Spaces:

BramVanroy

/

text-to-amr

like 3

Running

This work is part of the Horizon 2020 research project SignON (https://signon-project.eu/). Its goal is to bridge the communication gap between deaf, hard of hearing, and hearing people. Part of the research involves creating an interlingua to translate between different modalities. To this end, we will use abstract meaning representation (AMR). We trained models (available on the HF hub) and show-case its capabilities in this space. Visitors can generate a visual graph representation of text.

Under the hood, these models use MBART, which is quite large due to its massive vocabulary. The free-tier space is therefore very slow. For that reason I wish to request a community GPU grant to improve the user experience.