Create an overlay showing the image with mask highlighted

consistent with DMD: generate samples and train fake_score (SANA).

CUDA Phase A -> (I_P_kv, A, I_P_z, B_z), shapes/dtypes matching Triton.

Sample x_{t-1} from the model using DDIM. Same usage as p_sample().

Use DDIM to sample from the model and yield intermediate samples from each timestep of DDIM. Same usage as p_sample_loop_prog

Decode a batch of latents into images. Args: z: Input batch of latent vectors [B, C, T, H, W]. temb: Optional timeste

Stream-decode a latent video one latent frame at a time, yielding pixel chunks. Each iteration decodes a single latent frame and relies on th

Input: x: (B, H, W, C) weight: (C, K, K) bias: (C,) Output: y: (B, H, W, C)

Report whether the tensor contains any `nan` or `inf` entries. This is useful for detecting overflows/underflows and best to call rig

Compute the log-likelihood of a Gaussian distribution discretizing to a given image. :param x: the target images. It is assumed that this

Download a file from a remote url to a local path. The remote url can be a pipe: url, in which case the remainder of the url is treated as a c

The adaptive step size solver based on singlestep DPM-Solver. Args: x: A pytorch tensor. The initial value at time `t_T`

Dump a config file

Encode a batch of images into latents. Args: x: Input batch of images [B, C, T, H, W]. causal: Whether to use causal

Evaluate given image using detected objects on the global metadata specifications. Assumptions: * Metadata combines 'include' clauses wit

Expand the tensor `v` to the dim `dims`. Args: `v`: a PyTorch tensor with shape [N]. `dim`: a `int`. Returns: a P

Extracts the original image and greyscale mask from Gradio's ImageEditor data. - image_editor_data: dict with keys like 'background' and 'lay

Given a URL, return a local name for the shard.

Build pretrained Inception model for FID computation The Inception model for FID computation uses a different set of weights and has a slight

Find the latest checkpoint in the logdir.

Find the latest checkpoint in the logdir.

r""" Args: x(Tensor): Shape [B, L1, C] e(Tensor): Shape [B, C]

Forward pass of Sana. x: (N, C, H, W) tensor of spatial inputs (images or latent representations of images) t: (N,) tensor of

Forward pass of Sana. x: (N, C, T, H, W) tensor of spatial inputs (images or latent representations of images) t: (N,) tensor

Apply ``W += (B @ A) * strength`` for every LoRA pair into transformer_state. LoRA key convention (LTX official distilled LoRA): diffusio

Bidirectional fused GDN. Returns ``(B, N, H, D)``. Thin entry point kept for call-site stability; delegates to :func:`fused_bigdn_bidi_chunkw

denoise in chunk-wise manner, input/output is consistent with sample in SelfForcingFlowEuler.sample: - input/output: latents (B, C, T

Chunk-pipelined interactive generation. Runs stage-1 sampling, refiner AR blocks, and causal-VAE decode on three CUDA streams, emitti

Calculates the activations of the pool_3 layer for all images. Params: -- files : List of image files paths -- model : Instan

Get all checkpoints in the logdir sorted by step number.

Get all checkpoints in the logdir sorted by step number.