Initialize the Boogu-Image mixed single-double stream transformer model.

Append a tensor to list and persist it to disk.

From "Beta Sampling is All You Need" [arXiv:2407.12173] (Lee et. al, 2024)

Constructs an exponential noise schedule.

Constructs the noise schedule of Karras et al. (2022).

Constructs the noise schedule of Lu et al. (2022).

Get interleaved instruction embeddings from VLM (self.mllm), aligned with training: - Build VLM inputs via processor.apply_chat_templ

Helper function to get unpadding data from attention mask.

Helper function to get unpadding data from a 2D attention mask [B, L].

Return the best execution device for a possibly offloaded module.

Return the best execution device for a possibly offloaded module.

The index for the first timestep. It should be set from pipeline with `set_begin_index` method.

The index for the first timestep. It should be set from pipeline with `set_begin_index` method.

Forward pass matching the interface of RMSNorm. Args: x: Input tensor residual: Optional residual tensor to

Generate rotary position embeddings and attention mask for prompt tuning. Args: batch_size: Batch size devic

Forward pass of the transformer block. Args: hidden_states: Input hidden states tensor attention_mask:

Run one dual-stream (double-stream) block step. Returns updated `(img_hidden_states, instruct_hidden_states)`.

Forward pass: context/refiner -> dual-stream (double-stream) -> fusion -> single-stream -> projection.

r""" Fuses the LoRA parameters into the original parameters of the corresponding blocks. <Tip warning={true}> This is an exp

Accept batch-shaped inputs where each sample is: - None (represents no image) - [] (empty sample, can be co

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

r""" Save the LoRA parameters corresponding to the prompt_embedding. Arguments: save_directory (`str` or `os.PathLike`):

r""" Save the LoRA parameters corresponding to the UNet and text encoder. Arguments: save_directory (`str` or `os.PathLik

Ensures interchangeability with schedulers that need to scale the denoising model input depending on the current timestep. A

Sets the begin index for the scheduler. This function should be run from pipeline before the inference. Args: begin_inde

Sets the begin index for the scheduler. This function should be run from pipeline before the inference. Args: begin_inde

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

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

The index counter for current timestep. It will increase 1 after each scheduler step.

The index counter for current timestep. It will increase 1 after each scheduler step.

r""" Reverses the effect of [`pipe.fuse_lora()`](https://huggingface.co/docs/diffusers/main/en/api/loaders#diffusers.loaders.LoraBaseM

Validates if the device parameter format is correct.