Load an OLMo model from a checkpoint.

Generate token IDs using beam search. Note that by default ``beam_size`` is set to 1, which is greedy decoding. :param inpu

Run a query through an OpenAI model

Gets the default evaluator for task `name`

Get the font size based on the size of the image.

Use the Levenshtein distance (or edit distance) to determine which of the choices is most similar to the given prediction

Separate parameters into connector/vit/llm weight decay and non weight decay groups.

Separate parameters into weight decay and non weight decay groups.

Build a list of strings of what a user might type in to the model for the given example, and its responses, by applying a prompt template to t

Sample `Gumbel(phi)`. `phi` should have shape `(batch_size, num_classes)`.

Utility method to get a HTML rectangle element

Use this to load a state dict for an FSDP model and its optimizer via :module:`torch.distributed.checkpoint` functions. This should be used d

Load an OLMo config file and convert it into a MM-OLMo config

Preprocesses a single image Returns: crops: (n_crops, n_patches, patch_dim) individual crops, `n_crops` might

Calculates relaxed correctness. The correctness tolerates certain error ratio defined by max_relative_change. See https://arxiv.org/pdf/2203.

Resize an image while padding to preserve uts aspect ratio.

Returns the beams with the highest perturbed log probabilities.

Use this to save a state dict for an FSDP model and its optimizer via :module:`torch.distributed.checkpoint` functions. This should be used d

Score the final predictions found by beam search. Returns a tensor of the final sequence scores of shape `(batch_size, beam_size)`.

Divide in image of size [w, h] in up to max_num_patches of size patch_size

:param rank0_only: INFO and below messages will only be emitted on the rank0 process.