Resets the currently set image.

Load the WikiText-* language modeling dataset: https://paperswithcode.com/dataset/wikitext-2 https://paperswithcode.com/dataset/w

Compute self-attention. Returns attn output [B, n, L, d].

Positionally encode points that are normalized to [0,1].

Apply any filtering or masking to logits Parameters ---------- logits : mx.array, shape = (n_batch, vocab_size) p

Expects a mlx tensor with length 2 in the last dimension. Requires the original image size in (H, W) format.

Transforms masked indices of `x` conditioned on unmasked indices using `func`.

Calculates boxes in XYXY format around masks. Return [0,0,0,0] for an empty mask. For input shape C1xC2x...xHxW, the output shape is C1xC2x..

Timestamp tokens are above other special tokens' id range and are ignored by `decode()`. This method decodes given tokens with timest

Encode video to latent. Args: x: Video tensor [F, H, W, C] (channels-last) Returns: Latent tensor [

Load the enwik8 language modeling dataset: https://mattmahoney.net/dc/textdata.html

Positionally encode points that are not normalized to [0,1].

Generator yielding latents at each denoising step. First yield: conditioning tuple (for mx.eval by caller) Subsequent yields

Generate images to monitor the progress of the finetuning.

Compute the output size given input size and target long side length.

Get relative positional embeddings according to the relative positions of query and key sizes. Args: q_size (int): size of qu

Generate a grid image from a batch of images. Assumes input has shape (B, H, W, C).

Swap the linear layers in the transformer blocks with LoRA layers.

Load the stable diffusion text encoder from Hugging Face Hub.

Load a Whisper ASR model Parameters ---------- name_or_path : str one of the official model names listed by `whisper.availab

Perform a forward pass on the decoder and return per-token logits

Encodes masks to an uncompressed RLE, in the format expected by pycoco tools.

Load the MNIST dataset in 4 tensors: train images, train labels, test images, and test labels. Checks `save_dir` for already downloaded

Pad or trim the audio array to N_SAMPLES, as expected by the encoder.

Remove padding and upscale masks to the original image size. Args: masks (mx.array): Batched masks from the mask_decoder

Predict masks for the given input prompts, using the currently set image. Input prompts are batched mlx tensors and are expected to a

Normalize pixel values and pad to a square input.

Load the PTB language modeling dataset: https://paperswithcode.com/dataset/penn-treebank

Removes small disconnected regions and holes in a mask. Returns the mask and an indicator of if the mask has been modified.

Compute a binary mask from an uncompressed RLE.

Apply 3D RoPE using mx.fast.rope with reshapes. Args: x: Tensor of shape [B, L, H, D] grid_sizes: List of [frames, height, w