Compute log(alpha_t) of a given continuous-time label t in [0, T].

The noise predicition model function that is used for DPM-Solver.

The noise predicition model function that is used for DPM-Solver.

Create a wrapper function for the noise prediction model. DPM-Solver needs to solve the continuous-time diffusion ODEs. For DPMs trained on discr

Thanks to DPM-Solver for their code base

Multistep solver DPM-Solver-2 from time `t_prev_list[-1]` to time `t`. Args: x: A pytorch tensor. The initial value at t

Multistep solver DPM-Solver-3 from time `t_prev_list[-1]` to time `t`. Args: x: A pytorch tensor. The initial value at t

For some beta schedules such as cosine schedule, the log-SNR has numerical isssues. We clip the log-SNR near t=T within -5.1 to ensur

Sample x_{t-1} from the model at the given timestep. :param model: the model to sample from. :param x: the current tensor at

Generate samples from the model and yield intermediate samples from each timestep of diffusion. Arguments are the same as p_s

Prepend previous chunk's latent tail as left context. Args: z: Current latent chunk [B, C, T, H, W] prepend_prev_late

Print layer information, including parameter size and whether 8bit quantization is used Args: param_shape (tuple): parameter shap

Processes a directory containing PNG files, generates corresponding JSON files, and packages all files into a TAR file. It also counts the nu

Process mask with optional dilation and blur Args: mask: PIL Image mask blur_factor: int, blur radius (0 = n

Input: grad_y: (B, N, H*C) proj_weight: (H*C, H*C) vk_q: (B, N, H, C+1) Output: grad_vk_q: (B, N, H, C+1)

Project 3D points in camera frame to UCM image plane.

Project 3D points in camera frame to UCM image plane using FoV-based intrinsics.

Get the distribution q(x_t | x_0). :param x_start: the [N x C x ...] tensor of noiseless inputs. :param t: the number of diff

Run the LTX-2 refiner and return refined VAE latents. When ``block_size`` is ``None`` (default), uses the legacy single-shot path tha

Register a progress bar callback function Args: progress_fn: Callback function that takes current step and total steps a

Give position of A relative to B, factoring in object dimensions

Return an RGBA overlay of size ``(width, height)``.

r""" Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on Section 3.4 from [Common

Rescales betas to have zero terminal SNR Based on https://arxiv.org/pdf/2305.08891.pdf (Algorithm 1) Args: betas (`torch.FloatTensor`

Resolve a dataset description. This rebases the shards as necessary and loads any remote references. Dataset descriptions are JSON files. Th

Return an ffmpeg executable path, falling back to imageio-ffmpeg.

CUDA cam path: reads q/k/v [B,H,D,N] directly, writes fp32 [B,H,D,N] directly (no packing, no transpose). Phase B via Triton (skip_z). reuse_

Drive the three-stage chunked pipeline end-to-end. Args: stage1_chunk_iter: Iterator (typically from ``SelfForcingFlowEulerCa