MAGMA -- Multimodal Augmentation of Generative Models through Adapter-based Finetuning

Authors

repo (alphabetical)

Constantin (CoEich), Mayukh (Mayukhdeb), Sid (sdtblck)

paper

Constantin Eichenberg, Sidney Black, Samuel Weinbach, Aleph Alpha

Letitia Parcalabescu, Anette Frank, Heidelberg University

Abstract

Large-scale pretraining is fast becoming the norm in Vision-Language (VL) modeling. However, prevailing VL approaches are limited by the requirement for labeled data and the use of complex multi-step pretraining objectives. We present MAGMA - a simple method for augmenting generative language models with additional modalities using adapter-based finetuning. Building on Frozen, we train a series of VL models that autoregressively generate text from arbitrary combinations of visual and textual input. The pretraining is entirely end-to-end using a single language modeling objective, simplifying optimization compared to previous approaches. Importantly, the language model weights remain unchanged during training, allowing for transfer of encyclopedic knowledge and in-context learning abilities from language pretraining. MAGMA outperforms Frozen on open-ended generative tasks, achieving state of the art results on the OKVQA benchmark and competitive results on a range of other popular VL benchmarks, while pretraining on 0.2% of the number of samples used to train SimVLM.

Paper on arXiv: https://arxiv.org/abs/2112.05253

Examples (via Aleph Alpha playground)

Photos	Text & Technical

Model design

About the repository

In this repository we share the main parts of the codebase for training and inference of our MAGMA VL model. The main use of the repo is for downloading our pretrained weights and interacting with the model. We include a script for data parallel training with Deepspeed for finetuning our models or training a MAGMA model from scratch.

NOTE: The freely available model from this repo is only a demo. For the latest multimodal and multilingual models from Aleph Alpha check out our website.

Installation

Make sure PyTorch (Ver >= 1.9.0) and Torchvision are installed. See https://pytorch.org/get-started/locally/.

You can pip install from the git repository with:

pip install git+https://github.com/Aleph-Alpha/magma.git

Make sure that you also download the config:

mkdir configs; wget -O configs/MAGMA_v1.yml https://raw.githubusercontent.com/Aleph-Alpha/magma/master/configs/MAGMA_v1.yml

Or if you've cloned the repo, you can install all further requirements by:

pip install -r requirements.txt

Checkpoint

We also publish a model checkpoint that has been used for the publication. It is hosted on our infrastructure and downloads automatically. It can be downloaded manually here: https://bit.ly/aleph_alpha_magma_download

This checkpoint can also be played around with on a space managed by Heath Mitchell, AK, and Stella Biderman. (This is a 3rd party space, not managed by Aleph Alpha.)

Loading a model for inference

Downloads the checkpoint file into checkpoint_path if it's not already present.

from magma import Magma
from magma.image_input import ImageInput

model = Magma.from_checkpoint(
    config_path = "configs/MAGMA_v1.yml",
    checkpoint_path = "./mp_rank_00_model_states.pt",
    device = 'cuda:0'
)

inputs =[
    ## supports urls and path/to/image
    ImageInput('https://www.art-prints-on-demand.com/kunst/thomas_cole/woods_hi.jpg'),
    'Describe the painting:'
]

## returns a tensor of shape: (1, 149, 4096)
embeddings = model.preprocess_inputs(inputs)  

## returns a list of length embeddings.shape[0] (batch size)
output = model.generate(
    embeddings = embeddings,
    max_steps = 6,
    temperature = 0.7,
    top_k = 0,
)  

print(output[0]) ##  A cabin on a lake

Converting datasets to our format

To convert an image-caption dataset to our dataset class magma.datasets.ImgCptDataset, we suggest:

from magma.datasets.convert_datasets import convert_dataset

def my_dataset_iterator():
    """
    Implement an iterator for your dataset that for every datapoint yields a tuple
    image_path, {"captions": [...], "metadata": {...}, }, where image_path is the path to the image as a Path object, captions is a list of caption strings and metadata is an optional field.
    """

if __name__ == "__main__":
    convert_dataset(data_dir="/target/directory", ds_iterator=my_dataset_iterator())

How to train MAGMA

Run the training with:

deepspeed train.py --config path_to_my_config

To continue training from a deepspeed checkpoint, provide the checkpoint directory in the "load" config parameter.

WARNING: By default, instantiating magma via the init method instead of from_checkpoint loads the pretrained CLIP weights but not the pretrained gpt-j weights. For training MAGMA from scratch, download the gpt-j weights from this repo: https://github.com/finetuneanon/transformers and include them in the state dict after initializing the MAGMA model.