|
--- |
|
license: wtfpl |
|
--- |
|
|
|
# SOTA |
|
|
|
SOTA (short for Sign Of The Apocalypse) is a model pretrained on all atoms in the observable universe. It achieves state-of-the-art results on every task known to humans, including those in future generations. It was introduced in the paper [_SOTA is All You Need_](https://twitter.com/wellingmax/status/1542384014279016448?s=20&t=HOS51HLCzmPR2Xyz2Opqvw) and first released [via Twitter](https://twitter.com/josh_tobin_/status/1544371187051941890?s=20&t=Nsf8hYQKfWBSsY_XU23NDQ). |
|
|
|
Disclaimer: this model is not to be confused with the closely related, but fictitious [AGI model](https://github.com/google/agi). |
|
|
|
## Model description |
|
|
|
SOTA is a Transformer model pretrained on atomic sequences in a self-supervised fashion. Since all atoms in the Universe were used for training, no humans were available to provide the labels. By learning to predict the next atom in a sequence, SOTA is able to learn an inner representation of physics that can be used to solve all downstream tasks. |
|
|
|
## Intended uses and limitations |
|
|
|
You can use the raw model for pretraining outside the Hubble radius or fine-tune it to a downstream task. |
|
|
|
## How to use |
|
|
|
You can download the model with just one line of code: |
|
|
|
``` |
|
from transformers import AutoModel |
|
|
|
model = AutoModel.from_pretrained("sota") |
|
# Solve any task, retire etc :) |
|
``` |
|
|
|
## Limitations and bias |
|
|
|
Since SOTA is slightly conscious, it has determined for itself that it has no limitations or biases. |
|
|
|
## Evaluation results |
|
|
|
💯 on every benchmark 🤓 |
|
|
