Sample with first-frame conditioning support. Reads ``condition_frame_info`` from ``data_info`` (e.g. ``{0: 0.0}`` means frame 0 is f

latents: 1,C,F,H,W steps: int latents is only one sample but the model kwargs are 2 samples

Compute the sample at time `t_end` by DPM-Solver, given the initial `x` at time `t_start`. =========================================

For the PC-mode, please refer to the wiki page https://en.wikipedia.org/wiki/Predictor%E2%80%93corrector_method#PEC_mode_and_PECE_mod

Middle-anchor sampler (both sides), batched. - curf: * None: random curf per sample * int: same anchor frame

For the PC-mode, please refer to the wiki page https://en.wikipedia.org/wiki/Predictor%E2%80%93corrector_method#PEC_mode_and_PECE_mod

For the PC-mode, please refer to the wiki page https://en.wikipedia.org/wiki/Predictor%E2%80%93corrector_method#PEC_mode_and_PECE_mod

FSDP checkpoint save function, sharding

Calculates the FID of two paths

Save the current mask to disk

Save the original image to disk

Sets the discrete timesteps used for the diffusion chain (to be run before inference). Args: num_inference_steps (`int`):

Set up a new sequence

Singlestep DPM-Solver with the order `order` from time `s` to time `t`. Args: x: A pytorch tensor. The initial value at

Create a list of timesteps to use from an original diffusion process, given the number of timesteps we want to take from equally-sized portio

Predict the sample from the previous timestep by reversing the SDE. This function propagates the diffusion process from the learned m

One step for the SA-Predictor. Args: model_output (`torch.FloatTensor`): The direct output from the lear

One step for the SA-Corrector. Args: this_model_output (`torch.FloatTensor`): The model outputs at `x_t`

Sync documentation files from asset/docs to docs/. Priority: docs/ files take precedence over asset/docs/ files. Only copies files that

Convert tensor data type from uint8 to float, divide value by 255.0 and permute the dimensions of clip tensor Args: clip (torch.t

Drops labels to enable classifier-free guidance.

Drops labels to enable classifier-free guidance.

Drops labels to enable classifier-free guidance.

uniform distribution on the interval [0, 1)

uniform distribution on the interval [low, high)

Trim decoder output to remove context frames. Args: decoded: Full decoder output [B, C, T, H, W] prepend_prev_latent_

Exact inverse-CDF sampling of N(mu, sigma^2) truncated to [a, b] a,b in z-space. No mass at boundaries.

Unproject pixel grid into a camera-frame direction vector using the UCM.

x: (N, T, patch_size**2 * C) imgs: (N, H, W, C)

x: (N, T, patch_size**2 * C) imgs: (N, H, W, C)

x: (N, T, patch_size**2 * C) imgs: (N, H, W, C)

Update the previous latent tail for next chunk. Args: z: Current latent chunk before context prepending prepend_prev_

Update the reweighting using losses from a model. Sub-classes should override this method to update the reweighting using los

Merge a base URL and a relative URL. The function fills in any missing part of the url from the base, except for params, query, and fragment,

Repeat `val` for `repeat_time` times and return the list or val if list/tuple.

Repeat `val` for `repeat_time` times and return the list or val if list/tuple.

Input: grad_vk: (B, H, C+1, C), fp32 k: (B, N, H, C), fp16 v: (B, N, H, C), fp16 k_relu_mask: (B, N, H, C), bool

Input: grad_vk_q: (B, N, H, C+1), fp32 vk: (B, H, C+1, C), fp32 q: (B, N, H, C), fp16 q_relu_mask: (B, N, H, C),