Quantize a tensor to FP4 using the Transformer Engine backend. Args: x (torch.Tensor): The input tensor to quantize.

Quantize a tensor to FP4 using the Triton backend. Args: x (torch.Tensor): The input tensor to quantize. con

Quantize a tensor to FP4 using the PyTorch backend. Args: x (torch.Tensor): The input tensor to quantize. co

Quantize a tensor to FP4 using the CUDA backend. Args: x (torch.Tensor): The input tensor to quantize. confi

Register a new type of weight conversion.

Run any tasks that should be used to calibrate models for a given PTQ method and set of parameters before running evaluation.

Run tests on a B200 on Modal.

Sample one entry matching (block_pos, timestep_index) when 2D, or just timestep_index when 1D. Non-owned timestep buckets are transpa

Save a generated LongLive video tensor as an mp4 file.

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

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_index

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

Configure ``pipeline`` for NVFP4 inference from a generator checkpoint. Handles both supported NVFP4 backends: * ``model_quant_use_transform

r""" Args: x(Tensor): Shape [B, L, num_heads, C / num_heads] seq_lens(Tensor): Shape [B] grid_sizes(Tensor): Shape [B, 3],

r""" Forward pass through the diffusion model Args: x (List[Tensor]): List of input video tensors, each with shape [C_in,

x: A list of videos each with shape [C, T, H, W]. t: [B]. context: A list of text embeddings each with s

Start the evaluation coordinator.

Static blockwise FP4 fake quantization using Triton kernel. Args: x: [NUM_FP4_BLOCKS, BLOCK_SIZE] on CUDA. amax: [NUM_FP4_BLOCKS]

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.

Convert a string to a boolean. Supported true values: 'yes', 'true', 't', 'y', '1' Supported false values: 'no', 'false', 'f', 'n', '0'

The kernel's template content.

Start a training job on Modal.

Worker process for running PTQ experiments locally.