LLAVA

Arguments

Options

--llava_model_namestr

Help: Name of the LLAVA model to use

• Default: llava-hf/llava-1.5-7b-hf

--base_promptstr

Help: Base prompt for the LLAVA model

• Default: ``USER: <image>

<prompt> ASSISTANT:`` --load_in_4bit : 0|1|True|False -> bool

Help: Load the model in 4-bit mode

• Default: 1

--classification_promptstr

Help: Prompt to use for classification. If <classnames> is present, it will be replaced with the class names. If <datasetname> is present, it will be replaced with the dataset name

• Default: What object is in the photo ? Pick only one and answer with the class name <classnames>

Classes

class models.llava.FinalModel(dataset, args, denorm_transform, device)

Bases: Module

forward(x)

get_closest_classname(pred_class_name)

class models.llava.Llava(backbone, loss, args, transform, dataset=None)

Bases: ContinualModel

STATIC Continual Learning with LLAVA.

COMPATIBILITY: List[str] = ['class-il', 'domain-il', 'task-il', 'general-continual']

NAME: str = 'llava'

begin_task(dataset)

forward(x)

static get_parser(parser)

Return type:

ArgumentParser

observe(inputs, labels, not_aug_inputs, epoch=None)