Enables fused QKV projections. For self-attention modules, all projection matrices (i.e., query, key, value) are fused. For cross-a

See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298 Args: w (`tor

Wrapper function: transfrom velocity prediction model to denoiser Args: velocity: [batch_dim, ...] shaped tensor; velocity model o

Wrapper function: transfrom score prediction model to velocity Args: score: [batch_dim, ...] shaped tensor; score model output

returns a sampling function with given ODE settings Args: - sampling_method: type of sampler used in solving the ODE; default to be Do

returns a sampling function for calculating likelihood with given ODE settings Args: - sampling_method: type of sampler used in solvin

returns a sampling function with given SDE settings Args: - sampling_method: type of sampler used in solving the SDE; default to be Eu

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

Disables custom attention processors and sets the default attention implementation.

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

Disables the fused QKV projection if enabled. <Tip warning={true}> This API is 🧪 experimental. </Tip>