmas.max()
+        else:
+            init_noise_sigma = (sigmas.max() ** 2 + 1) ** 0.5
+
+        return EulerDiscreteSchedulerState.create(
+            common=common,
+            init_noise_sigma=init_noise_sigma,
+            timesteps=timesteps,
+            sigmas=sigmas,
+        )
+
+    def scale_model_input(self, state: EulerDiscreteSchedulerState, sample: jnp.ndarray, timestep: int) -> jnp.ndarray:
+        """
+        Scales the denoising model input by `(sigma**2 + 1) ** 0.5` to match the Euler algorithm.
+
+        Args:
+            state (`EulerDiscreteSchedulerState`):
+                the `FlaxEulerDiscreteScheduler` state data class instance.
+            sample (`jnp.ndarray`):
+                current instance of sample being created by diffusion process.
+            timestep (`int`):
+                current discrete timestep in the diffusion chain.
+
+        Returns:
+            `jnp.ndarray`: scaled input sample
+        """
+        (step_index,) = jnp.where(state.timesteps == timestep, size=1)
+        step_index = step_index[0]
+
+        sigma = state.sigmas[step_index]
+        sample = sample / ((sigma**2 + 1) ** 0.5)
+        return sample
+
+    def set_timesteps(
+        self, state: EulerDiscreteSchedulerState, num_inference_steps: int, shape: Tuple = ()
+    ) -> EulerDiscreteSchedulerState:
+        """
+        Sets the timesteps used for the diffusion chain. Supporting function to be run before inference.
+
+        Args:
+            state (`EulerDiscreteSchedulerState`):
+                the `FlaxEulerDiscreteScheduler` state data class instance.
+            num_inference_steps (`int`):
+                the number of diffusion steps used when generating samples with a pre-trained model.
+        """
+
+        if self.config.timestep_spacing == "linspace":
+            timesteps = jnp.linspace(self.config.num_train_timesteps - 1, 0, num_inference_steps, dtype=self.dtype)
+        elif self.config.timestep_spacing == "leading":
+            step_ratio = self.config.num_train_timesteps // num_inference_steps
+            timesteps = (jnp.arange(0, num_inference_steps) * step_ratio).round()[::-1].copy().astype(float)
+            timesteps += 1
+        else:
+            raise ValueError(
+                f"timestep_spacing must be one of ['linspace', 'leading'], got {self.config.timestep_spacing}"
+            )
+
+        sigmas = ((1 - state.common.alphas_cumprod) / state.common.alphas_cumprod) ** 0.5
+        sigmas = jnp.interp(timesteps, jnp.arange(0, len(sigmas)), sigmas)
+        sigmas = jnp.concatenate([sigmas, jnp.array([0.0], dtype=self.dtype)])
+
+        # standard deviation of the initial noise distribution
+        if self.config.timestep_spacing in ["linspace", "trailing"]:
+            init_noise_sigma = sigmas.max()
+        else:
+            init_noise_sigma = (sigmas.max() ** 2 + 1) ** 0.5
+
+        return state.replace(
+            timesteps=timesteps,
+            sigmas=sigmas,
+            num_inference_steps=num_inference_steps,
+            init_noise_sigma=init_noise_sigma,
+        )
+
+    def step(
+        self,
+        state: EulerDiscreteSchedulerState,
+        model_output: jnp.ndarray,
+        timestep: int,
+        sample: jnp.ndarray,
+        return_dict: bool = True,
+    ) -> Union[FlaxEulerDiscreteSchedulerOutput, Tuple]:
+        """
+        Predict the sample at the previous timestep by reversing the SDE. Core function to propagate the diffusion
+        process from the learned model outputs (most often the predicted noise).
+
+        Args:
+            state (`EulerDiscreteSchedulerState`):
+                the `FlaxEulerDiscreteScheduler` state data class instance.
+            model_output (`jnp.ndarray`): direct output from learned diffusion model.
+            timestep (`int`): current discrete timestep in the diffusion chain.
+            sample (`jnp.ndarray`):
+                current instance of sample being created by diffusion process.
+            order: coefficient for multi-step inference.
+            return_dict (`bool`): option for returning tuple rather than FlaxEulerDiscreteScheduler class
+
+        Returns:
+            [`FlaxEulerDiscreteScheduler`] or `tuple`: [`FlaxEulerDiscreteScheduler`] if `return_dict` is True,
+            otherwise a `tuple`. When returning a tuple, the first element is the sample tensor.
+
+        """
+        if state.num_inference_steps is None:
+            raise ValueError(
+                "Number of inference steps is 'None', you need to run 'set_timesteps' after creating the scheduler"
+            )
+
+        (step_index,) = jnp.where(state.timesteps == timestep, size=1)
+        step_index = step_index[0]
+
+        sigma = state.sigmas[step_index]
+
+        # 1. compute predicted original sample (x_0) from sigma-scaled predicted noise
+        if self.config.prediction_type == "epsilon":
+            pred_original_sample = sample - sigma * model_output
+        elif self.config.prediction_type == "v_prediction":
+            # * c_out + input * c_skip
+            pred_original_sample = model_output * (-sigma / (sigma**2 + 1) ** 0.5) + (sample / (sigma**2 + 1))
+        else:
+            raise ValueError(
+                f"prediction_type given as {self.config.prediction_type} must be one of `epsilon`, or `v_prediction`"
+            )
+
+        # 2. Convert to an ODE derivative
+        derivative = (sample - pred_original_sample) / sigma
+
+        # dt = sigma_down - sigma
+        dt = state.sigmas[step_index + 1] - sigma
+
+        prev_sample = sample + derivative * dt
+
+        if not return_dict:
+            return (prev_sample, state)
+
+        return FlaxEulerDiscreteSchedulerOutput(prev_sample=prev_sample, state=state)
+
+    def add_noise(
+        self,
+        state: EulerDiscreteSchedulerState,
+        original_samples: jnp.ndarray,
+        noise: jnp.ndarray,
+        timesteps: jnp.ndarray,
+    ) -> jnp.ndarray:
+        sigma = state.sigmas[timesteps].flatten()
+        sigma = broadcast_to_shape_from_left(sigma, noise.shape)
+
+        noisy_samples = original_samples + noise * sigma
+
+        return noisy_samples
+
+    def __len__(self):
+        return self.config.num_train_timesteps

pythonProject/.venv/Lib/site-packages/diffusers/schedulers/scheduling_flow_match_euler_discrete.py

@@ -0,0 +1,561 @@
+# Copyright 2025 Stability AI, Katherine Crowson 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.
+
+import math
+from dataclasses import dataclass
+from typing import List, Optional, Tuple, Union
+
+import numpy as np
+import torch
+
+from ..configuration_utils import ConfigMixin, register_to_config
+from ..utils import BaseOutput, is_scipy_available, logging
+from .scheduling_utils import SchedulerMixin
+
+
+if is_scipy_available():
+    import scipy.stats
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+@dataclass
+class FlowMatchEulerDiscreteSchedulerOutput(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.
+    """
+
+    prev_sample: torch.FloatTensor
+
+
+class FlowMatchEulerDiscreteScheduler(SchedulerMixin, ConfigMixin):
+    """
+    Euler scheduler.
+
+    This model inherits from [`SchedulerMixin`] and [`ConfigMixin`]. Check the superclass documentation for the generic
+    methods the library implements for all schedulers such as loading and saving.
+
+    Args:
+        num_train_timesteps (`int`, defaults to 1000):
+            The number of diffusion steps to train the model.
+        shift (`float`, defaults to 1.0):
+            The shift value for the timestep schedule.
+        use_dynamic_shifting (`bool`, defaults to False):
+            Whether to apply timestep shifting on-the-fly based on the image resolution.
+        base_shift (`float`, defaults to 0.5):
+            Value to stabilize image generation. Increasing `base_shift` reduces variation and image is more consistent
+            with desired output.
+        max_shift (`float`, defaults to 1.15):
+            Value change allowed to latent vectors. Increasing `max_shift` encourages more variation and image may be
+            more exaggerated or stylized.
+        base_image_seq_len (`int`, defaults to 256):
+            The base image sequence length.
+        max_image_seq_len (`int`, defaults to 4096):
+            The maximum image sequence length.
+        invert_sigmas (`bool`, defaults to False):
+            Whether to invert the sigmas.
+        shift_terminal (`float`, defaults to None):
+            The end value of the shifted timestep schedule.
+        use_karras_sigmas (`bool`, defaults to False):
+            Whether to use Karras sigmas for step sizes in the noise schedule during sampling.
+        use_exponential_sigmas (`bool`, defaults to False):
+            Whether to use exponential sigmas for step sizes in the noise schedule during sampling.
+        use_beta_sigmas (`bool`, defaults to False):
+            Whether to use beta sigmas for step sizes in the noise schedule during sampling.
+        time_shift_type (`str`, defaults to "exponential"):
+            The type of dynamic resolution-dependent timestep shifting to apply. Either "exponential" or "linear".
+        stochastic_sampling (`bool`, defaults to False):
+            Whether to use stochastic sampling.
+    """
+
+    _compatibles = []
+    order = 1
+
+    @register_to_config
+    def __init__(
+        self,
+        num_train_timesteps: int = 1000,
+        shift: float = 1.0,
+        use_dynamic_shifting: bool = False,
+        base_shift: Optional[float] = 0.5,
+        max_shift: Optional[float] = 1.15,
+        base_image_seq_len: Optional[int] = 256,
+        max_image_seq_len: Optional[int] = 4096,
+        invert_sigmas: bool = False,
+        shift_terminal: Optional[float] = None,
+        use_karras_sigmas: Optional[bool] = False,
+        use_exponential_sigmas: Optional[bool] = False,
+        use_beta_sigmas: Optional[bool] = False,
+        time_shift_type: str = "exponential",
+        stochastic_sampling: bool = False,
+    ):
+        if self.config.use_beta_sigmas and not is_scipy_available():
+            raise ImportError("Make sure to install scipy if you want to use beta sigmas.")
+        if sum([self.config.use_beta_sigmas, self.config.use_exponential_sigmas, self.config.use_karras_sigmas]) > 1:
+            raise ValueError(
+                "Only one of `config.use_beta_sigmas`, `config.use_exponential_sigmas`, `config.use_karras_sigmas` can be used."
+            )
+        if time_shift_type not in {"exponential", "linear"}:
+            raise ValueError("`time_shift_type` must either be 'exponential' or 'linear'.")
+
+        timesteps = np.linspace(1, num_train_timesteps, num_train_timesteps, dtype=np.float32)[::-1].copy()
+        timesteps = torch.from_numpy(timesteps).to(dtype=torch.float32)
+
+        sigmas = timesteps / num_train_timesteps
+        if not use_dynamic_shifting:
+            # when use_dynamic_shifting is True, we apply the timestep shifting on the fly based on the image resolution
+            sigmas = shift * sigmas / (1 + (shift - 1) * sigmas)
+
+        self.timesteps = sigmas * num_train_timesteps
+
+        self._step_index = None
+        self._begin_index = None
+
+        self._shift = shift
+
+        self.sigmas = sigmas.to("cpu")  # to avoid too much CPU/GPU communication
+        self.sigma_min = self.sigmas[-1].item()
+        self.sigma_max = self.sigmas[0].item()
+
+    @property
+    def shift(self):
+        """
+        The value used for shifting.
+        """
+        return self._shift
+
+    @property
+    def step_index(self):
+        """
+        The index counter for current timestep. It will increase 1 after each scheduler step.
+        """
+        return self._step_index
+
+    @property
+    def begin_index(self):
+        """
+        The index for the first timestep. It should be set from pipeline with `set_begin_index` method.
+        """
+        return self._begin_index
+
+    # Copied from diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler.set_begin_index
+    def set_begin_index(self, begin_index: int = 0):
+        """
+        Sets the begin index for the scheduler. This function should be run from pipeline before the inference.
+
+        Args:
+            begin_index (`int`):
+                The begin index for the scheduler.
+        """
+        self._begin_index = begin_index
+
+    def set_shift(self, shift: float):
+        self._shift = shift
+
+    def scale_noise(
+        self,
+        sample: torch.FloatTensor,
+        timestep: Union[float, torch.FloatTensor],
+        noise: Optional[torch.FloatTensor] = None,
+    ) -> torch.FloatTensor:
+        """
+        Forward process in flow-matching
+
+        Args:
+            sample (`torch.FloatTensor`):
+                The input sample.
+            timestep (`int`, *optional*):
+                The current timestep in the diffusion chain.
+
+        Returns:
+            `torch.FloatTensor`:
+                A scaled input sample.
+        """
+        # Make sure sigmas and timesteps have the same device and dtype as original_samples
+        sigmas = self.sigmas.to(device=sample.device, dtype=sample.dtype)
+
+        if sample.device.type == "mps" and torch.is_floating_point(timestep):
+            # mps does not support float64
+            schedule_timesteps = self.timesteps.to(sample.device, dtype=torch.float32)
+            timestep = timestep.to(sample.device, dtype=torch.float32)
+        else:
+            schedule_timesteps = self.timesteps.to(sample.device)
+            timestep = timestep.to(sample.device)
+
+        # self.begin_index is None when scheduler is used for training, or pipeline does not implement set_begin_index
+        if self.begin_index is None:
+            step_indices = [self.index_for_timestep(t, schedule_timesteps) for t in timestep]
+        elif self.step_index is not None:
+            # add_noise is called after first denoising step (for inpainting)
+            step_indices = [self.step_index] * timestep.shape[0]
+        else:
+            # add noise is called before first denoising step to create initial latent(img2img)
+            step_indices = [self.begin_index] * timestep.shape[0]
+
+        sigma = sigmas[step_indices].flatten()
+        while len(sigma.shape) < len(sample.shape):
+            sigma = sigma.unsqueeze(-1)
+
+        sample = sigma * noise + (1.0 - sigma) * sample
+
+        return sample
+
+    def _sigma_to_t(self, sigma):
+        return sigma * self.config.num_train_timesteps
+
+    def time_shift(self, mu: float, sigma: float, t: torch.Tensor):
+        if self.config.time_shift_type == "exponential":
+            return self._time_shift_exponential(mu, sigma, t)
+        elif self.config.time_shift_type == "linear":
+            return self._time_shift_linear(mu, sigma, t)
+
+    def stretch_shift_to_terminal(self, t: torch.Tensor) -> torch.Tensor:
+        r"""
+        Stretches and shifts the timestep schedule to ensure it terminates at the configured `shift_terminal` config
+        value.
+
+        Reference:
+        https://github.com/Lightricks/LTX-Video/blob/a01a171f8fe3d99dce2728d60a73fecf4d4238ae/ltx_video/schedulers/rf.py#L51
+
+        Args:
+            t (`torch.Tensor`):
+                A tensor of timesteps to be stretched and shifted.
+
+        Returns:
+            `torch.Tensor`:
+                A tensor of adjusted timesteps such that the final value equals `self.config.shift_terminal`.
+        """
+        one_minus_z = 1 - t
+        scale_factor = one_minus_z[-1] / (1 - self.config.shift_terminal)
+        stretched_t = 1 - (one_minus_z / scale_factor)
+        return stretched_t
+
+    def set_timesteps(
+        self,
+        num_inference_steps: Optional[int] = None,
+        device: Union[str, torch.device] = None,
+        sigmas: Optional[List[float]] = None,
+        mu: Optional[float] = None,
+        timesteps: Optional[List[float]] = None,
+    ):
+        """
+        Sets the discrete timesteps used for the diffusion chain (to be run before inference).
+
+        Args:
+            num_inference_steps (`int`, *optional*):
+                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.
+            sigmas (`List[float]`, *optional*):
+                Custom values for sigmas to be used for each diffusion step. If `None`, the sigmas are computed
+                automatically.
+            mu (`float`, *optional*):
+                Determines the amount of shifting applied to sigmas when performing resolution-dependent timestep
+                shifting.
+            timesteps (`List[float]`, *optional*):
+                Custom values for timesteps to be used for each diffusion step. If `None`, the timesteps are computed
+                automatically.
+        """
+        if self.config.use_dynamic_shifting and mu is None:
+            raise ValueError("`mu` must be passed when `use_dynamic_shifting` is set to be `True`")
+
+        if sigmas is not None and timesteps is not None:
+            if len(sigmas) != len(timesteps):
+                raise ValueError("`sigmas` and `timesteps` should have the same length")
+
+        if num_inference_steps is not None:
+            if (sigmas is not None and len(sigmas) != num_inference_steps) or (
+                timesteps is not None and len(timesteps) != num_inference_steps
+            ):
+                raise ValueError(
+                    "`sigmas` and `timesteps` should have the same length as num_inference_steps, if `num_inference_steps` is provided"
+                )
+        else:
+            num_inference_steps = len(sigmas) if sigmas is not None else len(timesteps)
+
+        self.num_inference_steps = num_inference_steps
+
+        # 1. Prepare default sigmas
+        is_timesteps_provided = timesteps is not None
+
+        if is_timesteps_provided:
+            timesteps = np.array(timesteps).astype(np.float32)
+
+        if sigmas is None:
+            if timesteps is None:
+                timesteps = np.linspace(
+                    self._sigma_to_t(self.sigma_max), self._sigma_to_t(self.sigma_min), num_inference_steps
+                )
+            sigmas = timesteps / self.config.num_train_timesteps
+        else:
+            sigmas = np.array(sigmas).astype(np.float32)
+            num_inference_steps = len(sigmas)
+
+        # 2. Perform timestep shifting. Either no shifting is applied, or resolution-dependent shifting of
+        #    "exponential" or "linear" type is applied
+        if self.config.use_dynamic_shifting:
+            sigmas = self.time_shift(mu, 1.0, sigmas)
+        else:
+            sigmas = self.shift * sigmas / (1 + (self.shift - 1) * sigmas)
+
+        # 3. If required, stretch the sigmas schedule to terminate at the configured `shift_terminal` value
+        if self.config.shift_terminal:
+            sigmas = self.stretch_shift_to_terminal(sigmas)
+
+        # 4. If required, convert sigmas to one of karras, exponential, or beta sigma schedules
+        if self.config.use_karras_sigmas:
+            sigmas = self._convert_to_karras(in_sigmas=sigmas, num_inference_steps=num_inference_steps)
+        elif self.config.use_exponential_sigmas:
+            sigmas = self._convert_to_exponential(in_sigmas=sigmas, num_inference_steps=num_inference_steps)
+        elif self.config.use_beta_sigmas:
+            sigmas = self._convert_to_beta(in_sigmas=sigmas, num_inference_steps=num_inference_steps)
+
+        # 5. Convert sigmas and timesteps to tensors and move to specified device
+        sigmas = torch.from_numpy(sigmas).to(dtype=torch.float32, device=device)
+        if not is_timesteps_provided:
+            timesteps = sigmas * self.config.num_train_timesteps
+        else:
+            timesteps = torch.from_numpy(timesteps).to(dtype=torch.float32, device=device)
+
+        # 6. Append the terminal sigma value.
+        #    If a model requires inverted sigma schedule for denoising but timesteps without inversion, the
+        #    `invert_sigmas` flag can be set to `True`. This case is only required in Mochi
+        if self.config.invert_sigmas:
+            sigmas = 1.0 - sigmas
+            timesteps = sigmas * self.config.num_train_timesteps
+            sigmas = torch.cat([sigmas, torch.ones(1, device=sigmas.device)])
+        else:
+            sigmas = torch.cat([sigmas, torch.zeros(1, device=sigmas.device)])
+
+        self.timesteps = timesteps
+        self.sigmas = sigmas
+        self._step_index = None
+        self._begin_index = None
+
+    def index_for_timestep(self, timestep, schedule_timesteps=None):
+        if schedule_timesteps is None:
+            schedule_timesteps = self.timesteps
+
+        indices = (schedule_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)
+        pos = 1 if len(indices) > 1 else 0
+
+        return indices[pos].item()
+
+    def _init_step_index(self, timestep):
+        if self.begin_index is None:
+            if isinstance(timestep, torch.Tensor):
+                timestep = timestep.to(self.timesteps.device)
+            self._step_index = self.index_for_timestep(timestep)
+        else:
+            self._step_index = self._begin_index
+
+    def step(
+        self,
+        model_output: torch.FloatTensor,
+        timestep: Union[float, torch.FloatTensor],
+        sample: torch.FloatTensor,
+        s_churn: float = 0.0,
+        s_tmin: float = 0.0,
+        s_tmax: float = float("inf"),
+        s_noise: float = 1.0,
+        generator: Optional[torch.Generator] = None,
+        per_token_timesteps: Optional[torch.Tensor] = None,
+        return_dict: bool = True,
+    ) -> Union[FlowMatchEulerDiscreteSchedulerOutput, 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.
+            s_churn (`float`):
+            s_tmin  (`float`):
+            s_tmax  (`float`):
+            s_noise (`float`, defaults to 1.0):
+                Scaling factor for noise added to the sample.
+            generator (`torch.Generator`, *optional*):
+                A random number generator.
+            per_token_timesteps (`torch.Tensor`, *optional*):
+                The timesteps for each token in the sample.
+            return_dict (`bool`):
+                Whether or not to return a
+                [`~schedulers.scheduling_flow_match_euler_discrete.FlowMatchEulerDiscreteSchedulerOutput`] or tuple.
+
+        Returns:
+            [`~schedulers.scheduling_flow_match_euler_discrete.FlowMatchEulerDiscreteSchedulerOutput`] or `tuple`:
+                If return_dict is `True`,
+                [`~schedulers.scheduling_flow_match_euler_discrete.FlowMatchEulerDiscreteSchedulerOutput`] is returned,
+                otherwise a tuple is returned where the first element is the sample tensor.
+        """
+
+        if (
+            isinstance(timestep, int)
+            or isinstance(timestep, torch.IntTensor)
+            or isinstance(timestep, torch.LongTensor)
+        ):
+            raise ValueError(
+                (
+                    "Passing integer indices (e.g. from `enumerate(timesteps)`) as timesteps to"
+                    " `FlowMatchEulerDiscreteScheduler.step()` is not supported. Make sure to pass"
+                    " one of the `scheduler.timesteps` as a timestep."
+                ),
+            )
+
+        if self.step_index is None:
+            self._init_step_index(timestep)
+
+        # Upcast to avoid precision issues when computing prev_sample
+        sample = sample.to(torch.float32)
+
+        if per_token_timesteps is not None:
+            per_token_sigmas = per_token_timesteps / self.config.num_train_timesteps
+
+            sigmas = self.sigmas[:, None, None]
+            lower_mask = sigmas < per_token_sigmas[None] - 1e-6
+            lower_sigmas = lower_mask * sigmas
+            lower_sigmas, _ = lower_sigmas.max(dim=0)
+
+            current_sigma = per_token_sigmas[..., None]
+            next_sigma = lower_sigmas[..., None]
+            dt = current_sigma - next_sigma
+        else:
+            sigma_idx = self.step_index
+            sigma = self.sigmas[sigma_idx]
+            sigma_next = self.sigmas[sigma_idx + 1]
+
+            current_sigma = sigma
+            next_sigma = sigma_next
+            dt = sigma_next - sigma
+
+        if self.config.stochastic_sampling:
+            x0 = sample - current_sigma * model_output
+            noise = torch.randn_like(sample)
+            prev_sample = (1.0 - next_sigma) * x0 + next_sigma * noise
+        else:
+            prev_sample = sample + dt * model_output
+
+        # upon completion increase step index by one
+        self._step_index += 1
+        if per_token_timesteps is None:
+            # Cast sample back to model compatible dtype
+            prev_sample = prev_sample.to(model_output.dtype)
+
+        if not return_dict:
+            return (prev_sample,)
+
+        return FlowMatchEulerDiscreteSchedulerOutput(prev_sample=prev_sample)
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._convert_to_karras
+    def _convert_to_karras(self, in_sigmas: torch.Tensor, num_inference_steps) -> torch.Tensor:
+        """Constructs the noise schedule of Karras et al. (2022)."""
+
+        # Hack to make sure that other schedulers which copy this function don't break
+        # TODO: Add this logic to the other schedulers
+        if hasattr(self.config, "sigma_min"):
+            sigma_min = self.config.sigma_min
+        else:
+            sigma_min = None
+
+        if hasattr(self.config, "sigma_max"):
+            sigma_max = self.config.sigma_max
+        else:
+            sigma_max = None
+
+        sigma_min = sigma_min if sigma_min is not None else in_sigmas[-1].item()
+        sigma_max = sigma_max if sigma_max is not None else in_sigmas[0].item()
+
+        rho = 7.0  # 7.0 is the value used in the paper
+        ramp = np.linspace(0, 1, num_inference_steps)
+        min_inv_rho = sigma_min ** (1 / rho)
+        max_inv_rho = sigma_max ** (1 / rho)
+        sigmas = (max_inv_rho + ramp * (min_inv_rho - max_inv_rho)) ** rho
+        return sigmas
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._convert_to_exponential
+    def _convert_to_exponential(self, in_sigmas: torch.Tensor, num_inference_steps: int) -> torch.Tensor:
+        """Constructs an exponential noise schedule."""
+
+        # Hack to make sure that other schedulers which copy this function don't break
+        # TODO: Add this logic to the other schedulers
+        if hasattr(self.config, "sigma_min"):
+            sigma_min = self.config.sigma_min
+        else:
+            sigma_min = None
+
+        if hasattr(self.config, "sigma_max"):
+            sigma_max = self.config.sigma_max
+        else:
+            sigma_max = None
+
+        sigma_min = sigma_min if sigma_min is not None else in_sigmas[-1].item()
+        sigma_max = sigma_max if sigma_max is not None else in_sigmas[0].item()
+
+        sigmas = np.exp(np.linspace(math.log(sigma_max), math.log(sigma_min), num_inference_steps))
+        return sigmas
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._convert_to_beta
+    def _convert_to_beta(
+        self, in_sigmas: torch.Tensor, num_inference_steps: int, alpha: float = 0.6, beta: float = 0.6
+    ) -> torch.Tensor:
+        """From "Beta Sampling is All You Need" [arXiv:2407.12173] (Lee et. al, 2024)"""
+
+        # Hack to make sure that other schedulers which copy this function don't break
+        # TODO: Add this logic to the other schedulers
+        if hasattr(self.config, "sigma_min"):
+            sigma_min = self.config.sigma_min
+        else:
+            sigma_min = None
+
+        if hasattr(self.config, "sigma_max"):
+            sigma_max = self.config.sigma_max
+        else:
+            sigma_max = None
+
+        sigma_min = sigma_min if sigma_min is not None else in_sigmas[-1].item()
+        sigma_max = sigma_max if sigma_max is not None else in_sigmas[0].item()
+
+        sigmas = np.array(
+            [
+                sigma_min + (ppf * (sigma_max - sigma_min))
+                for ppf in [
+                    scipy.stats.beta.ppf(timestep, alpha, beta)
+                    for timestep in 1 - np.linspace(0, 1, num_inference_steps)
+                ]
+            ]
+        )
+        return sigmas
+
+    def _time_shift_exponential(self, mu, sigma, t):
+        return math.exp(mu) / (math.exp(mu) + (1 / t - 1) ** sigma)
+
+    def _time_shift_linear(self, mu, sigma, t):
+        return mu / (mu + (1 / t - 1) ** sigma)
+
+    def __len__(self):
+        return self.config.num_train_timesteps

pythonProject/.venv/Lib/site-packages/diffusers/schedulers/scheduling_flow_match_heun_discrete.py

@@ -0,0 +1,321 @@
+# Copyright 2025 Stability AI, Katherine Crowson 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.
+
+from dataclasses import dataclass
+from typing import Optional, Tuple, Union
+
+import numpy as np
+import torch
+
+from ..configuration_utils import ConfigMixin, register_to_config
+from ..utils import BaseOutput, logging
+from ..utils.torch_utils import randn_tensor
+from .scheduling_utils import SchedulerMixin
+
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+@dataclass
+class FlowMatchHeunDiscreteSchedulerOutput(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.
+    """
+
+    prev_sample: torch.FloatTensor
+
+
+class FlowMatchHeunDiscreteScheduler(SchedulerMixin, ConfigMixin):
+    """
+    Heun scheduler.
+
+    This model inherits from [`SchedulerMixin`] and [`ConfigMixin`]. Check the superclass documentation for the generic
+    methods the library implements for all schedulers such as loading and saving.
+
+    Args:
+        num_train_timesteps (`int`, defaults to 1000):
+            The number of diffusion steps to train the model.
+        timestep_spacing (`str`, defaults to `"linspace"`):
+            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.
+        shift (`float`, defaults to 1.0):
+            The shift value for the timestep schedule.
+    """
+
+    _compatibles = []
+    order = 2
+
+    @register_to_config
+    def __init__(
+        self,
+        num_train_timesteps: int = 1000,
+        shift: float = 1.0,
+    ):
+        timesteps = np.linspace(1, num_train_timesteps, num_train_timesteps, dtype=np.float32)[::-1].copy()
+        timesteps = torch.from_numpy(timesteps).to(dtype=torch.float32)
+
+        sigmas = timesteps / num_train_timesteps
+        sigmas = shift * sigmas / (1 + (shift - 1) * sigmas)
+
+        self.timesteps = sigmas * num_train_timesteps
+
+        self._step_index = None
+        self._begin_index = None
+
+        self.sigmas = sigmas.to("cpu")  # to avoid too much CPU/GPU communication
+        self.sigma_min = self.sigmas[-1].item()
+        self.sigma_max = self.sigmas[0].item()
+
+    @property
+    def step_index(self):
+        """
+        The index counter for current timestep. It will increase 1 after each scheduler step.
+        """
+        return self._step_index
+
+    @property
+    def begin_index(self):
+        """
+        The index for the first timestep. It should be set from pipeline with `set_begin_index` method.
+        """
+        return self._begin_index
+
+    # Copied from diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler.set_begin_index
+    def set_begin_index(self, begin_index: int = 0):
+        """
+        Sets the begin index for the scheduler. This function should be run from pipeline before the inference.
+
+        Args:
+            begin_index (`int`):
+                The begin index for the scheduler.
+        """
+        self._begin_index = begin_index
+
+    def scale_noise(
+        self,
+        sample: torch.FloatTensor,
+        timestep: Union[float, torch.FloatTensor],
+        noise: Optional[torch.FloatTensor] = None,
+    ) -> torch.FloatTensor:
+        """
+        Forward process in flow-matching
+
+        Args:
+            sample (`torch.FloatTensor`):
+                The input sample.
+            timestep (`int`, *optional*):
+                The current timestep in the diffusion chain.
+
+        Returns:
+            `torch.FloatTensor`:
+                A scaled input sample.
+        """
+        if self.step_index is None:
+            self._init_step_index(timestep)
+
+        sigma = self.sigmas[self.step_index]
+        sample = sigma * noise + (1.0 - sigma) * sample
+
+        return sample
+
+    def _sigma_to_t(self, sigma):
+        return sigma * self.config.num_train_timesteps
+
+    def set_timesteps(self, num_inference_steps: int, device: Union[str, torch.device] = 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.
+        """
+        self.num_inference_steps = num_inference_steps
+
+        timesteps = np.linspace(
+            self._sigma_to_t(self.sigma_max), self._sigma_to_t(self.sigma_min), num_inference_steps
+        )
+
+        sigmas = timesteps / self.config.num_train_timesteps
+        sigmas = self.config.shift * sigmas / (1 + (self.config.shift - 1) * sigmas)
+        sigmas = torch.from_numpy(sigmas).to(dtype=torch.float32, device=device)
+
+        timesteps = sigmas * self.config.num_train_timesteps
+        timesteps = torch.cat([timesteps[:1], timesteps[1:].repeat_interleave(2)])
+        self.timesteps = timesteps.to(device=device)
+
+        sigmas = torch.cat([sigmas, torch.zeros(1, device=sigmas.device)])
+        self.sigmas = torch.cat([sigmas[:1], sigmas[1:-1].repeat_interleave(2), sigmas[-1:]])
+
+        # empty dt and derivative
+        self.prev_derivative = None
+        self.dt = None
+
+        self._step_index = None
+        self._begin_index = None
+
+    def index_for_timestep(self, timestep, schedule_timesteps=None):
+        if schedule_timesteps is None:
+            schedule_timesteps = self.timesteps
+
+        indices = (schedule_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)
+        pos = 1 if len(indices) > 1 else 0
+
+        return indices[pos].item()
+
+    def _init_step_index(self, timestep):
+        if self.begin_index is None:
+            if isinstance(timestep, torch.Tensor):
+                timestep = timestep.to(self.timesteps.device)
+            self._step_index = self.index_for_timestep(timestep)
+        else:
+            self._step_index = self._begin_index
+
+    @property
+    def state_in_first_order(self):
+        return self.dt is None
+
+    def step(
+        self,
+        model_output: torch.FloatTensor,
+        timestep: Union[float, torch.FloatTensor],
+        sample: torch.FloatTensor,
+        s_churn: float = 0.0,
+        s_tmin: float = 0.0,
+        s_tmax: float = float("inf"),
+        s_noise: float = 1.0,
+        generator: Optional[torch.Generator] = None,
+        return_dict: bool = True,
+    ) -> Union[FlowMatchHeunDiscreteSchedulerOutput, 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.
+            s_churn (`float`):
+            s_tmin  (`float`):
+            s_tmax  (`float`):
+            s_noise (`float`, defaults to 1.0):
+                Scaling factor for noise added to the sample.
+            generator (`torch.Generator`, *optional*):
+                A random number generator.
+            return_dict (`bool`):
+                Whether or not to return a
+                [`~schedulers.scheduling_flow_match_heun_discrete.FlowMatchHeunDiscreteSchedulerOutput`] tuple.
+
+        Returns:
+            [`~schedulers.scheduling_flow_match_heun_discrete.FlowMatchHeunDiscreteSchedulerOutput`] or `tuple`:
+                If return_dict is `True`,
+                [`~schedulers.scheduling_flow_match_heun_discrete.FlowMatchHeunDiscreteSchedulerOutput`] is returned,
+                otherwise a tuple is returned where the first element is the sample tensor.
+        """
+
+        if (
+            isinstance(timestep, int)
+            or isinstance(timestep, torch.IntTensor)
+            or isinstance(timestep, torch.LongTensor)
+        ):
+            raise ValueError(
+                (
+                    "Passing integer indices (e.g. from `enumerate(timesteps)`) as timesteps to"
+                    " `FlowMatchHeunDiscreteScheduler.step()` is not supported. Make sure to pass"
+                    " one of the `scheduler.timesteps` as a timestep."
+                ),
+            )
+
+        if self.step_index is None:
+            self._init_step_index(timestep)
+
+        # Upcast to avoid precision issues when computing prev_sample
+        sample = sample.to(torch.float32)
+
+        if self.state_in_first_order:
+            sigma = self.sigmas[self.step_index]
+            sigma_next = self.sigmas[self.step_index + 1]
+        else:
+            # 2nd order / Heun's method
+            sigma = self.sigmas[self.step_index - 1]
+            sigma_next = self.sigmas[self.step_index]
+
+        gamma = min(s_churn / (len(self.sigmas) - 1), 2**0.5 - 1) if s_tmin <= sigma <= s_tmax else 0.0
+
+        sigma_hat = sigma * (gamma + 1)
+
+        if gamma > 0:
+            noise = randn_tensor(
+                model_output.shape, dtype=model_output.dtype, device=model_output.device, generator=generator
+            )
+            eps = noise * s_noise
+            sample = sample + eps * (sigma_hat**2 - sigma**2) ** 0.5
+
+        if self.state_in_first_order:
+            # 1. compute predicted original sample (x_0) from sigma-scaled predicted noise
+            denoised = sample - model_output * sigma
+            # 2. convert to an ODE derivative for 1st order
+            derivative = (sample - denoised) / sigma_hat
+            # 3. Delta timestep
+            dt = sigma_next - sigma_hat
+
+            # store for 2nd order step
+            self.prev_derivative = derivative
+            self.dt = dt
+            self.sample = sample
+        else:
+            # 1. compute predicted original sample (x_0) from sigma-scaled predicted noise
+            denoised = sample - model_output * sigma_next
+            # 2. 2nd order / Heun's method
+            derivative = (sample - denoised) / sigma_next
+            derivative = 0.5 * (self.prev_derivative + derivative)
+
+            # 3. take prev timestep & sample
+            dt = self.dt
+            sample = self.sample
+
+            # free dt and derivative
+            # Note, this puts the scheduler in "first order mode"
+            self.prev_derivative = None
+            self.dt = None
+            self.sample = None
+
+        prev_sample = sample + derivative * dt
+        # Cast sample back to model compatible dtype
+        prev_sample = prev_sample.to(model_output.dtype)
+
+        # upon completion increase step index by one
+        self._step_index += 1
+
+        if not return_dict:
+            return (prev_sample,)
+
+        return FlowMatchHeunDiscreteSchedulerOutput(prev_sample=prev_sample)
+
+    def __len__(self):
+        return self.config.num_train_timesteps

pythonProject/.venv/Lib/site-packages/fsspec/__init__.py

@@ -0,0 +1,71 @@
+from . import caching
+from ._version import __version__  # noqa: F401
+from .callbacks import Callback
+from .compression import available_compressions
+from .core import get_fs_token_paths, open, open_files, open_local, url_to_fs
+from .exceptions import FSTimeoutError
+from .mapping import FSMap, get_mapper
+from .registry import (
+    available_protocols,
+    filesystem,
+    get_filesystem_class,
+    register_implementation,
+    registry,
+)
+from .spec import AbstractFileSystem
+
+__all__ = [
+    "AbstractFileSystem",
+    "FSTimeoutError",
+    "FSMap",
+    "filesystem",
+    "register_implementation",
+    "get_filesystem_class",
+    "get_fs_token_paths",
+    "get_mapper",
+    "open",
+    "open_files",
+    "open_local",
+    "registry",
+    "caching",
+    "Callback",
+    "available_protocols",
+    "available_compressions",
+    "url_to_fs",
+]
+
+
+def process_entries():
+    try:
+        from importlib.metadata import entry_points
+    except ImportError:
+        return
+    if entry_points is not None:
+        try:
+            eps = entry_points()
+        except TypeError:
+            pass  # importlib-metadata < 0.8
+        else:
+            if hasattr(eps, "select"):  # Python 3.10+ / importlib_metadata >= 3.9.0
+                specs = eps.select(group="fsspec.specs")
+            else:
+                specs = eps.get("fsspec.specs", [])
+            registered_names = {}
+            for spec in specs:
+                err_msg = f"Unable to load filesystem from {spec}"
+                name = spec.name
+                if name in registered_names:
+                    continue
+                registered_names[name] = True
+                register_implementation(
+                    name,
+                    spec.value.replace(":", "."),
+                    errtxt=err_msg,
+                    # We take our implementations as the ones to overload with if
+                    # for some reason we encounter some, may be the same, already
+                    # registered
+                    clobber=True,
+                )
+
+
+process_entries()

pythonProject/.venv/Lib/site-packages/fsspec/_version.py

@@ -0,0 +1,34 @@
+# file generated by setuptools-scm
+# don't change, don't track in version control
+
+__all__ = [
+    "__version__",
+    "__version_tuple__",
+    "version",
+    "version_tuple",
+    "__commit_id__",
+    "commit_id",
+]
+
+TYPE_CHECKING = False
+if TYPE_CHECKING:
+    from typing import Tuple
+    from typing import Union
+
+    VERSION_TUPLE = Tuple[Union[int, str], ...]
+    COMMIT_ID = Union[str, None]
+else:
+    VERSION_TUPLE = object
+    COMMIT_ID = object
+
+version: str
+__version__: str
+__version_tuple__: VERSION_TUPLE
+version_tuple: VERSION_TUPLE
+commit_id: COMMIT_ID
+__commit_id__: COMMIT_ID
+
+__version__ = version = '2025.9.0'
+__version_tuple__ = version_tuple = (2025, 9, 0)
+
+__commit_id__ = commit_id = None

pythonProject/.venv/Lib/site-packages/fsspec/implementations/arrow.py

@@ -0,0 +1,304 @@
+import errno
+import io
+import os
+import secrets
+import shutil
+from contextlib import suppress
+from functools import cached_property, wraps
+from urllib.parse import parse_qs
+
+from fsspec.spec import AbstractFileSystem
+from fsspec.utils import (
+    get_package_version_without_import,
+    infer_storage_options,
+    mirror_from,
+    tokenize,
+)
+
+
+def wrap_exceptions(func):
+    @wraps(func)
+    def wrapper(*args, **kwargs):
+        try:
+            return func(*args, **kwargs)
+        except OSError as exception:
+            if not exception.args:
+                raise
+
+            message, *args = exception.args
+            if isinstance(message, str) and "does not exist" in message:
+                raise FileNotFoundError(errno.ENOENT, message) from exception
+            else:
+                raise
+
+    return wrapper
+
+
+PYARROW_VERSION = None
+
+
+class ArrowFSWrapper(AbstractFileSystem):
+    """FSSpec-compatible wrapper of pyarrow.fs.FileSystem.
+
+    Parameters
+    ----------
+    fs : pyarrow.fs.FileSystem
+
+    """
+
+    root_marker = "/"
+
+    def __init__(self, fs, **kwargs):
+        global PYARROW_VERSION
+        PYARROW_VERSION = get_package_version_without_import("pyarrow")
+        self.fs = fs
+        super().__init__(**kwargs)
+
+    @property
+    def protocol(self):
+        return self.fs.type_name
+
+    @cached_property
+    def fsid(self):
+        return "hdfs_" + tokenize(self.fs.host, self.fs.port)
+
+    @classmethod
+    def _strip_protocol(cls, path):
+        ops = infer_storage_options(path)
+        path = ops["path"]
+        if path.startswith("//"):
+            # special case for "hdfs://path" (without the triple slash)
+            path = path[1:]
+        return path
+
+    def ls(self, path, detail=False, **kwargs):
+        path = self._strip_protocol(path)
+        from pyarrow.fs import FileSelector
+
+        entries = [
+            self._make_entry(entry)
+            for entry in self.fs.get_file_info(FileSelector(path))
+        ]
+        if detail:
+            return entries
+        else:
+            return [entry["name"] for entry in entries]
+
+    def info(self, path, **kwargs):
+        path = self._strip_protocol(path)
+        [info] = self.fs.get_file_info([path])
+        return self._make_entry(info)
+
+    def exists(self, path):
+        path = self._strip_protocol(path)
+        try:
+            self.info(path)
+        except FileNotFoundError:
+            return False
+        else:
+            return True
+
+    def _make_entry(self, info):
+        from pyarrow.fs import FileType
+
+        if info.type is FileType.Directory:
+            kind = "directory"
+        elif info.type is FileType.File:
+            kind = "file"
+        elif info.type is FileType.NotFound:
+            raise FileNotFoundError(errno.ENOENT, os.strerror(errno.ENOENT), info.path)
+        else:
+            kind = "other"
+
+        return {
+            "name": info.path,
+            "size": info.size,
+            "type": kind,
+            "mtime": info.mtime,
+        }
+
+    @wrap_exceptions
+    def cp_file(self, path1, path2, **kwargs):
+        path1 = self._strip_protocol(path1).rstrip("/")
+        path2 = self._strip_protocol(path2).rstrip("/")
+
+        with self._open(path1, "rb") as lstream:
+            tmp_fname = f"{path2}.tmp.{secrets.token_hex(6)}"
+            try:
+                with self.open(tmp_fname, "wb") as rstream:
+                    shutil.copyfileobj(lstream, rstream)
+                self.fs.move(tmp_fname, path2)
+            except BaseException:
+                with suppress(FileNotFoundError):
+                    self.fs.delete_file(tmp_fname)
+                raise
+
+    @wrap_exceptions
+    def mv(self, path1, path2, **kwargs):
+        path1 = self._strip_protocol(path1).rstrip("/")
+        path2 = self._strip_protocol(path2).rstrip("/")
+        self.fs.move(path1, path2)
+
+    @wrap_exceptions
+    def rm_file(self, path):
+        path = self._strip_protocol(path)
+        self.fs.delete_file(path)
+
+    @wrap_exceptions
+    def rm(self, path, recursive=False, maxdepth=None):
+        path = self._strip_protocol(path).rstrip("/")
+        if self.isdir(path):
+            if recursive:
+                self.fs.delete_dir(path)
+            else:
+                raise ValueError("Can't delete directories without recursive=False")
+        else:
+            self.fs.delete_file(path)
+
+    @wrap_exceptions
+    def _open(self, path, mode="rb", block_size=None, seekable=True, **kwargs):
+        if mode == "rb":
+            if seekable:
+                method = self.fs.open_input_file
+            else:
+                method = self.fs.open_input_stream
+        elif mode == "wb":
+            method = self.fs.open_output_stream
+        elif mode == "ab":
+            method = self.fs.open_append_stream
+        else:
+            raise ValueError(f"unsupported mode for Arrow filesystem: {mode!r}")
+
+        _kwargs = {}
+        if mode != "rb" or not seekable:
+            if int(PYARROW_VERSION.split(".")[0]) >= 4:
+                # disable compression auto-detection
+                _kwargs["compression"] = None
+        stream = method(path, **_kwargs)
+
+        return ArrowFile(self, stream, path, mode, block_size, **kwargs)
+
+    @wrap_exceptions
+    def mkdir(self, path, create_parents=True, **kwargs):
+        path = self._strip_protocol(path)
+        if create_parents:
+            self.makedirs(path, exist_ok=True)
+        else:
+            self.fs.create_dir(path, recursive=False)
+
+    @wrap_exceptions
+    def makedirs(self, path, exist_ok=False):
+        path = self._strip_protocol(path)
+        self.fs.create_dir(path, recursive=True)
+
+    @wrap_exceptions
+    def rmdir(self, path):
+        path = self._strip_protocol(path)
+        self.fs.delete_dir(path)
+
+    @wrap_exceptions
+    def modified(self, path):
+        path = self._strip_protocol(path)
+        return self.fs.get_file_info(path).mtime
+
+    def cat_file(self, path, start=None, end=None, **kwargs):
+        kwargs["seekable"] = start not in [None, 0]
+        return super().cat_file(path, start=None, end=None, **kwargs)
+
+    def get_file(self, rpath, lpath, **kwargs):
+        kwargs["seekable"] = False
+        super().get_file(rpath, lpath, **kwargs)
+
+
+@mirror_from(
+    "stream",
+    [
+        "read",
+        "seek",
+        "tell",
+        "write",
+        "readable",
+        "writable",
+        "close",
+        "size",
+        "seekable",
+    ],
+)
+class ArrowFile(io.IOBase):
+    def __init__(self, fs, stream, path, mode, block_size=None, **kwargs):
+        self.path = path
+        self.mode = mode
+
+        self.fs = fs
+        self.stream = stream
+
+        self.blocksize = self.block_size = block_size
+        self.kwargs = kwargs
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, *args):
+        return self.close()
+
+
+class HadoopFileSystem(ArrowFSWrapper):
+    """A wrapper on top of the pyarrow.fs.HadoopFileSystem
+    to connect it's interface with fsspec"""
+
+    protocol = "hdfs"
+
+    def __init__(
+        self,
+        host="default",
+        port=0,
+        user=None,
+        kerb_ticket=None,
+        replication=3,
+        extra_conf=None,
+        **kwargs,
+    ):
+        """
+
+        Parameters
+        ----------
+        host: str
+            Hostname, IP or "default" to try to read from Hadoop config
+        port: int
+            Port to connect on, or default from Hadoop config if 0
+        user: str or None
+            If given, connect as this username
+        kerb_ticket: str or None
+            If given, use this ticket for authentication
+        replication: int
+            set replication factor of file for write operations. default value is 3.
+        extra_conf: None or dict
+            Passed on to HadoopFileSystem
+        """
+        from pyarrow.fs import HadoopFileSystem
+
+        fs = HadoopFileSystem(
+            host=host,
+            port=port,
+            user=user,
+            kerb_ticket=kerb_ticket,
+            replication=replication,
+            extra_conf=extra_conf,
+        )
+        super().__init__(fs=fs, **kwargs)
+
+    @staticmethod
+    def _get_kwargs_from_urls(path):
+        ops = infer_storage_options(path)
+        out = {}
+        if ops.get("host", None):
+            out["host"] = ops["host"]
+        if ops.get("username", None):
+            out["user"] = ops["username"]
+        if ops.get("port", None):
+            out["port"] = ops["port"]
+        if ops.get("url_query", None):
+            queries = parse_qs(ops["url_query"])
+            if queries.get("replication", None):
+                out["replication"] = int(queries["replication"][0])
+        return out

pythonProject/.venv/Lib/site-packages/fsspec/implementations/asyn_wrapper.py

@@ -0,0 +1,122 @@
+import asyncio
+import functools
+import inspect
+
+import fsspec
+from fsspec.asyn import AsyncFileSystem, running_async
+
+
+def async_wrapper(func, obj=None, semaphore=None):
+    """
+    Wraps a synchronous function to make it awaitable.
+
+    Parameters
+    ----------
+    func : callable
+        The synchronous function to wrap.
+    obj : object, optional
+        The instance to bind the function to, if applicable.
+    semaphore : asyncio.Semaphore, optional
+        A semaphore to limit concurrent calls.
+
+    Returns
+    -------
+    coroutine
+        An awaitable version of the function.
+    """
+
+    @functools.wraps(func)
+    async def wrapper(*args, **kwargs):
+        if semaphore:
+            async with semaphore:
+                return await asyncio.to_thread(func, *args, **kwargs)
+        return await asyncio.to_thread(func, *args, **kwargs)
+
+    return wrapper
+
+
+class AsyncFileSystemWrapper(AsyncFileSystem):
+    """
+    A wrapper class to convert a synchronous filesystem into an asynchronous one.
+
+    This class takes an existing synchronous filesystem implementation and wraps all
+    its methods to provide an asynchronous interface.
+
+    Parameters
+    ----------
+    sync_fs : AbstractFileSystem
+        The synchronous filesystem instance to wrap.
+    """
+
+    protocol = "asyncwrapper", "async_wrapper"
+    cachable = False
+
+    def __init__(
+        self,
+        fs=None,
+        asynchronous=None,
+        target_protocol=None,
+        target_options=None,
+        semaphore=None,
+        max_concurrent_tasks=None,
+        **kwargs,
+    ):
+        if asynchronous is None:
+            asynchronous = running_async()
+        super().__init__(asynchronous=asynchronous, **kwargs)
+        if fs is not None:
+            self.sync_fs = fs
+        else:
+            self.sync_fs = fsspec.filesystem(target_protocol, **target_options)
+        self.protocol = self.sync_fs.protocol
+        self.semaphore = semaphore
+        self._wrap_all_sync_methods()
+
+    @property
+    def fsid(self):
+        return f"async_{self.sync_fs.fsid}"
+
+    def _wrap_all_sync_methods(self):
+        """
+        Wrap all synchronous methods of the underlying filesystem with asynchronous versions.
+        """
+        excluded_methods = {"open"}
+        for method_name in dir(self.sync_fs):
+            if method_name.startswith("_") or method_name in excluded_methods:
+                continue
+
+            attr = inspect.getattr_static(self.sync_fs, method_name)
+            if isinstance(attr, property):
+                continue
+
+            method = getattr(self.sync_fs, method_name)
+            if callable(method) and not inspect.iscoroutinefunction(method):
+                async_method = async_wrapper(method, obj=self, semaphore=self.semaphore)
+                setattr(self, f"_{method_name}", async_method)
+
+    @classmethod
+    def wrap_class(cls, sync_fs_class):
+        """
+        Create a new class that can be used to instantiate an AsyncFileSystemWrapper
+        with lazy instantiation of the underlying synchronous filesystem.
+
+        Parameters
+        ----------
+        sync_fs_class : type
+            The class of the synchronous filesystem to wrap.
+
+        Returns
+        -------
+        type
+            A new class that wraps the provided synchronous filesystem class.
+        """
+
+        class GeneratedAsyncFileSystemWrapper(cls):
+            def __init__(self, *args, **kwargs):
+                sync_fs = sync_fs_class(*args, **kwargs)
+                super().__init__(sync_fs)
+
+        GeneratedAsyncFileSystemWrapper.__name__ = (
+            f"Async{sync_fs_class.__name__}Wrapper"
+        )
+        return GeneratedAsyncFileSystemWrapper

pythonProject/.venv/Lib/site-packages/fsspec/implementations/cache_mapper.py

@@ -0,0 +1,75 @@
+from __future__ import annotations
+
+import abc
+import hashlib
+
+from fsspec.implementations.local import make_path_posix
+
+
+class AbstractCacheMapper(abc.ABC):
+    """Abstract super-class for mappers from remote URLs to local cached
+    basenames.
+    """
+
+    @abc.abstractmethod
+    def __call__(self, path: str) -> str: ...
+
+    def __eq__(self, other: object) -> bool:
+        # Identity only depends on class. When derived classes have attributes
+        # they will need to be included.
+        return isinstance(other, type(self))
+
+    def __hash__(self) -> int:
+        # Identity only depends on class. When derived classes have attributes
+        # they will need to be included.
+        return hash(type(self))
+
+
+class BasenameCacheMapper(AbstractCacheMapper):
+    """Cache mapper that uses the basename of the remote URL and a fixed number
+    of directory levels above this.
+
+    The default is zero directory levels, meaning different paths with the same
+    basename will have the same cached basename.
+    """
+
+    def __init__(self, directory_levels: int = 0):
+        if directory_levels < 0:
+            raise ValueError(
+                "BasenameCacheMapper requires zero or positive directory_levels"
+            )
+        self.directory_levels = directory_levels
+
+        # Separator for directories when encoded as strings.
+        self._separator = "_@_"
+
+    def __call__(self, path: str) -> str:
+        path = make_path_posix(path)
+        prefix, *bits = path.rsplit("/", self.directory_levels + 1)
+        if bits:
+            return self._separator.join(bits)
+        else:
+            return prefix  # No separator found, simple filename
+
+    def __eq__(self, other: object) -> bool:
+        return super().__eq__(other) and self.directory_levels == other.directory_levels
+
+    def __hash__(self) -> int:
+        return super().__hash__() ^ hash(self.directory_levels)
+
+
+class HashCacheMapper(AbstractCacheMapper):
+    """Cache mapper that uses a hash of the remote URL."""
+
+    def __call__(self, path: str) -> str:
+        return hashlib.sha256(path.encode()).hexdigest()
+
+
+def create_cache_mapper(same_names: bool) -> AbstractCacheMapper:
+    """Factory method to create cache mapper for backward compatibility with
+    ``CachingFileSystem`` constructor using ``same_names`` kwarg.
+    """
+    if same_names:
+        return BasenameCacheMapper()
+    else:
+        return HashCacheMapper()

pythonProject/.venv/Lib/site-packages/fsspec/implementations/cache_metadata.py

@@ -0,0 +1,233 @@
+from __future__ import annotations
+
+import os
+import pickle
+import time
+from typing import TYPE_CHECKING
+
+from fsspec.utils import atomic_write
+
+try:
+    import ujson as json
+except ImportError:
+    if not TYPE_CHECKING:
+        import json
+
+if TYPE_CHECKING:
+    from collections.abc import Iterator
+    from typing import Any, Literal
+
+    from typing_extensions import TypeAlias
+
+    from .cached import CachingFileSystem
+
+    Detail: TypeAlias = dict[str, Any]
+
+
+class CacheMetadata:
+    """Cache metadata.
+
+    All reading and writing of cache metadata is performed by this class,
+    accessing the cached files and blocks is not.
+
+    Metadata is stored in a single file per storage directory in JSON format.
+    For backward compatibility, also reads metadata stored in pickle format
+    which is converted to JSON when next saved.
+    """
+
+    def __init__(self, storage: list[str]):
+        """
+
+        Parameters
+        ----------
+        storage: list[str]
+            Directories containing cached files, must be at least one. Metadata
+            is stored in the last of these directories by convention.
+        """
+        if not storage:
+            raise ValueError("CacheMetadata expects at least one storage location")
+
+        self._storage = storage
+        self.cached_files: list[Detail] = [{}]
+
+        # Private attribute to force saving of metadata in pickle format rather than
+        # JSON for use in tests to confirm can read both pickle and JSON formats.
+        self._force_save_pickle = False
+
+    def _load(self, fn: str) -> Detail:
+        """Low-level function to load metadata from specific file"""
+        try:
+            with open(fn, "r") as f:
+                loaded = json.load(f)
+        except ValueError:
+            with open(fn, "rb") as f:
+                loaded = pickle.load(f)
+        for c in loaded.values():
+            if isinstance(c.get("blocks"), list):
+                c["blocks"] = set(c["blocks"])
+        return loaded
+
+    def _save(self, metadata_to_save: Detail, fn: str) -> None:
+        """Low-level function to save metadata to specific file"""
+        if self._force_save_pickle:
+            with atomic_write(fn) as f:
+                pickle.dump(metadata_to_save, f)
+        else:
+            with atomic_write(fn, mode="w") as f:
+                json.dump(metadata_to_save, f)
+
+    def _scan_locations(
+        self, writable_only: bool = False
+    ) -> Iterator[tuple[str, str, bool]]:
+        """Yield locations (filenames) where metadata is stored, and whether
+        writable or not.
+
+        Parameters
+        ----------
+        writable: bool
+            Set to True to only yield writable locations.
+
+        Returns
+        -------
+        Yields (str, str, bool)
+        """
+        n = len(self._storage)
+        for i, storage in enumerate(self._storage):
+            writable = i == n - 1
+            if writable_only and not writable:
+                continue
+            yield os.path.join(storage, "cache"), storage, writable
+
+    def check_file(
+        self, path: str, cfs: CachingFileSystem | None
+    ) -> Literal[False] | tuple[Detail, str]:
+        """If path is in cache return its details, otherwise return ``False``.
+
+        If the optional CachingFileSystem is specified then it is used to
+        perform extra checks to reject possible matches, such as if they are
+        too old.
+        """
+        for (fn, base, _), cache in zip(self._scan_locations(), self.cached_files):
+            if path not in cache:
+                continue
+            detail = cache[path].copy()
+
+            if cfs is not None:
+                if cfs.check_files and detail["uid"] != cfs.fs.ukey(path):
+                    # Wrong file as determined by hash of file properties
+                    continue
+                if cfs.expiry and time.time() - detail["time"] > cfs.expiry:
+                    # Cached file has expired
+                    continue
+
+            fn = os.path.join(base, detail["fn"])
+            if os.path.exists(fn):
+                return detail, fn
+        return False
+
+    def clear_expired(self, expiry_time: int) -> tuple[list[str], bool]:
+        """Remove expired metadata from the cache.
+
+        Returns names of files corresponding to expired metadata and a boolean
+        flag indicating whether the writable cache is empty. Caller is
+        responsible for deleting the expired files.
+        """
+        expired_files = []
+        for path, detail in self.cached_files[-1].copy().items():
+            if time.time() - detail["time"] > expiry_time:
+                fn = detail.get("fn", "")
+                if not fn:
+                    raise RuntimeError(
+                        f"Cache metadata does not contain 'fn' for {path}"
+                    )
+                fn = os.path.join(self._storage[-1], fn)
+                expired_files.append(fn)
+                self.cached_files[-1].pop(path)
+
+        if self.cached_files[-1]:
+            cache_path = os.path.join(self._storage[-1], "cache")
+            self._save(self.cached_files[-1], cache_path)
+
+        writable_cache_empty = not self.cached_files[-1]
+        return expired_files, writable_cache_empty
+
+    def load(self) -> None:
+        """Load all metadata from disk and store in ``self.cached_files``"""
+        cached_files = []
+        for fn, _, _ in self._scan_locations():
+            if os.path.exists(fn):
+                # TODO: consolidate blocks here
+                cached_files.append(self._load(fn))
+            else:
+                cached_files.append({})
+        self.cached_files = cached_files or [{}]
+
+    def on_close_cached_file(self, f: Any, path: str) -> None:
+        """Perform side-effect actions on closing a cached file.
+
+        The actual closing of the file is the responsibility of the caller.
+        """
+        # File must be writeble, so in self.cached_files[-1]
+        c = self.cached_files[-1][path]
+        if c["blocks"] is not True and len(c["blocks"]) * f.blocksize >= f.size:
+            c["blocks"] = True
+
+    def pop_file(self, path: str) -> str | None:
+        """Remove metadata of cached file.
+
+        If path is in the cache, return the filename of the cached file,
+        otherwise return ``None``.  Caller is responsible for deleting the
+        cached file.
+        """
+        details = self.check_file(path, None)
+        if not details:
+            return None
+        _, fn = details
+        if fn.startswith(self._storage[-1]):
+            self.cached_files[-1].pop(path)
+            self.save()
+        else:
+            raise PermissionError(
+                "Can only delete cached file in last, writable cache location"
+            )
+        return fn
+
+    def save(self) -> None:
+        """Save metadata to disk"""
+        for (fn, _, writable), cache in zip(self._scan_locations(), self.cached_files):
+            if not writable:
+                continue
+
+            if os.path.exists(fn):
+                cached_files = self._load(fn)
+                for k, c in cached_files.items():
+                    if k in cache:
+                        if c["blocks"] is True or cache[k]["blocks"] is True:
+                            c["blocks"] = True
+                        else:
+                            # self.cached_files[*][*]["blocks"] must continue to
+                            # point to the same set object so that updates
+                            # performed by MMapCache are propagated back to
+                            # self.cached_files.
+                            blocks = cache[k]["blocks"]
+                            blocks.update(c["blocks"])
+                            c["blocks"] = blocks
+                        c["time"] = max(c["time"], cache[k]["time"])
+                        c["uid"] = cache[k]["uid"]
+
+                # Files can be added to cache after it was written once
+                for k, c in cache.items():
+                    if k not in cached_files:
+                        cached_files[k] = c
+            else:
+                cached_files = cache
+            cache = {k: v.copy() for k, v in cached_files.items()}
+            for c in cache.values():
+                if isinstance(c["blocks"], set):
+                    c["blocks"] = list(c["blocks"])
+            self._save(cache, fn)
+            self.cached_files[-1] = cached_files
+
+    def update_file(self, path: str, detail: Detail) -> None:
+        """Update metadata for specific file in memory, do not save"""
+        self.cached_files[-1][path] = detail

pythonProject/.venv/Lib/site-packages/fsspec/implementations/cached.py

@@ -0,0 +1,998 @@
+from __future__ import annotations
+
+import inspect
+import logging
+import os
+import tempfile
+import time
+import weakref
+from shutil import rmtree
+from typing import TYPE_CHECKING, Any, Callable, ClassVar
+
+from fsspec import AbstractFileSystem, filesystem
+from fsspec.callbacks import DEFAULT_CALLBACK
+from fsspec.compression import compr
+from fsspec.core import BaseCache, MMapCache
+from fsspec.exceptions import BlocksizeMismatchError
+from fsspec.implementations.cache_mapper import create_cache_mapper
+from fsspec.implementations.cache_metadata import CacheMetadata
+from fsspec.implementations.local import LocalFileSystem
+from fsspec.spec import AbstractBufferedFile
+from fsspec.transaction import Transaction
+from fsspec.utils import infer_compression
+
+if TYPE_CHECKING:
+    from fsspec.implementations.cache_mapper import AbstractCacheMapper
+
+logger = logging.getLogger("fsspec.cached")
+
+
+class WriteCachedTransaction(Transaction):
+    def complete(self, commit=True):
+        rpaths = [f.path for f in self.files]
+        lpaths = [f.fn for f in self.files]
+        if commit:
+            self.fs.put(lpaths, rpaths)
+        self.files.clear()
+        self.fs._intrans = False
+        self.fs._transaction = None
+        self.fs = None  # break cycle
+
+
+class CachingFileSystem(AbstractFileSystem):
+    """Locally caching filesystem, layer over any other FS
+
+    This class implements chunk-wise local storage of remote files, for quick
+    access after the initial download. The files are stored in a given
+    directory with hashes of URLs for the filenames. If no directory is given,
+    a temporary one is used, which should be cleaned up by the OS after the
+    process ends. The files themselves are sparse (as implemented in
+    :class:`~fsspec.caching.MMapCache`), so only the data which is accessed
+    takes up space.
+
+    Restrictions:
+
+    - the block-size must be the same for each access of a given file, unless
+      all blocks of the file have already been read
+    - caching can only be applied to file-systems which produce files
+      derived from fsspec.spec.AbstractBufferedFile ; LocalFileSystem is also
+      allowed, for testing
+    """
+
+    protocol: ClassVar[str | tuple[str, ...]] = ("blockcache", "cached")
+
+    def __init__(
+        self,
+        target_protocol=None,
+        cache_storage="TMP",
+        cache_check=10,
+        check_files=False,
+        expiry_time=604800,
+        target_options=None,
+        fs=None,
+        same_names: bool | None = None,
+        compression=None,
+        cache_mapper: AbstractCacheMapper | None = None,
+        **kwargs,
+    ):
+        """
+
+        Parameters
+        ----------
+        target_protocol: str (optional)
+            Target filesystem protocol. Provide either this or ``fs``.
+        cache_storage: str or list(str)
+            Location to store files. If "TMP", this is a temporary directory,
+            and will be cleaned up by the OS when this process ends (or later).
+            If a list, each location will be tried in the order given, but
+            only the last will be considered writable.
+        cache_check: int
+            Number of seconds between reload of cache metadata
+        check_files: bool
+            Whether to explicitly see if the UID of the remote file matches
+            the stored one before using. Warning: some file systems such as
+            HTTP cannot reliably give a unique hash of the contents of some
+            path, so be sure to set this option to False.
+        expiry_time: int
+            The time in seconds after which a local copy is considered useless.
+            Set to falsy to prevent expiry. The default is equivalent to one
+            week.
+        target_options: dict or None
+            Passed to the instantiation of the FS, if fs is None.
+        fs: filesystem instance
+            The target filesystem to run against. Provide this or ``protocol``.
+        same_names: bool (optional)
+            By default, target URLs are hashed using a ``HashCacheMapper`` so
+            that files from different backends with the same basename do not
+            conflict. If this argument is ``true``, a ``BasenameCacheMapper``
+            is used instead. Other cache mapper options are available by using
+            the ``cache_mapper`` keyword argument. Only one of this and
+            ``cache_mapper`` should be specified.
+        compression: str (optional)
+            To decompress on download. Can be 'infer' (guess from the URL name),
+            one of the entries in ``fsspec.compression.compr``, or None for no
+            decompression.
+        cache_mapper: AbstractCacheMapper (optional)
+            The object use to map from original filenames to cached filenames.
+            Only one of this and ``same_names`` should be specified.
+        """
+        super().__init__(**kwargs)
+        if fs is None and target_protocol is None:
+            raise ValueError(
+                "Please provide filesystem instance(fs) or target_protocol"
+            )
+        if not (fs is None) ^ (target_protocol is None):
+            raise ValueError(
+                "Both filesystems (fs) and target_protocol may not be both given."
+            )
+        if cache_storage == "TMP":
+            tempdir = tempfile.mkdtemp()
+            storage = [tempdir]
+            weakref.finalize(self, self._remove_tempdir, tempdir)
+        else:
+            if isinstance(cache_storage, str):
+                storage = [cache_storage]
+            else:
+                storage = cache_storage
+        os.makedirs(storage[-1], exist_ok=True)
+        self.storage = storage
+        self.kwargs = target_options or {}
+        self.cache_check = cache_check
+        self.check_files = check_files
+        self.expiry = expiry_time
+        self.compression = compression
+
+        # Size of cache in bytes. If None then the size is unknown and will be
+        # recalculated the next time cache_size() is called. On writes to the
+        # cache this is reset to None.
+        self._cache_size = None
+
+        if same_names is not None and cache_mapper is not None:
+            raise ValueError(
+                "Cannot specify both same_names and cache_mapper in "
+                "CachingFileSystem.__init__"
+            )
+        if cache_mapper is not None:
+            self._mapper = cache_mapper
+        else:
+            self._mapper = create_cache_mapper(
+                same_names if same_names is not None else False
+            )
+
+        self.target_protocol = (
+            target_protocol
+            if isinstance(target_protocol, str)
+            else (fs.protocol if isinstance(fs.protocol, str) else fs.protocol[0])
+        )
+        self._metadata = CacheMetadata(self.storage)
+        self.load_cache()
+        self.fs = fs if fs is not None else filesystem(target_protocol, **self.kwargs)
+
+        def _strip_protocol(path):
+            # acts as a method, since each instance has a difference target
+            return self.fs._strip_protocol(type(self)._strip_protocol(path))
+
+        self._strip_protocol: Callable = _strip_protocol
+
+    @staticmethod
+    def _remove_tempdir(tempdir):
+        try:
+            rmtree(tempdir)
+        except Exception:
+            pass
+
+    def _mkcache(self):
+        os.makedirs(self.storage[-1], exist_ok=True)
+
+    def cache_size(self):
+        """Return size of cache in bytes.
+
+        If more than one cache directory is in use, only the size of the last
+        one (the writable cache directory) is returned.
+        """
+        if self._cache_size is None:
+            cache_dir = self.storage[-1]
+            self._cache_size = filesystem("file").du(cache_dir, withdirs=True)
+        return self._cache_size
+
+    def load_cache(self):
+        """Read set of stored blocks from file"""
+        self._metadata.load()
+        self._mkcache()
+        self.last_cache = time.time()
+
+    def save_cache(self):
+        """Save set of stored blocks from file"""
+        self._mkcache()
+        self._metadata.save()
+        self.last_cache = time.time()
+        self._cache_size = None
+
+    def _check_cache(self):
+        """Reload caches if time elapsed or any disappeared"""
+        self._mkcache()
+        if not self.cache_check:
+            # explicitly told not to bother checking
+            return
+        timecond = time.time() - self.last_cache > self.cache_check
+        existcond = all(os.path.exists(storage) for storage in self.storage)
+        if timecond or not existcond:
+            self.load_cache()
+
+    def _check_file(self, path):
+        """Is path in cache and still valid"""
+        path = self._strip_protocol(path)
+        self._check_cache()
+        return self._metadata.check_file(path, self)
+
+    def clear_cache(self):
+        """Remove all files and metadata from the cache
+
+        In the case of multiple cache locations, this clears only the last one,
+        which is assumed to be the read/write one.
+        """
+        rmtree(self.storage[-1])
+        self.load_cache()
+        self._cache_size = None
+
+    def clear_expired_cache(self, expiry_time=None):
+        """Remove all expired files and metadata from the cache
+
+        In the case of multiple cache locations, this clears only the last one,
+        which is assumed to be the read/write one.
+
+        Parameters
+        ----------
+        expiry_time: int
+            The time in seconds after which a local copy is considered useless.
+            If not defined the default is equivalent to the attribute from the
+            file caching instantiation.
+        """
+
+        if not expiry_time:
+            expiry_time = self.expiry
+
+        self._check_cache()
+
+        expired_files, writable_cache_empty = self._metadata.clear_expired(expiry_time)
+        for fn in expired_files:
+            if os.path.exists(fn):
+                os.remove(fn)
+
+        if writable_cache_empty:
+            rmtree(self.storage[-1])
+            self.load_cache()
+
+        self._cache_size = None
+
+    def pop_from_cache(self, path):
+        """Remove cached version of given file
+
+        Deletes local copy of the given (remote) path. If it is found in a cache
+        location which is not the last, it is assumed to be read-only, and
+        raises PermissionError
+        """
+        path = self._strip_protocol(path)
+        fn = self._metadata.pop_file(path)
+        if fn is not None:
+            os.remove(fn)
+        self._cache_size = None
+
+    def _open(
+        self,
+        path,
+        mode="rb",
+        block_size=None,
+        autocommit=True,
+        cache_options=None,
+        **kwargs,
+    ):
+        """Wrap the target _open
+
+        If the whole file exists in the cache, just open it locally and
+        return that.
+
+        Otherwise, open the file on the target FS, and make it have a mmap
+        cache pointing to the location which we determine, in our cache.
+        The ``blocks`` instance is shared, so as the mmap cache instance
+        updates, so does the entry in our ``cached_files`` attribute.
+        We monkey-patch this file, so that when it closes, we call
+        ``close_and_update`` to save the state of the blocks.
+        """
+        path = self._strip_protocol(path)
+
+        path = self.fs._strip_protocol(path)
+        if "r" not in mode:
+            return self.fs._open(
+                path,
+                mode=mode,
+                block_size=block_size,
+                autocommit=autocommit,
+                cache_options=cache_options,
+                **kwargs,
+            )
+        detail = self._check_file(path)
+        if detail:
+            # file is in cache
+            detail, fn = detail
+            hash, blocks = detail["fn"], detail["blocks"]
+            if blocks is True:
+                # stored file is complete
+                logger.debug("Opening local copy of %s", path)
+                return open(fn, mode)
+            # TODO: action where partial file exists in read-only cache
+            logger.debug("Opening partially cached copy of %s", path)
+        else:
+            hash = self._mapper(path)
+            fn = os.path.join(self.storage[-1], hash)
+            blocks = set()
+            detail = {
+                "original": path,
+                "fn": hash,
+                "blocks": blocks,
+                "time": time.time(),
+                "uid": self.fs.ukey(path),
+            }
+            self._metadata.update_file(path, detail)
+            logger.debug("Creating local sparse file for %s", path)
+
+        # explicitly submitting the size to the open call will avoid extra
+        # operations when opening. This is particularly relevant
+        # for any file that is read over a network, e.g. S3.
+        size = detail.get("size")
+
+        # call target filesystems open
+        self._mkcache()
+        f = self.fs._open(
+            path,
+            mode=mode,
+            block_size=block_size,
+            autocommit=autocommit,
+            cache_options=cache_options,
+            cache_type="none",
+            size=size,
+            **kwargs,
+        )
+
+        # set size if not already set
+        if size is None:
+            detail["size"] = f.size
+            self._metadata.update_file(path, detail)
+
+        if self.compression:
+            comp = (
+                infer_compression(path)
+                if self.compression == "infer"
+                else self.compression
+            )
+            f = compr[comp](f, mode="rb")
+        if "blocksize" in detail:
+            if detail["blocksize"] != f.blocksize:
+                raise BlocksizeMismatchError(
+                    f"Cached file must be reopened with same block"
+                    f" size as original (old: {detail['blocksize']},"
+                    f" new {f.blocksize})"
+                )
+        else:
+            detail["blocksize"] = f.blocksize
+
+        def _fetch_ranges(ranges):
+            return self.fs.cat_ranges(
+                [path] * len(ranges),
+                [r[0] for r in ranges],
+                [r[1] for r in ranges],
+                **kwargs,
+            )
+
+        multi_fetcher = None if self.compression else _fetch_ranges
+        f.cache = MMapCache(
+            f.blocksize, f._fetch_range, f.size, fn, blocks, multi_fetcher=multi_fetcher
+        )
+        close = f.close
+        f.close = lambda: self.close_and_update(f, close)
+        self.save_cache()
+        return f
+
+    def _parent(self, path):
+        return self.fs._parent(path)
+
+    def hash_name(self, path: str, *args: Any) -> str:
+        # Kept for backward compatibility with downstream libraries.
+        # Ignores extra arguments, previously same_name boolean.
+        return self._mapper(path)
+
+    def close_and_update(self, f, close):
+        """Called when a file is closing, so store the set of blocks"""
+        if f.closed:
+            return
+        path = self._strip_protocol(f.path)
+        self._metadata.on_close_cached_file(f, path)
+        try:
+            logger.debug("going to save")
+            self.save_cache()
+            logger.debug("saved")
+        except OSError:
+            logger.debug("Cache saving failed while closing file")
+        except NameError:
+            logger.debug("Cache save failed due to interpreter shutdown")
+        close()
+        f.closed = True
+
+    def ls(self, path, detail=True):
+        return self.fs.ls(path, detail)
+
+    def __getattribute__(self, item):
+        if item in {
+            "load_cache",
+            "_open",
+            "save_cache",
+            "close_and_update",
+            "__init__",
+            "__getattribute__",
+            "__reduce__",
+            "_make_local_details",
+            "open",
+            "cat",
+            "cat_file",
+            "_cat_file",
+            "cat_ranges",
+            "_cat_ranges",
+            "get",
+            "read_block",
+            "tail",
+            "head",
+            "info",
+            "ls",
+            "exists",
+            "isfile",
+            "isdir",
+            "_check_file",
+            "_check_cache",
+            "_mkcache",
+            "clear_cache",
+            "clear_expired_cache",
+            "pop_from_cache",
+            "local_file",
+            "_paths_from_path",
+            "get_mapper",
+            "open_many",
+            "commit_many",
+            "hash_name",
+            "__hash__",
+            "__eq__",
+            "to_json",
+            "to_dict",
+            "cache_size",
+            "pipe_file",
+            "pipe",
+            "start_transaction",
+            "end_transaction",
+        }:
+            # all the methods defined in this class. Note `open` here, since
+            # it calls `_open`, but is actually in superclass
+            return lambda *args, **kw: getattr(type(self), item).__get__(self)(
+                *args, **kw
+            )
+        if item in ["__reduce_ex__"]:
+            raise AttributeError
+        if item in ["transaction"]:
+            # property
+            return type(self).transaction.__get__(self)
+        if item in {"_cache", "transaction_type", "protocol"}:
+            # class attributes
+            return getattr(type(self), item)
+        if item == "__class__":
+            return type(self)
+        d = object.__getattribute__(self, "__dict__")
+        fs = d.get("fs", None)  # fs is not immediately defined
+        if item in d:
+            return d[item]
+        elif fs is not None:
+            if item in fs.__dict__:
+                # attribute of instance
+                return fs.__dict__[item]
+            # attributed belonging to the target filesystem
+            cls = type(fs)
+            m = getattr(cls, item)
+            if (inspect.isfunction(m) or inspect.isdatadescriptor(m)) and (
+                not hasattr(m, "__self__") or m.__self__ is None
+            ):
+                # instance method
+                return m.__get__(fs, cls)
+            return m  # class method or attribute
+        else:
+            # attributes of the superclass, while target is being set up
+            return super().__getattribute__(item)
+
+    def __eq__(self, other):
+        """Test for equality."""
+        if self is other:
+            return True
+        if not isinstance(other, type(self)):
+            return False
+        return (
+            self.storage == other.storage
+            and self.kwargs == other.kwargs
+            and self.cache_check == other.cache_check
+            and self.check_files == other.check_files
+            and self.expiry == other.expiry
+            and self.compression == other.compression
+            and self._mapper == other._mapper
+            and self.target_protocol == other.target_protocol
+        )
+
+    def __hash__(self):
+        """Calculate hash."""
+        return (
+            hash(tuple(self.storage))
+            ^ hash(str(self.kwargs))
+            ^ hash(self.cache_check)
+            ^ hash(self.check_files)
+            ^ hash(self.expiry)
+            ^ hash(self.compression)
+            ^ hash(self._mapper)
+            ^ hash(self.target_protocol)
+        )
+
+
+class WholeFileCacheFileSystem(CachingFileSystem):
+    """Caches whole remote files on first access
+
+    This class is intended as a layer over any other file system, and
+    will make a local copy of each file accessed, so that all subsequent
+    reads are local. This is similar to ``CachingFileSystem``, but without
+    the block-wise functionality and so can work even when sparse files
+    are not allowed. See its docstring for definition of the init
+    arguments.
+
+    The class still needs access to the remote store for listing files,
+    and may refresh cached files.
+    """
+
+    protocol = "filecache"
+    local_file = True
+
+    def open_many(self, open_files, **kwargs):
+        paths = [of.path for of in open_files]
+        if "r" in open_files.mode:
+            self._mkcache()
+        else:
+            return [
+                LocalTempFile(
+                    self.fs,
+                    path,
+                    mode=open_files.mode,
+                    fn=os.path.join(self.storage[-1], self._mapper(path)),
+                    **kwargs,
+                )
+                for path in paths
+            ]
+
+        if self.compression:
+            raise NotImplementedError
+        details = [self._check_file(sp) for sp in paths]
+        downpath = [p for p, d in zip(paths, details) if not d]
+        downfn0 = [
+            os.path.join(self.storage[-1], self._mapper(p))
+            for p, d in zip(paths, details)
+        ]  # keep these path names for opening later
+        downfn = [fn for fn, d in zip(downfn0, details) if not d]
+        if downpath:
+            # skip if all files are already cached and up to date
+            self.fs.get(downpath, downfn)
+
+            # update metadata - only happens when downloads are successful
+            newdetail = [
+                {
+                    "original": path,
+                    "fn": self._mapper(path),
+                    "blocks": True,
+                    "time": time.time(),
+                    "uid": self.fs.ukey(path),
+                }
+                for path in downpath
+            ]
+            for path, detail in zip(downpath, newdetail):
+                self._metadata.update_file(path, detail)
+            self.save_cache()
+
+        def firstpart(fn):
+            # helper to adapt both whole-file and simple-cache
+            return fn[1] if isinstance(fn, tuple) else fn
+
+        return [
+            open(firstpart(fn0) if fn0 else fn1, mode=open_files.mode)
+            for fn0, fn1 in zip(details, downfn0)
+        ]
+
+    def commit_many(self, open_files):
+        self.fs.put([f.fn for f in open_files], [f.path for f in open_files])
+        [f.close() for f in open_files]
+        for f in open_files:
+            # in case autocommit is off, and so close did not already delete
+            try:
+                os.remove(f.name)
+            except FileNotFoundError:
+                pass
+        self._cache_size = None
+
+    def _make_local_details(self, path):
+        hash = self._mapper(path)
+        fn = os.path.join(self.storage[-1], hash)
+        detail = {
+            "original": path,
+            "fn": hash,
+            "blocks": True,
+            "time": time.time(),
+            "uid": self.fs.ukey(path),
+        }
+        self._metadata.update_file(path, detail)
+        logger.debug("Copying %s to local cache", path)
+        return fn
+
+    def cat(
+        self,
+        path,
+        recursive=False,
+        on_error="raise",
+        callback=DEFAULT_CALLBACK,
+        **kwargs,
+    ):
+        paths = self.expand_path(
+            path, recursive=recursive, maxdepth=kwargs.get("maxdepth")
+        )
+        getpaths = []
+        storepaths = []
+        fns = []
+        out = {}
+        for p in paths.copy():
+            try:
+                detail = self._check_file(p)
+                if not detail:
+                    fn = self._make_local_details(p)
+                    getpaths.append(p)
+                    storepaths.append(fn)
+                else:
+                    detail, fn = detail if isinstance(detail, tuple) else (None, detail)
+                fns.append(fn)
+            except Exception as e:
+                if on_error == "raise":
+                    raise
+                if on_error == "return":
+                    out[p] = e
+                paths.remove(p)
+
+        if getpaths:
+            self.fs.get(getpaths, storepaths)
+            self.save_cache()
+
+        callback.set_size(len(paths))
+        for p, fn in zip(paths, fns):
+            with open(fn, "rb") as f:
+                out[p] = f.read()
+            callback.relative_update(1)
+        if isinstance(path, str) and len(paths) == 1 and recursive is False:
+            out = out[paths[0]]
+        return out
+
+    def _open(self, path, mode="rb", **kwargs):
+        path = self._strip_protocol(path)
+        if "r" not in mode:
+            hash = self._mapper(path)
+            fn = os.path.join(self.storage[-1], hash)
+            user_specified_kwargs = {
+                k: v
+                for k, v in kwargs.items()
+                # those kwargs were added by open(), we don't want them
+                if k not in ["autocommit", "block_size", "cache_options"]
+            }
+            return LocalTempFile(self, path, mode=mode, fn=fn, **user_specified_kwargs)
+        detail = self._check_file(path)
+        if detail:
+            detail, fn = detail
+            _, blocks = detail["fn"], detail["blocks"]
+            if blocks is True:
+                logger.debug("Opening local copy of %s", path)
+
+                # In order to support downstream filesystems to be able to
+                # infer the compression from the original filename, like
+                # the `TarFileSystem`, let's extend the `io.BufferedReader`
+                # fileobject protocol by adding a dedicated attribute
+                # `original`.
+                f = open(fn, mode)
+                f.original = detail.get("original")
+                return f
+            else:
+                raise ValueError(
+                    f"Attempt to open partially cached file {path}"
+                    f" as a wholly cached file"
+                )
+        else:
+            fn = self._make_local_details(path)
+        kwargs["mode"] = mode
+
+        # call target filesystems open
+        self._mkcache()
+        if self.compression:
+            with self.fs._open(path, **kwargs) as f, open(fn, "wb") as f2:
+                if isinstance(f, AbstractBufferedFile):
+                    # want no type of caching if just downloading whole thing
+                    f.cache = BaseCache(0, f.cache.fetcher, f.size)
+                comp = (
+                    infer_compression(path)
+                    if self.compression == "infer"
+                    else self.compression
+                )
+                f = compr[comp](f, mode="rb")
+                data = True
+                while data:
+                    block = getattr(f, "blocksize", 5 * 2**20)
+                    data = f.read(block)
+                    f2.write(data)
+        else:
+            self.fs.get_file(path, fn)
+        self.save_cache()
+        return self._open(path, mode)
+
+
+class SimpleCacheFileSystem(WholeFileCacheFileSystem):
+    """Caches whole remote files on first access
+
+    This class is intended as a layer over any other file system, and
+    will make a local copy of each file accessed, so that all subsequent
+    reads are local. This implementation only copies whole files, and
+    does not keep any metadata about the download time or file details.
+    It is therefore safer to use in multi-threaded/concurrent situations.
+
+    This is the only of the caching filesystems that supports write: you will
+    be given a real local open file, and upon close and commit, it will be
+    uploaded to the target filesystem; the writability or the target URL is
+    not checked until that time.
+
+    """
+
+    protocol = "simplecache"
+    local_file = True
+    transaction_type = WriteCachedTransaction
+
+    def __init__(self, **kwargs):
+        kw = kwargs.copy()
+        for key in ["cache_check", "expiry_time", "check_files"]:
+            kw[key] = False
+        super().__init__(**kw)
+        for storage in self.storage:
+            if not os.path.exists(storage):
+                os.makedirs(storage, exist_ok=True)
+
+    def _check_file(self, path):
+        self._check_cache()
+        sha = self._mapper(path)
+        for storage in self.storage:
+            fn = os.path.join(storage, sha)
+            if os.path.exists(fn):
+                return fn
+
+    def save_cache(self):
+        pass
+
+    def load_cache(self):
+        pass
+
+    def pipe_file(self, path, value=None, **kwargs):
+        if self._intrans:
+            with self.open(path, "wb") as f:
+                f.write(value)
+        else:
+            super().pipe_file(path, value)
+
+    def ls(self, path, detail=True, **kwargs):
+        path = self._strip_protocol(path)
+        details = []
+        try:
+            details = self.fs.ls(
+                path, detail=True, **kwargs
+            ).copy()  # don't edit original!
+        except FileNotFoundError as e:
+            ex = e
+        else:
+            ex = None
+        if self._intrans:
+            path1 = path.rstrip("/") + "/"
+            for f in self.transaction.files:
+                if f.path == path:
+                    details.append(
+                        {"name": path, "size": f.size or f.tell(), "type": "file"}
+                    )
+                elif f.path.startswith(path1):
+                    if f.path.count("/") == path1.count("/"):
+                        details.append(
+                            {"name": f.path, "size": f.size or f.tell(), "type": "file"}
+                        )
+                    else:
+                        dname = "/".join(f.path.split("/")[: path1.count("/") + 1])
+                        details.append({"name": dname, "size": 0, "type": "directory"})
+        if ex is not None and not details:
+            raise ex
+        if detail:
+            return details
+        return sorted(_["name"] for _ in details)
+
+    def info(self, path, **kwargs):
+        path = self._strip_protocol(path)
+        if self._intrans:
+            f = [_ for _ in self.transaction.files if _.path == path]
+            if f:
+                size = os.path.getsize(f[0].fn) if f[0].closed else f[0].tell()
+                return {"name": path, "size": size, "type": "file"}
+            f = any(_.path.startswith(path + "/") for _ in self.transaction.files)
+            if f:
+                return {"name": path, "size": 0, "type": "directory"}
+        return self.fs.info(path, **kwargs)
+
+    def pipe(self, path, value=None, **kwargs):
+        if isinstance(path, str):
+            self.pipe_file(self._strip_protocol(path), value, **kwargs)
+        elif isinstance(path, dict):
+            for k, v in path.items():
+                self.pipe_file(self._strip_protocol(k), v, **kwargs)
+        else:
+            raise ValueError("path must be str or dict")
+
+    async def _cat_file(self, path, start=None, end=None, **kwargs):
+        logger.debug("async cat_file %s", path)
+        path = self._strip_protocol(path)
+        sha = self._mapper(path)
+        fn = self._check_file(path)
+
+        if not fn:
+            fn = os.path.join(self.storage[-1], sha)
+            await self.fs._get_file(path, fn, **kwargs)
+
+        with open(fn, "rb") as f:  # noqa ASYNC230
+            if start:
+                f.seek(start)
+            size = -1 if end is None else end - f.tell()
+            return f.read(size)
+
+    async def _cat_ranges(
+        self, paths, starts, ends, max_gap=None, on_error="return", **kwargs
+    ):
+        logger.debug("async cat ranges %s", paths)
+        lpaths = []
+        rset = set()
+        download = []
+        rpaths = []
+        for p in paths:
+            fn = self._check_file(p)
+            if fn is None and p not in rset:
+                sha = self._mapper(p)
+                fn = os.path.join(self.storage[-1], sha)
+                download.append(fn)
+                rset.add(p)
+                rpaths.append(p)
+            lpaths.append(fn)
+        if download:
+            await self.fs._get(rpaths, download, on_error=on_error)
+
+        return LocalFileSystem().cat_ranges(
+            lpaths, starts, ends, max_gap=max_gap, on_error=on_error, **kwargs
+        )
+
+    def cat_ranges(
+        self, paths, starts, ends, max_gap=None, on_error="return", **kwargs
+    ):
+        logger.debug("cat ranges %s", paths)
+        lpaths = [self._check_file(p) for p in paths]
+        rpaths = [p for l, p in zip(lpaths, paths) if l is False]
+        lpaths = [l for l, p in zip(lpaths, paths) if l is False]
+        self.fs.get(rpaths, lpaths)
+        paths = [self._check_file(p) for p in paths]
+        return LocalFileSystem().cat_ranges(
+            paths, starts, ends, max_gap=max_gap, on_error=on_error, **kwargs
+        )
+
+    def _open(self, path, mode="rb", **kwargs):
+        path = self._strip_protocol(path)
+        sha = self._mapper(path)
+
+        if "r" not in mode:
+            fn = os.path.join(self.storage[-1], sha)
+            user_specified_kwargs = {
+                k: v
+                for k, v in kwargs.items()
+                if k not in ["autocommit", "block_size", "cache_options"]
+            }  # those were added by open()
+            return LocalTempFile(
+                self,
+                path,
+                mode=mode,
+                autocommit=not self._intrans,
+                fn=fn,
+                **user_specified_kwargs,
+            )
+        fn = self._check_file(path)
+        if fn:
+            return open(fn, mode)
+
+        fn = os.path.join(self.storage[-1], sha)
+        logger.debug("Copying %s to local cache", path)
+        kwargs["mode"] = mode
+
+        self._mkcache()
+        self._cache_size = None
+        if self.compression:
+            with self.fs._open(path, **kwargs) as f, open(fn, "wb") as f2:
+                if isinstance(f, AbstractBufferedFile):
+                    # want no type of caching if just downloading whole thing
+                    f.cache = BaseCache(0, f.cache.fetcher, f.size)
+                comp = (
+                    infer_compression(path)
+                    if self.compression == "infer"
+                    else self.compression
+                )
+                f = compr[comp](f, mode="rb")
+                data = True
+                while data:
+                    block = getattr(f, "blocksize", 5 * 2**20)
+                    data = f.read(block)
+                    f2.write(data)
+        else:
+            self.fs.get_file(path, fn)
+        return self._open(path, mode)
+
+
+class LocalTempFile:
+    """A temporary local file, which will be uploaded on commit"""
+
+    def __init__(self, fs, path, fn, mode="wb", autocommit=True, seek=0, **kwargs):
+        self.fn = fn
+        self.fh = open(fn, mode)
+        self.mode = mode
+        if seek:
+            self.fh.seek(seek)
+        self.path = path
+        self.size = None
+        self.fs = fs
+        self.closed = False
+        self.autocommit = autocommit
+        self.kwargs = kwargs
+
+    def __reduce__(self):
+        # always open in r+b to allow continuing writing at a location
+        return (
+            LocalTempFile,
+            (self.fs, self.path, self.fn, "r+b", self.autocommit, self.tell()),
+        )
+
+    def __enter__(self):
+        return self.fh
+
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        self.close()
+
+    def close(self):
+        # self.size = self.fh.tell()
+        if self.closed:
+            return
+        self.fh.close()
+        self.closed = True
+        if self.autocommit:
+            self.commit()
+
+    def discard(self):
+        self.fh.close()
+        os.remove(self.fn)
+
+    def commit(self):
+        self.fs.put(self.fn, self.path, **self.kwargs)
+        # we do not delete the local copy, it's still in the cache.
+
+    @property
+    def name(self):
+        return self.fn
+
+    def __repr__(self) -> str:
+        return f"LocalTempFile: {self.path}"
+
+    def __getattr__(self, item):
+        return getattr(self.fh, item)

pythonProject/.venv/Lib/site-packages/fsspec/implementations/dask.py

@@ -0,0 +1,152 @@
+import dask
+from distributed.client import Client, _get_global_client
+from distributed.worker import Worker
+
+from fsspec import filesystem
+from fsspec.spec import AbstractBufferedFile, AbstractFileSystem
+from fsspec.utils import infer_storage_options
+
+
+def _get_client(client):
+    if client is None:
+        return _get_global_client()
+    elif isinstance(client, Client):
+        return client
+    else:
+        # e.g., connection string
+        return Client(client)
+
+
+def _in_worker():
+    return bool(Worker._instances)
+
+
+class DaskWorkerFileSystem(AbstractFileSystem):
+    """View files accessible to a worker as any other remote file-system
+
+    When instances are run on the worker, uses the real filesystem. When
+    run on the client, they call the worker to provide information or data.
+
+    **Warning** this implementation is experimental, and read-only for now.
+    """
+
+    def __init__(
+        self, target_protocol=None, target_options=None, fs=None, client=None, **kwargs
+    ):
+        super().__init__(**kwargs)
+        if not (fs is None) ^ (target_protocol is None):
+            raise ValueError(
+                "Please provide one of filesystem instance (fs) or"
+                " target_protocol, not both"
+            )
+        self.target_protocol = target_protocol
+        self.target_options = target_options
+        self.worker = None
+        self.client = client
+        self.fs = fs
+        self._determine_worker()
+
+    @staticmethod
+    def _get_kwargs_from_urls(path):
+        so = infer_storage_options(path)
+        if "host" in so and "port" in so:
+            return {"client": f"{so['host']}:{so['port']}"}
+        else:
+            return {}
+
+    def _determine_worker(self):
+        if _in_worker():
+            self.worker = True
+            if self.fs is None:
+                self.fs = filesystem(
+                    self.target_protocol, **(self.target_options or {})
+                )
+        else:
+            self.worker = False
+            self.client = _get_client(self.client)
+            self.rfs = dask.delayed(self)
+
+    def mkdir(self, *args, **kwargs):
+        if self.worker:
+            self.fs.mkdir(*args, **kwargs)
+        else:
+            self.rfs.mkdir(*args, **kwargs).compute()
+
+    def rm(self, *args, **kwargs):
+        if self.worker:
+            self.fs.rm(*args, **kwargs)
+        else:
+            self.rfs.rm(*args, **kwargs).compute()
+
+    def copy(self, *args, **kwargs):
+        if self.worker:
+            self.fs.copy(*args, **kwargs)
+        else:
+            self.rfs.copy(*args, **kwargs).compute()
+
+    def mv(self, *args, **kwargs):
+        if self.worker:
+            self.fs.mv(*args, **kwargs)
+        else:
+            self.rfs.mv(*args, **kwargs).compute()
+
+    def ls(self, *args, **kwargs):
+        if self.worker:
+            return self.fs.ls(*args, **kwargs)
+        else:
+            return self.rfs.ls(*args, **kwargs).compute()
+
+    def _open(
+        self,
+        path,
+        mode="rb",
+        block_size=None,
+        autocommit=True,
+        cache_options=None,
+        **kwargs,
+    ):
+        if self.worker:
+            return self.fs._open(
+                path,
+                mode=mode,
+                block_size=block_size,
+                autocommit=autocommit,
+                cache_options=cache_options,
+                **kwargs,
+            )
+        else:
+            return DaskFile(
+                fs=self,
+                path=path,
+                mode=mode,
+                block_size=block_size,
+                autocommit=autocommit,
+                cache_options=cache_options,
+                **kwargs,
+            )
+
+    def fetch_range(self, path, mode, start, end):
+        if self.worker:
+            with self._open(path, mode) as f:
+                f.seek(start)
+                return f.read(end - start)
+        else:
+            return self.rfs.fetch_range(path, mode, start, end).compute()
+
+
+class DaskFile(AbstractBufferedFile):
+    def __init__(self, mode="rb", **kwargs):
+        if mode != "rb":
+            raise ValueError('Remote dask files can only be opened in "rb" mode')
+        super().__init__(**kwargs)
+
+    def _upload_chunk(self, final=False):
+        pass
+
+    def _initiate_upload(self):
+        """Create remote file/upload"""
+        pass
+
+    def _fetch_range(self, start, end):
+        """Get the specified set of bytes from remote"""
+        return self.fs.fetch_range(self.path, self.mode, start, end)

pythonProject/.venv/Lib/site-packages/fsspec/implementations/data.py

@@ -0,0 +1,58 @@
+import base64
+import io
+from typing import Optional
+from urllib.parse import unquote
+
+from fsspec import AbstractFileSystem
+
+
+class DataFileSystem(AbstractFileSystem):
+    """A handy decoder for data-URLs
+
+    Example
+    -------
+    >>> with fsspec.open("data:,Hello%2C%20World%21") as f:
+    ...     print(f.read())
+    b"Hello, World!"
+
+    See https://developer.mozilla.org/en-US/docs/Web/HTTP/Basics_of_HTTP/Data_URLs
+    """
+
+    protocol = "data"
+
+    def __init__(self, **kwargs):
+        """No parameters for this filesystem"""
+        super().__init__(**kwargs)
+
+    def cat_file(self, path, start=None, end=None, **kwargs):
+        pref, data = path.split(",", 1)
+        if pref.endswith("base64"):
+            return base64.b64decode(data)[start:end]
+        return unquote(data).encode()[start:end]
+
+    def info(self, path, **kwargs):
+        pref, name = path.split(",", 1)
+        data = self.cat_file(path)
+        mime = pref.split(":", 1)[1].split(";", 1)[0]
+        return {"name": name, "size": len(data), "type": "file", "mimetype": mime}
+
+    def _open(
+        self,
+        path,
+        mode="rb",
+        block_size=None,
+        autocommit=True,
+        cache_options=None,
+        **kwargs,
+    ):
+        if "r" not in mode:
+            raise ValueError("Read only filesystem")
+        return io.BytesIO(self.cat_file(path))
+
+    @staticmethod
+    def encode(data: bytes, mime: Optional[str] = None):
+        """Format the given data into data-URL syntax
+
+        This version always base64 encodes, even when the data is ascii/url-safe.
+        """
+        return f"data:{mime or ''};base64,{base64.b64encode(data).decode()}"

pythonProject/.venv/Lib/site-packages/fsspec/implementations/dbfs.py

@@ -0,0 +1,496 @@
+from __future__ import annotations
+
+import base64
+import urllib
+
+import requests
+from requests.adapters import HTTPAdapter, Retry
+from typing_extensions import override
+
+from fsspec import AbstractFileSystem
+from fsspec.spec import AbstractBufferedFile
+
+
+class DatabricksException(Exception):
+    """
+    Helper class for exceptions raised in this module.
+    """
+
+    def __init__(self, error_code, message, details=None):
+        """Create a new DatabricksException"""
+        super().__init__(message)
+
+        self.error_code = error_code
+        self.message = message
+        self.details = details
+
+
+class DatabricksFileSystem(AbstractFileSystem):
+    """
+    Get access to the Databricks filesystem implementation over HTTP.
+    Can be used inside and outside of a databricks cluster.
+    """
+
+    def __init__(self, instance, token, **kwargs):
+        """
+        Create a new DatabricksFileSystem.
+
+        Parameters
+        ----------
+        instance: str
+            The instance URL of the databricks cluster.
+            For example for an Azure databricks cluster, this
+            has the form adb-<some-number>.<two digits>.azuredatabricks.net.
+        token: str
+            Your personal token. Find out more
+            here: https://docs.databricks.com/dev-tools/api/latest/authentication.html
+        """
+        self.instance = instance
+        self.token = token
+        self.session = requests.Session()
+        self.retries = Retry(
+            total=10,
+            backoff_factor=0.05,
+            status_forcelist=[408, 429, 500, 502, 503, 504],
+        )
+
+        self.session.mount("https://", HTTPAdapter(max_retries=self.retries))
+        self.session.headers.update({"Authorization": f"Bearer {self.token}"})
+
+        super().__init__(**kwargs)
+
+    @override
+    def _ls_from_cache(self, path) -> list[dict[str, str | int]] | None:
+        """Check cache for listing
+
+        Returns listing, if found (may be empty list for a directory that
+        exists but contains nothing), None if not in cache.
+        """
+        self.dircache.pop(path.rstrip("/"), None)
+
+        parent = self._parent(path)
+        if parent in self.dircache:
+            for entry in self.dircache[parent]:
+                if entry["name"] == path.rstrip("/"):
+                    if entry["type"] != "directory":
+                        return [entry]
+                    return []
+            raise FileNotFoundError(path)
+
+    def ls(self, path, detail=True, **kwargs):
+        """
+        List the contents of the given path.
+
+        Parameters
+        ----------
+        path: str
+            Absolute path
+        detail: bool
+            Return not only the list of filenames,
+            but also additional information on file sizes
+            and types.
+        """
+        try:
+            out = self._ls_from_cache(path)
+        except FileNotFoundError:
+            # This happens if the `path`'s parent was cached, but `path` is not
+            # there. This suggests that `path` is new since the parent was
+            # cached. Attempt to invalidate parent's cache before continuing.
+            self.dircache.pop(self._parent(path), None)
+            out = None
+
+        if not out:
+            try:
+                r = self._send_to_api(
+                    method="get", endpoint="list", json={"path": path}
+                )
+            except DatabricksException as e:
+                if e.error_code == "RESOURCE_DOES_NOT_EXIST":
+                    raise FileNotFoundError(e.message) from e
+
+                raise
+            files = r.get("files", [])
+            out = [
+                {
+                    "name": o["path"],
+                    "type": "directory" if o["is_dir"] else "file",
+                    "size": o["file_size"],
+                }
+                for o in files
+            ]
+            self.dircache[path] = out
+
+        if detail:
+            return out
+        return [o["name"] for o in out]
+
+    def makedirs(self, path, exist_ok=True):
+        """
+        Create a given absolute path and all of its parents.
+
+        Parameters
+        ----------
+        path: str
+            Absolute path to create
+        exist_ok: bool
+            If false, checks if the folder
+            exists before creating it (and raises an
+            Exception if this is the case)
+        """
+        if not exist_ok:
+            try:
+                # If the following succeeds, the path is already present
+                self._send_to_api(
+                    method="get", endpoint="get-status", json={"path": path}
+                )
+                raise FileExistsError(f"Path {path} already exists")
+            except DatabricksException as e:
+                if e.error_code == "RESOURCE_DOES_NOT_EXIST":
+                    pass
+
+        try:
+            self._send_to_api(method="post", endpoint="mkdirs", json={"path": path})
+        except DatabricksException as e:
+            if e.error_code == "RESOURCE_ALREADY_EXISTS":
+                raise FileExistsError(e.message) from e
+
+            raise
+        self.invalidate_cache(self._parent(path))
+
+    def mkdir(self, path, create_parents=True, **kwargs):
+        """
+        Create a given absolute path and all of its parents.
+
+        Parameters
+        ----------
+        path: str
+            Absolute path to create
+        create_parents: bool
+            Whether to create all parents or not.
+            "False" is not implemented so far.
+        """
+        if not create_parents:
+            raise NotImplementedError
+
+        self.mkdirs(path, **kwargs)
+
+    def rm(self, path, recursive=False, **kwargs):
+        """
+        Remove the file or folder at the given absolute path.
+
+        Parameters
+        ----------
+        path: str
+            Absolute path what to remove
+        recursive: bool
+            Recursively delete all files in a folder.
+        """
+        try:
+            self._send_to_api(
+                method="post",
+                endpoint="delete",
+                json={"path": path, "recursive": recursive},
+            )
+        except DatabricksException as e:
+            # This is not really an exception, it just means
+            # not everything was deleted so far
+            if e.error_code == "PARTIAL_DELETE":
+                self.rm(path=path, recursive=recursive)
+            elif e.error_code == "IO_ERROR":
+                # Using the same exception as the os module would use here
+                raise OSError(e.message) from e
+
+            raise
+        self.invalidate_cache(self._parent(path))
+
+    def mv(
+        self, source_path, destination_path, recursive=False, maxdepth=None, **kwargs
+    ):
+        """
+        Move a source to a destination path.
+
+        A note from the original [databricks API manual]
+        (https://docs.databricks.com/dev-tools/api/latest/dbfs.html#move).
+
+        When moving a large number of files the API call will time out after
+        approximately 60s, potentially resulting in partially moved data.
+        Therefore, for operations that move more than 10k files, we strongly
+        discourage using the DBFS REST API.
+
+        Parameters
+        ----------
+        source_path: str
+            From where to move (absolute path)
+        destination_path: str
+            To where to move (absolute path)
+        recursive: bool
+            Not implemented to far.
+        maxdepth:
+            Not implemented to far.
+        """
+        if recursive:
+            raise NotImplementedError
+        if maxdepth:
+            raise NotImplementedError
+
+        try:
+            self._send_to_api(
+                method="post",
+                endpoint="move",
+                json={"source_path": source_path, "destination_path": destination_path},
+            )
+        except DatabricksException as e:
+            if e.error_code == "RESOURCE_DOES_NOT_EXIST":
+                raise FileNotFoundError(e.message) from e
+            elif e.error_code == "RESOURCE_ALREADY_EXISTS":
+                raise FileExistsError(e.message) from e
+
+            raise
+        self.invalidate_cache(self._parent(source_path))
+        self.invalidate_cache(self._parent(destination_path))
+
+    def _open(self, path, mode="rb", block_size="default", **kwargs):
+        """
+        Overwrite the base class method to make sure to create a DBFile.
+        All arguments are copied from the base method.
+
+        Only the default blocksize is allowed.
+        """
+        return DatabricksFile(self, path, mode=mode, block_size=block_size, **kwargs)
+
+    def _send_to_api(self, method, endpoint, json):
+        """
+        Send the given json to the DBFS API
+        using a get or post request (specified by the argument `method`).
+
+        Parameters
+        ----------
+        method: str
+            Which http method to use for communication; "get" or "post".
+        endpoint: str
+            Where to send the request to (last part of the API URL)
+        json: dict
+            Dictionary of information to send
+        """
+        if method == "post":
+            session_call = self.session.post
+        elif method == "get":
+            session_call = self.session.get
+        else:
+            raise ValueError(f"Do not understand method {method}")
+
+        url = urllib.parse.urljoin(f"https://{self.instance}/api/2.0/dbfs/", endpoint)
+
+        r = session_call(url, json=json)
+
+        # The DBFS API will return a json, also in case of an exception.
+        # We want to preserve this information as good as possible.
+        try:
+            r.raise_for_status()
+        except requests.HTTPError as e:
+            # try to extract json error message
+            # if that fails, fall back to the original exception
+            try:
+                exception_json = e.response.json()
+            except Exception:
+                raise e from None
+
+            raise DatabricksException(**exception_json) from e
+
+        return r.json()
+
+    def _create_handle(self, path, overwrite=True):
+        """
+        Internal function to create a handle, which can be used to
+        write blocks of a file to DBFS.
+        A handle has a unique identifier which needs to be passed
+        whenever written during this transaction.
+        The handle is active for 10 minutes - after that a new
+        write transaction needs to be created.
+        Make sure to close the handle after you are finished.
+
+        Parameters
+        ----------
+        path: str
+            Absolute path for this file.
+        overwrite: bool
+            If a file already exist at this location, either overwrite
+            it or raise an exception.
+        """
+        try:
+            r = self._send_to_api(
+                method="post",
+                endpoint="create",
+                json={"path": path, "overwrite": overwrite},
+            )
+            return r["handle"]
+        except DatabricksException as e:
+            if e.error_code == "RESOURCE_ALREADY_EXISTS":
+                raise FileExistsError(e.message) from e
+
+            raise
+
+    def _close_handle(self, handle):
+        """
+        Close a handle, which was opened by :func:`_create_handle`.
+
+        Parameters
+        ----------
+        handle: str
+            Which handle to close.
+        """
+        try:
+            self._send_to_api(method="post", endpoint="close", json={"handle": handle})
+        except DatabricksException as e:
+            if e.error_code == "RESOURCE_DOES_NOT_EXIST":
+                raise FileNotFoundError(e.message) from e
+
+            raise
+
+    def _add_data(self, handle, data):
+        """
+        Upload data to an already opened file handle
+        (opened by :func:`_create_handle`).
+        The maximal allowed data size is 1MB after
+        conversion to base64.
+        Remember to close the handle when you are finished.
+
+        Parameters
+        ----------
+        handle: str
+            Which handle to upload data to.
+        data: bytes
+            Block of data to add to the handle.
+        """
+        data = base64.b64encode(data).decode()
+        try:
+            self._send_to_api(
+                method="post",
+                endpoint="add-block",
+                json={"handle": handle, "data": data},
+            )
+        except DatabricksException as e:
+            if e.error_code == "RESOURCE_DOES_NOT_EXIST":
+                raise FileNotFoundError(e.message) from e
+            elif e.error_code == "MAX_BLOCK_SIZE_EXCEEDED":
+                raise ValueError(e.message) from e
+
+            raise
+
+    def _get_data(self, path, start, end):
+        """
+        Download data in bytes from a given absolute path in a block
+        from [start, start+length].
+        The maximum number of allowed bytes to read is 1MB.
+
+        Parameters
+        ----------
+        path: str
+            Absolute path to download data from
+        start: int
+            Start position of the block
+        end: int
+            End position of the block
+        """
+        try:
+            r = self._send_to_api(
+                method="get",
+                endpoint="read",
+                json={"path": path, "offset": start, "length": end - start},
+            )
+            return base64.b64decode(r["data"])
+        except DatabricksException as e:
+            if e.error_code == "RESOURCE_DOES_NOT_EXIST":
+                raise FileNotFoundError(e.message) from e
+            elif e.error_code in ["INVALID_PARAMETER_VALUE", "MAX_READ_SIZE_EXCEEDED"]:
+                raise ValueError(e.message) from e
+
+            raise
+
+    def invalidate_cache(self, path=None):
+        if path is None:
+            self.dircache.clear()
+        else:
+            self.dircache.pop(path, None)
+        super().invalidate_cache(path)
+
+
+class DatabricksFile(AbstractBufferedFile):
+    """
+    Helper class for files referenced in the DatabricksFileSystem.
+    """
+
+    DEFAULT_BLOCK_SIZE = 1 * 2**20  # only allowed block size
+
+    def __init__(
+        self,
+        fs,
+        path,
+        mode="rb",
+        block_size="default",
+        autocommit=True,
+        cache_type="readahead",
+        cache_options=None,
+        **kwargs,
+    ):
+        """
+        Create a new instance of the DatabricksFile.
+
+        The blocksize needs to be the default one.
+        """
+        if block_size is None or block_size == "default":
+            block_size = self.DEFAULT_BLOCK_SIZE
+
+        assert block_size == self.DEFAULT_BLOCK_SIZE, (
+            f"Only the default block size is allowed, not {block_size}"
+        )
+
+        super().__init__(
+            fs,
+            path,
+            mode=mode,
+            block_size=block_size,
+            autocommit=autocommit,
+            cache_type=cache_type,
+            cache_options=cache_options or {},
+            **kwargs,
+        )
+
+    def _initiate_upload(self):
+        """Internal function to start a file upload"""
+        self.handle = self.fs._create_handle(self.path)
+
+    def _upload_chunk(self, final=False):
+        """Internal function to add a chunk of data to a started upload"""
+        self.buffer.seek(0)
+        data = self.buffer.getvalue()
+
+        data_chunks = [
+            data[start:end] for start, end in self._to_sized_blocks(len(data))
+        ]
+
+        for data_chunk in data_chunks:
+            self.fs._add_data(handle=self.handle, data=data_chunk)
+
+        if final:
+            self.fs._close_handle(handle=self.handle)
+            return True
+
+    def _fetch_range(self, start, end):
+        """Internal function to download a block of data"""
+        return_buffer = b""
+        length = end - start
+        for chunk_start, chunk_end in self._to_sized_blocks(length, start):
+            return_buffer += self.fs._get_data(
+                path=self.path, start=chunk_start, end=chunk_end
+            )
+
+        return return_buffer
+
+    def _to_sized_blocks(self, length, start=0):
+        """Helper function to split a range from 0 to total_length into blocksizes"""
+        end = start + length
+        for data_chunk in range(start, end, self.blocksize):
+            data_start = data_chunk
+            data_end = min(end, data_chunk + self.blocksize)
+            yield data_start, data_end

pythonProject/.venv/Lib/site-packages/fsspec/implementations/dirfs.py

@@ -0,0 +1,388 @@
+from .. import filesystem
+from ..asyn import AsyncFileSystem
+
+
+class DirFileSystem(AsyncFileSystem):
+    """Directory prefix filesystem
+
+    The DirFileSystem is a filesystem-wrapper. It assumes every path it is dealing with
+    is relative to the `path`. After performing the necessary paths operation it
+    delegates everything to the wrapped filesystem.
+    """
+
+    protocol = "dir"
+
+    def __init__(
+        self,
+        path=None,
+        fs=None,
+        fo=None,
+        target_protocol=None,
+        target_options=None,
+        **storage_options,
+    ):
+        """
+        Parameters
+        ----------
+        path: str
+            Path to the directory.
+        fs: AbstractFileSystem
+            An instantiated filesystem to wrap.
+        target_protocol, target_options:
+            if fs is none, construct it from these
+        fo: str
+            Alternate for path; do not provide both
+        """
+        super().__init__(**storage_options)
+        if fs is None:
+            fs = filesystem(protocol=target_protocol, **(target_options or {}))
+        path = path or fo
+
+        if self.asynchronous and not fs.async_impl:
+            raise ValueError("can't use asynchronous with non-async fs")
+
+        if fs.async_impl and self.asynchronous != fs.asynchronous:
+            raise ValueError("both dirfs and fs should be in the same sync/async mode")
+
+        self.path = fs._strip_protocol(path)
+        self.fs = fs
+
+    def _join(self, path):
+        if isinstance(path, str):
+            if not self.path:
+                return path
+            if not path:
+                return self.path
+            return self.fs.sep.join((self.path, self._strip_protocol(path)))
+        if isinstance(path, dict):
+            return {self._join(_path): value for _path, value in path.items()}
+        return [self._join(_path) for _path in path]
+
+    def _relpath(self, path):
+        if isinstance(path, str):
+            if not self.path:
+                return path
+            # We need to account for S3FileSystem returning paths that do not
+            # start with a '/'
+            if path == self.path or (
+                self.path.startswith(self.fs.sep) and path == self.path[1:]
+            ):
+                return ""
+            prefix = self.path + self.fs.sep
+            if self.path.startswith(self.fs.sep) and not path.startswith(self.fs.sep):
+                prefix = prefix[1:]
+            assert path.startswith(prefix)
+            return path[len(prefix) :]
+        return [self._relpath(_path) for _path in path]
+
+    # Wrappers below
+
+    @property
+    def sep(self):
+        return self.fs.sep
+
+    async def set_session(self, *args, **kwargs):
+        return await self.fs.set_session(*args, **kwargs)
+
+    async def _rm_file(self, path, **kwargs):
+        return await self.fs._rm_file(self._join(path), **kwargs)
+
+    def rm_file(self, path, **kwargs):
+        return self.fs.rm_file(self._join(path), **kwargs)
+
+    async def _rm(self, path, *args, **kwargs):
+        return await self.fs._rm(self._join(path), *args, **kwargs)
+
+    def rm(self, path, *args, **kwargs):
+        return self.fs.rm(self._join(path), *args, **kwargs)
+
+    async def _cp_file(self, path1, path2, **kwargs):
+        return await self.fs._cp_file(self._join(path1), self._join(path2), **kwargs)
+
+    def cp_file(self, path1, path2, **kwargs):
+        return self.fs.cp_file(self._join(path1), self._join(path2), **kwargs)
+
+    async def _copy(
+        self,
+        path1,
+        path2,
+        *args,
+        **kwargs,
+    ):
+        return await self.fs._copy(
+            self._join(path1),
+            self._join(path2),
+            *args,
+            **kwargs,
+        )
+
+    def copy(self, path1, path2, *args, **kwargs):
+        return self.fs.copy(
+            self._join(path1),
+            self._join(path2),
+            *args,
+            **kwargs,
+        )
+
+    async def _pipe(self, path, *args, **kwargs):
+        return await self.fs._pipe(self._join(path), *args, **kwargs)
+
+    def pipe(self, path, *args, **kwargs):
+        return self.fs.pipe(self._join(path), *args, **kwargs)
+
+    async def _pipe_file(self, path, *args, **kwargs):
+        return await self.fs._pipe_file(self._join(path), *args, **kwargs)
+
+    def pipe_file(self, path, *args, **kwargs):
+        return self.fs.pipe_file(self._join(path), *args, **kwargs)
+
+    async def _cat_file(self, path, *args, **kwargs):
+        return await self.fs._cat_file(self._join(path), *args, **kwargs)
+
+    def cat_file(self, path, *args, **kwargs):
+        return self.fs.cat_file(self._join(path), *args, **kwargs)
+
+    async def _cat(self, path, *args, **kwargs):
+        ret = await self.fs._cat(
+            self._join(path),
+            *args,
+            **kwargs,
+        )
+
+        if isinstance(ret, dict):
+            return {self._relpath(key): value for key, value in ret.items()}
+
+        return ret
+
+    def cat(self, path, *args, **kwargs):
+        ret = self.fs.cat(
+            self._join(path),
+            *args,
+            **kwargs,
+        )
+
+        if isinstance(ret, dict):
+            return {self._relpath(key): value for key, value in ret.items()}
+
+        return ret
+
+    async def _put_file(self, lpath, rpath, **kwargs):
+        return await self.fs._put_file(lpath, self._join(rpath), **kwargs)
+
+    def put_file(self, lpath, rpath, **kwargs):
+        return self.fs.put_file(lpath, self._join(rpath), **kwargs)
+
+    async def _put(
+        self,
+        lpath,
+        rpath,
+        *args,
+        **kwargs,
+    ):
+        return await self.fs._put(
+            lpath,
+            self._join(rpath),
+            *args,
+            **kwargs,
+        )
+
+    def put(self, lpath, rpath, *args, **kwargs):
+        return self.fs.put(
+            lpath,
+            self._join(rpath),
+            *args,
+            **kwargs,
+        )
+
+    async def _get_file(self, rpath, lpath, **kwargs):
+        return await self.fs._get_file(self._join(rpath), lpath, **kwargs)
+
+    def get_file(self, rpath, lpath, **kwargs):
+        return self.fs.get_file(self._join(rpath), lpath, **kwargs)
+
+    async def _get(self, rpath, *args, **kwargs):
+        return await self.fs._get(self._join(rpath), *args, **kwargs)
+
+    def get(self, rpath, *args, **kwargs):
+        return self.fs.get(self._join(rpath), *args, **kwargs)
+
+    async def _isfile(self, path):
+        return await self.fs._isfile(self._join(path))
+
+    def isfile(self, path):
+        return self.fs.isfile(self._join(path))
+
+    async def _isdir(self, path):
+        return await self.fs._isdir(self._join(path))
+
+    def isdir(self, path):
+        return self.fs.isdir(self._join(path))
+
+    async def _size(self, path):
+        return await self.fs._size(self._join(path))
+
+    def size(self, path):
+        return self.fs.size(self._join(path))
+
+    async def _exists(self, path):
+        return await self.fs._exists(self._join(path))
+
+    def exists(self, path):
+        return self.fs.exists(self._join(path))
+
+    async def _info(self, path, **kwargs):
+        info = await self.fs._info(self._join(path), **kwargs)
+        info = info.copy()
+        info["name"] = self._relpath(info["name"])
+        return info
+
+    def info(self, path, **kwargs):
+        info = self.fs.info(self._join(path), **kwargs)
+        info = info.copy()
+        info["name"] = self._relpath(info["name"])
+        return info
+
+    async def _ls(self, path, detail=True, **kwargs):
+        ret = (await self.fs._ls(self._join(path), detail=detail, **kwargs)).copy()
+        if detail:
+            out = []
+            for entry in ret:
+                entry = entry.copy()
+                entry["name"] = self._relpath(entry["name"])
+                out.append(entry)
+            return out
+
+        return self._relpath(ret)
+
+    def ls(self, path, detail=True, **kwargs):
+        ret = self.fs.ls(self._join(path), detail=detail, **kwargs).copy()
+        if detail:
+            out = []
+            for entry in ret:
+                entry = entry.copy()
+                entry["name"] = self._relpath(entry["name"])
+                out.append(entry)
+            return out
+
+        return self._relpath(ret)
+
+    async def _walk(self, path, *args, **kwargs):
+        async for root, dirs, files in self.fs._walk(self._join(path), *args, **kwargs):
+            yield self._relpath(root), dirs, files
+
+    def walk(self, path, *args, **kwargs):
+        for root, dirs, files in self.fs.walk(self._join(path), *args, **kwargs):
+            yield self._relpath(root), dirs, files
+
+    async def _glob(self, path, **kwargs):
+        detail = kwargs.get("detail", False)
+        ret = await self.fs._glob(self._join(path), **kwargs)
+        if detail:
+            return {self._relpath(path): info for path, info in ret.items()}
+        return self._relpath(ret)
+
+    def glob(self, path, **kwargs):
+        detail = kwargs.get("detail", False)
+        ret = self.fs.glob(self._join(path), **kwargs)
+        if detail:
+            return {self._relpath(path): info for path, info in ret.items()}
+        return self._relpath(ret)
+
+    async def _du(self, path, *args, **kwargs):
+        total = kwargs.get("total", True)
+        ret = await self.fs._du(self._join(path), *args, **kwargs)
+        if total:
+            return ret
+
+        return {self._relpath(path): size for path, size in ret.items()}
+
+    def du(self, path, *args, **kwargs):
+        total = kwargs.get("total", True)
+        ret = self.fs.du(self._join(path), *args, **kwargs)
+        if total:
+            return ret
+
+        return {self._relpath(path): size for path, size in ret.items()}
+
+    async def _find(self, path, *args, **kwargs):
+        detail = kwargs.get("detail", False)
+        ret = await self.fs._find(self._join(path), *args, **kwargs)
+        if detail:
+            return {self._relpath(path): info for path, info in ret.items()}
+        return self._relpath(ret)
+
+    def find(self, path, *args, **kwargs):
+        detail = kwargs.get("detail", False)
+        ret = self.fs.find(self._join(path), *args, **kwargs)
+        if detail:
+            return {self._relpath(path): info for path, info in ret.items()}
+        return self._relpath(ret)
+
+    async def _expand_path(self, path, *args, **kwargs):
+        return self._relpath(
+            await self.fs._expand_path(self._join(path), *args, **kwargs)
+        )
+
+    def expand_path(self, path, *args, **kwargs):
+        return self._relpath(self.fs.expand_path(self._join(path), *args, **kwargs))
+
+    async def _mkdir(self, path, *args, **kwargs):
+        return await self.fs._mkdir(self._join(path), *args, **kwargs)
+
+    def mkdir(self, path, *args, **kwargs):
+        return self.fs.mkdir(self._join(path), *args, **kwargs)
+
+    async def _makedirs(self, path, *args, **kwargs):
+        return await self.fs._makedirs(self._join(path), *args, **kwargs)
+
+    def makedirs(self, path, *args, **kwargs):
+        return self.fs.makedirs(self._join(path), *args, **kwargs)
+
+    def rmdir(self, path):
+        return self.fs.rmdir(self._join(path))
+
+    def mv(self, path1, path2, **kwargs):
+        return self.fs.mv(
+            self._join(path1),
+            self._join(path2),
+            **kwargs,
+        )
+
+    def touch(self, path, **kwargs):
+        return self.fs.touch(self._join(path), **kwargs)
+
+    def created(self, path):
+        return self.fs.created(self._join(path))
+
+    def modified(self, path):
+        return self.fs.modified(self._join(path))
+
+    def sign(self, path, *args, **kwargs):
+        return self.fs.sign(self._join(path), *args, **kwargs)
+
+    def __repr__(self):
+        return f"{self.__class__.__qualname__}(path='{self.path}', fs={self.fs})"
+
+    def open(
+        self,
+        path,
+        *args,
+        **kwargs,
+    ):
+        return self.fs.open(
+            self._join(path),
+            *args,
+            **kwargs,
+        )
+
+    async def open_async(
+        self,
+        path,
+        *args,
+        **kwargs,
+    ):
+        return await self.fs.open_async(
+            self._join(path),
+            *args,
+            **kwargs,
+        )

pythonProject/.venv/Lib/site-packages/fsspec/utils.py

@@ -0,0 +1,737 @@
+from __future__ import annotations
+
+import contextlib
+import logging
+import math
+import os
+import re
+import sys
+import tempfile
+from collections.abc import Iterable, Iterator, Sequence
+from functools import partial
+from hashlib import md5
+from importlib.metadata import version
+from typing import (
+    IO,
+    TYPE_CHECKING,
+    Any,
+    Callable,
+    TypeVar,
+)
+from urllib.parse import urlsplit
+
+if TYPE_CHECKING:
+    import pathlib
+
+    from typing_extensions import TypeGuard
+
+    from fsspec.spec import AbstractFileSystem
+
+
+DEFAULT_BLOCK_SIZE = 5 * 2**20
+
+T = TypeVar("T")
+
+
+def infer_storage_options(
+    urlpath: str, inherit_storage_options: dict[str, Any] | None = None
+) -> dict[str, Any]:
+    """Infer storage options from URL path and merge it with existing storage
+    options.
+
+    Parameters
+    ----------
+    urlpath: str or unicode
+        Either local absolute file path or URL (hdfs://namenode:8020/file.csv)
+    inherit_storage_options: dict (optional)
+        Its contents will get merged with the inferred information from the
+        given path
+
+    Returns
+    -------
+    Storage options dict.
+
+    Examples
+    --------
+    >>> infer_storage_options('/mnt/datasets/test.csv')  # doctest: +SKIP
+    {"protocol": "file", "path", "/mnt/datasets/test.csv"}
+    >>> infer_storage_options(
+    ...     'hdfs://username:pwd@node:123/mnt/datasets/test.csv?q=1',
+    ...     inherit_storage_options={'extra': 'value'},
+    ... )  # doctest: +SKIP
+    {"protocol": "hdfs", "username": "username", "password": "pwd",
+    "host": "node", "port": 123, "path": "/mnt/datasets/test.csv",
+    "url_query": "q=1", "extra": "value"}
+    """
+    # Handle Windows paths including disk name in this special case
+    if (
+        re.match(r"^[a-zA-Z]:[\\/]", urlpath)
+        or re.match(r"^[a-zA-Z0-9]+://", urlpath) is None
+    ):
+        return {"protocol": "file", "path": urlpath}
+
+    parsed_path = urlsplit(urlpath)
+    protocol = parsed_path.scheme or "file"
+    if parsed_path.fragment:
+        path = "#".join([parsed_path.path, parsed_path.fragment])
+    else:
+        path = parsed_path.path
+    if protocol == "file":
+        # Special case parsing file protocol URL on Windows according to:
+        # https://msdn.microsoft.com/en-us/library/jj710207.aspx
+        windows_path = re.match(r"^/([a-zA-Z])[:|]([\\/].*)$", path)
+        if windows_path:
+            drive, path = windows_path.groups()
+            path = f"{drive}:{path}"
+
+    if protocol in ["http", "https"]:
+        # for HTTP, we don't want to parse, as requests will anyway
+        return {"protocol": protocol, "path": urlpath}
+
+    options: dict[str, Any] = {"protocol": protocol, "path": path}
+
+    if parsed_path.netloc:
+        # Parse `hostname` from netloc manually because `parsed_path.hostname`
+        # lowercases the hostname which is not always desirable (e.g. in S3):
+        # https://github.com/dask/dask/issues/1417
+        options["host"] = parsed_path.netloc.rsplit("@", 1)[-1].rsplit(":", 1)[0]
+
+        if protocol in ("s3", "s3a", "gcs", "gs"):
+            options["path"] = options["host"] + options["path"]
+        else:
+            options["host"] = options["host"]
+        if parsed_path.port:
+            options["port"] = parsed_path.port
+        if parsed_path.username:
+            options["username"] = parsed_path.username
+        if parsed_path.password:
+            options["password"] = parsed_path.password
+
+    if parsed_path.query:
+        options["url_query"] = parsed_path.query
+    if parsed_path.fragment:
+        options["url_fragment"] = parsed_path.fragment
+
+    if inherit_storage_options:
+        update_storage_options(options, inherit_storage_options)
+
+    return options
+
+
+def update_storage_options(
+    options: dict[str, Any], inherited: dict[str, Any] | None = None
+) -> None:
+    if not inherited:
+        inherited = {}
+    collisions = set(options) & set(inherited)
+    if collisions:
+        for collision in collisions:
+            if options.get(collision) != inherited.get(collision):
+                raise KeyError(
+                    f"Collision between inferred and specified storage "
+                    f"option:\n{collision}"
+                )
+    options.update(inherited)
+
+
+# Compression extensions registered via fsspec.compression.register_compression
+compressions: dict[str, str] = {}
+
+
+def infer_compression(filename: str) -> str | None:
+    """Infer compression, if available, from filename.
+
+    Infer a named compression type, if registered and available, from filename
+    extension. This includes builtin (gz, bz2, zip) compressions, as well as
+    optional compressions. See fsspec.compression.register_compression.
+    """
+    extension = os.path.splitext(filename)[-1].strip(".").lower()
+    if extension in compressions:
+        return compressions[extension]
+    return None
+
+
+def build_name_function(max_int: float) -> Callable[[int], str]:
+    """Returns a function that receives a single integer
+    and returns it as a string padded by enough zero characters
+    to align with maximum possible integer
+
+    >>> name_f = build_name_function(57)
+
+    >>> name_f(7)
+    '07'
+    >>> name_f(31)
+    '31'
+    >>> build_name_function(1000)(42)
+    '0042'
+    >>> build_name_function(999)(42)
+    '042'
+    >>> build_name_function(0)(0)
+    '0'
+    """
+    # handle corner cases max_int is 0 or exact power of 10
+    max_int += 1e-8
+
+    pad_length = int(math.ceil(math.log10(max_int)))
+
+    def name_function(i: int) -> str:
+        return str(i).zfill(pad_length)
+
+    return name_function
+
+
+def seek_delimiter(file: IO[bytes], delimiter: bytes, blocksize: int) -> bool:
+    r"""Seek current file to file start, file end, or byte after delimiter seq.
+
+    Seeks file to next chunk delimiter, where chunks are defined on file start,
+    a delimiting sequence, and file end. Use file.tell() to see location afterwards.
+    Note that file start is a valid split, so must be at offset > 0 to seek for
+    delimiter.
+
+    Parameters
+    ----------
+    file: a file
+    delimiter: bytes
+        a delimiter like ``b'\n'`` or message sentinel, matching file .read() type
+    blocksize: int
+        Number of bytes to read from the file at once.
+
+
+    Returns
+    -------
+    Returns True if a delimiter was found, False if at file start or end.
+
+    """
+
+    if file.tell() == 0:
+        # beginning-of-file, return without seek
+        return False
+
+    # Interface is for binary IO, with delimiter as bytes, but initialize last
+    # with result of file.read to preserve compatibility with text IO.
+    last: bytes | None = None
+    while True:
+        current = file.read(blocksize)
+        if not current:
+            # end-of-file without delimiter
+            return False
+        full = last + current if last else current
+        try:
+            if delimiter in full:
+                i = full.index(delimiter)
+                file.seek(file.tell() - (len(full) - i) + len(delimiter))
+                return True
+            elif len(current) < blocksize:
+                # end-of-file without delimiter
+                return False
+        except (OSError, ValueError):
+            pass
+        last = full[-len(delimiter) :]
+
+
+def read_block(
+    f: IO[bytes],
+    offset: int,
+    length: int | None,
+    delimiter: bytes | None = None,
+    split_before: bool = False,
+) -> bytes:
+    """Read a block of bytes from a file
+
+    Parameters
+    ----------
+    f: File
+        Open file
+    offset: int
+        Byte offset to start read
+    length: int
+        Number of bytes to read, read through end of file if None
+    delimiter: bytes (optional)
+        Ensure reading starts and stops at delimiter bytestring
+    split_before: bool (optional)
+        Start/stop read *before* delimiter bytestring.
+
+
+    If using the ``delimiter=`` keyword argument we ensure that the read
+    starts and stops at delimiter boundaries that follow the locations
+    ``offset`` and ``offset + length``.  If ``offset`` is zero then we
+    start at zero, regardless of delimiter.  The bytestring returned WILL
+    include the terminating delimiter string.
+
+    Examples
+    --------
+
+    >>> from io import BytesIO  # doctest: +SKIP
+    >>> f = BytesIO(b'Alice, 100\\nBob, 200\\nCharlie, 300')  # doctest: +SKIP
+    >>> read_block(f, 0, 13)  # doctest: +SKIP
+    b'Alice, 100\\nBo'
+
+    >>> read_block(f, 0, 13, delimiter=b'\\n')  # doctest: +SKIP
+    b'Alice, 100\\nBob, 200\\n'
+
+    >>> read_block(f, 10, 10, delimiter=b'\\n')  # doctest: +SKIP
+    b'Bob, 200\\nCharlie, 300'
+    """
+    if delimiter:
+        f.seek(offset)
+        found_start_delim = seek_delimiter(f, delimiter, 2**16)
+        if length is None:
+            return f.read()
+        start = f.tell()
+        length -= start - offset
+
+        f.seek(start + length)
+        found_end_delim = seek_delimiter(f, delimiter, 2**16)
+        end = f.tell()
+
+        # Adjust split location to before delimiter if seek found the
+        # delimiter sequence, not start or end of file.
+        if found_start_delim and split_before:
+            start -= len(delimiter)
+
+        if found_end_delim and split_before:
+            end -= len(delimiter)
+
+        offset = start
+        length = end - start
+
+    f.seek(offset)
+
+    # TODO: allow length to be None and read to the end of the file?
+    assert length is not None
+    b = f.read(length)
+    return b
+
+
+def tokenize(*args: Any, **kwargs: Any) -> str:
+    """Deterministic token
+
+    (modified from dask.base)
+
+    >>> tokenize([1, 2, '3'])
+    '9d71491b50023b06fc76928e6eddb952'
+
+    >>> tokenize('Hello') == tokenize('Hello')
+    True
+    """
+    if kwargs:
+        args += (kwargs,)
+    try:
+        h = md5(str(args).encode())
+    except ValueError:
+        # FIPS systems: https://github.com/fsspec/filesystem_spec/issues/380
+        h = md5(str(args).encode(), usedforsecurity=False)
+    return h.hexdigest()
+
+
+def stringify_path(filepath: str | os.PathLike[str] | pathlib.Path) -> str:
+    """Attempt to convert a path-like object to a string.
+
+    Parameters
+    ----------
+    filepath: object to be converted
+
+    Returns
+    -------
+    filepath_str: maybe a string version of the object
+
+    Notes
+    -----
+    Objects supporting the fspath protocol are coerced according to its
+    __fspath__ method.
+
+    For backwards compatibility with older Python version, pathlib.Path
+    objects are specially coerced.
+
+    Any other object is passed through unchanged, which includes bytes,
+    strings, buffers, or anything else that's not even path-like.
+    """
+    if isinstance(filepath, str):
+        return filepath
+    elif hasattr(filepath, "__fspath__"):
+        return filepath.__fspath__()
+    elif hasattr(filepath, "path"):
+        return filepath.path
+    else:
+        return filepath  # type: ignore[return-value]
+
+
+def make_instance(
+    cls: Callable[..., T], args: Sequence[Any], kwargs: dict[str, Any]
+) -> T:
+    inst = cls(*args, **kwargs)
+    inst._determine_worker()  # type: ignore[attr-defined]
+    return inst
+
+
+def common_prefix(paths: Iterable[str]) -> str:
+    """For a list of paths, find the shortest prefix common to all"""
+    parts = [p.split("/") for p in paths]
+    lmax = min(len(p) for p in parts)
+    end = 0
+    for i in range(lmax):
+        end = all(p[i] == parts[0][i] for p in parts)
+        if not end:
+            break
+    i += end
+    return "/".join(parts[0][:i])
+
+
+def other_paths(
+    paths: list[str],
+    path2: str | list[str],
+    exists: bool = False,
+    flatten: bool = False,
+) -> list[str]:
+    """In bulk file operations, construct a new file tree from a list of files
+
+    Parameters
+    ----------
+    paths: list of str
+        The input file tree
+    path2: str or list of str
+        Root to construct the new list in. If this is already a list of str, we just
+        assert it has the right number of elements.
+    exists: bool (optional)
+        For a str destination, it is already exists (and is a dir), files should
+        end up inside.
+    flatten: bool (optional)
+        Whether to flatten the input directory tree structure so that the output files
+        are in the same directory.
+
+    Returns
+    -------
+    list of str
+    """
+
+    if isinstance(path2, str):
+        path2 = path2.rstrip("/")
+
+        if flatten:
+            path2 = ["/".join((path2, p.split("/")[-1])) for p in paths]
+        else:
+            cp = common_prefix(paths)
+            if exists:
+                cp = cp.rsplit("/", 1)[0]
+            if not cp and all(not s.startswith("/") for s in paths):
+                path2 = ["/".join([path2, p]) for p in paths]
+            else:
+                path2 = [p.replace(cp, path2, 1) for p in paths]
+    else:
+        assert len(paths) == len(path2)
+    return path2
+
+
+def is_exception(obj: Any) -> bool:
+    return isinstance(obj, BaseException)
+
+
+def isfilelike(f: Any) -> TypeGuard[IO[bytes]]:
+    return all(hasattr(f, attr) for attr in ["read", "close", "tell"])
+
+
+def get_protocol(url: str) -> str:
+    url = stringify_path(url)
+    parts = re.split(r"(\:\:|\://)", url, maxsplit=1)
+    if len(parts) > 1:
+        return parts[0]
+    return "file"
+
+
+def can_be_local(path: str) -> bool:
+    """Can the given URL be used with open_local?"""
+    from fsspec import get_filesystem_class
+
+    try:
+        return getattr(get_filesystem_class(get_protocol(path)), "local_file", False)
+    except (ValueError, ImportError):
+        # not in registry or import failed
+        return False
+
+
+def get_package_version_without_import(name: str) -> str | None:
+    """For given package name, try to find the version without importing it
+
+    Import and package.__version__ is still the backup here, so an import
+    *might* happen.
+
+    Returns either the version string, or None if the package
+    or the version was not readily  found.
+    """
+    if name in sys.modules:
+        mod = sys.modules[name]
+        if hasattr(mod, "__version__"):
+            return mod.__version__
+    try:
+        return version(name)
+    except:  # noqa: E722
+        pass
+    try:
+        import importlib
+
+        mod = importlib.import_module(name)
+        return mod.__version__
+    except (ImportError, AttributeError):
+        return None
+
+
+def setup_logging(
+    logger: logging.Logger | None = None,
+    logger_name: str | None = None,
+    level: str = "DEBUG",
+    clear: bool = True,
+) -> logging.Logger:
+    if logger is None and logger_name is None:
+        raise ValueError("Provide either logger object or logger name")
+    logger = logger or logging.getLogger(logger_name)
+    handle = logging.StreamHandler()
+    formatter = logging.Formatter(
+        "%(asctime)s - %(name)s - %(levelname)s - %(funcName)s -- %(message)s"
+    )
+    handle.setFormatter(formatter)
+    if clear:
+        logger.handlers.clear()
+    logger.addHandler(handle)
+    logger.setLevel(level)
+    return logger
+
+
+def _unstrip_protocol(name: str, fs: AbstractFileSystem) -> str:
+    return fs.unstrip_protocol(name)
+
+
+def mirror_from(
+    origin_name: str, methods: Iterable[str]
+) -> Callable[[type[T]], type[T]]:
+    """Mirror attributes and methods from the given
+    origin_name attribute of the instance to the
+    decorated class"""
+
+    def origin_getter(method: str, self: Any) -> Any:
+        origin = getattr(self, origin_name)
+        return getattr(origin, method)
+
+    def wrapper(cls: type[T]) -> type[T]:
+        for method in methods:
+            wrapped_method = partial(origin_getter, method)
+            setattr(cls, method, property(wrapped_method))
+        return cls
+
+    return wrapper
+
+
+@contextlib.contextmanager
+def nullcontext(obj: T) -> Iterator[T]:
+    yield obj
+
+
+def merge_offset_ranges(
+    paths: list[str],
+    starts: list[int] | int,
+    ends: list[int] | int,
+    max_gap: int = 0,
+    max_block: int | None = None,
+    sort: bool = True,
+) -> tuple[list[str], list[int], list[int]]:
+    """Merge adjacent byte-offset ranges when the inter-range
+    gap is <= `max_gap`, and when the merged byte range does not
+    exceed `max_block` (if specified). By default, this function
+    will re-order the input paths and byte ranges to ensure sorted
+    order. If the user can guarantee that the inputs are already
+    sorted, passing `sort=False` will skip the re-ordering.
+    """
+    # Check input
+    if not isinstance(paths, list):
+        raise TypeError
+    if not isinstance(starts, list):
+        starts = [starts] * len(paths)
+    if not isinstance(ends, list):
+        ends = [ends] * len(paths)
+    if len(starts) != len(paths) or len(ends) != len(paths):
+        raise ValueError
+
+    # Early Return
+    if len(starts) <= 1:
+        return paths, starts, ends
+
+    starts = [s or 0 for s in starts]
+    # Sort by paths and then ranges if `sort=True`
+    if sort:
+        paths, starts, ends = (
+            list(v)
+            for v in zip(
+                *sorted(
+                    zip(paths, starts, ends),
+                )
+            )
+        )
+
+    if paths:
+        # Loop through the coupled `paths`, `starts`, and
+        # `ends`, and merge adjacent blocks when appropriate
+        new_paths = paths[:1]
+        new_starts = starts[:1]
+        new_ends = ends[:1]
+        for i in range(1, len(paths)):
+            if paths[i] == paths[i - 1] and new_ends[-1] is None:
+                continue
+            elif (
+                paths[i] != paths[i - 1]
+                or ((starts[i] - new_ends[-1]) > max_gap)
+                or (max_block is not None and (ends[i] - new_starts[-1]) > max_block)
+            ):
+                # Cannot merge with previous block.
+                # Add new `paths`, `starts`, and `ends` elements
+                new_paths.append(paths[i])
+                new_starts.append(starts[i])
+                new_ends.append(ends[i])
+            else:
+                # Merge with previous block by updating the
+                # last element of `ends`
+                new_ends[-1] = ends[i]
+        return new_paths, new_starts, new_ends
+
+    # `paths` is empty. Just return input lists
+    return paths, starts, ends
+
+
+def file_size(filelike: IO[bytes]) -> int:
+    """Find length of any open read-mode file-like"""
+    pos = filelike.tell()
+    try:
+        return filelike.seek(0, 2)
+    finally:
+        filelike.seek(pos)
+
+
+@contextlib.contextmanager
+def atomic_write(path: str, mode: str = "wb"):
+    """
+    A context manager that opens a temporary file next to `path` and, on exit,
+    replaces `path` with the temporary file, thereby updating `path`
+    atomically.
+    """
+    fd, fn = tempfile.mkstemp(
+        dir=os.path.dirname(path), prefix=os.path.basename(path) + "-"
+    )
+    try:
+        with open(fd, mode) as fp:
+            yield fp
+    except BaseException:
+        with contextlib.suppress(FileNotFoundError):
+            os.unlink(fn)
+        raise
+    else:
+        os.replace(fn, path)
+
+
+def _translate(pat, STAR, QUESTION_MARK):
+    # Copied from: https://github.com/python/cpython/pull/106703.
+    res: list[str] = []
+    add = res.append
+    i, n = 0, len(pat)
+    while i < n:
+        c = pat[i]
+        i = i + 1
+        if c == "*":
+            # compress consecutive `*` into one
+            if (not res) or res[-1] is not STAR:
+                add(STAR)
+        elif c == "?":
+            add(QUESTION_MARK)
+        elif c == "[":
+            j = i
+            if j < n and pat[j] == "!":
+                j = j + 1
+            if j < n and pat[j] == "]":
+                j = j + 1
+            while j < n and pat[j] != "]":
+                j = j + 1
+            if j >= n:
+                add("\\[")
+            else:
+                stuff = pat[i:j]
+                if "-" not in stuff:
+                    stuff = stuff.replace("\\", r"\\")
+                else:
+                    chunks = []
+                    k = i + 2 if pat[i] == "!" else i + 1
+                    while True:
+                        k = pat.find("-", k, j)
+                        if k < 0:
+                            break
+                        chunks.append(pat[i:k])
+                        i = k + 1
+                        k = k + 3
+                    chunk = pat[i:j]
+                    if chunk:
+                        chunks.append(chunk)
+                    else:
+                        chunks[-1] += "-"
+                    # Remove empty ranges -- invalid in RE.
+                    for k in range(len(chunks) - 1, 0, -1):
+                        if chunks[k - 1][-1] > chunks[k][0]:
+                            chunks[k - 1] = chunks[k - 1][:-1] + chunks[k][1:]
+                            del chunks[k]
+                    # Escape backslashes and hyphens for set difference (--).
+                    # Hyphens that create ranges shouldn't be escaped.
+                    stuff = "-".join(
+                        s.replace("\\", r"\\").replace("-", r"\-") for s in chunks
+                    )
+                # Escape set operations (&&, ~~ and ||).
+                stuff = re.sub(r"([&~|])", r"\\\1", stuff)
+                i = j + 1
+                if not stuff:
+                    # Empty range: never match.
+                    add("(?!)")
+                elif stuff == "!":
+                    # Negated empty range: match any character.
+                    add(".")
+                else:
+                    if stuff[0] == "!":
+                        stuff = "^" + stuff[1:]
+                    elif stuff[0] in ("^", "["):
+                        stuff = "\\" + stuff
+                    add(f"[{stuff}]")
+        else:
+            add(re.escape(c))
+    assert i == n
+    return res
+
+
+def glob_translate(pat):
+    # Copied from: https://github.com/python/cpython/pull/106703.
+    # The keyword parameters' values are fixed to:
+    # recursive=True, include_hidden=True, seps=None
+    """Translate a pathname with shell wildcards to a regular expression."""
+    if os.path.altsep:
+        seps = os.path.sep + os.path.altsep
+    else:
+        seps = os.path.sep
+    escaped_seps = "".join(map(re.escape, seps))
+    any_sep = f"[{escaped_seps}]" if len(seps) > 1 else escaped_seps
+    not_sep = f"[^{escaped_seps}]"
+    one_last_segment = f"{not_sep}+"
+    one_segment = f"{one_last_segment}{any_sep}"
+    any_segments = f"(?:.+{any_sep})?"
+    any_last_segments = ".*"
+    results = []
+    parts = re.split(any_sep, pat)
+    last_part_idx = len(parts) - 1
+    for idx, part in enumerate(parts):
+        if part == "*":
+            results.append(one_segment if idx < last_part_idx else one_last_segment)
+            continue
+        if part == "**":
+            results.append(any_segments if idx < last_part_idx else any_last_segments)
+            continue
+        elif "**" in part:
+            raise ValueError(
+                "Invalid pattern: '**' can only be an entire path component"
+            )
+        if part:
+            results.extend(_translate(part, f"{not_sep}*", not_sep))
+        if idx < last_part_idx:
+            results.append(any_sep)
+    res = "".join(results)
+    return rf"(?s:{res})\Z"