Skip to content

Latest commit

 

History

History
72 lines (46 loc) · 2.41 KB

README.md

File metadata and controls

72 lines (46 loc) · 2.41 KB
Raw

[Colab]

RUDOLPH 🦌🎄☃️

One Hyper-Tasking Transformer can be creative as DALL-E and GPT-3 and smart as CLIP

RUssian Decoder On Language Picture Hyper-tasking (RUDOLPH) is a text-image-text transformer designed for an easy fine-tuning for a range of tasks: from generating images by text description and image classification to visual question answering and more. This model demonstrates the power of Hyper-tasking Transformers.

Hyper-tasking model is a generalized multi-tasking model, i.e., the model that can solve almost all tasks within supported modalities, mandatory including mutual pairwise translations between modalities (two modalities in case of RUDOLPH: images and Russian texts).

Models

The following table shows the values of the parameters corresponding to different RUDOLPH versions.

350M 1.3B 2.7B
l 64 128 384
r 64 128 128
m 16 32 24
n 16 32 24

Sparse Attention Mask

350M

row - col - row - [last] conv

1.3B

row - col - row - [last] conv

2.7B

row - col - row - [last] conv

Installing

pip install rudolph==0.0.1rc10

Usage and Fine-Tuning

Usage and fine-tuning examples for different versions of RUDOLPH can be found in jupyters folder.

Citation

@misc{github2022ruDolph,
  title         = {RUDOLPH: One Hyper-Tasking Transformer can be creative as DALL-E and GPT-3 and smart as CLIP},
  author        = {AIRI},
  year          = {2022},
  howpublished  = {\url{https://github.com/ai-forever/ru-dolph}},
}

Supported by