Spaces:
Runtime error
Apply for a GPU community grant: Academic project
This project implements Few-Shot Sentiment Analysis for the Tunisian Dialect using MAML (Model-Agnostic Meta-Learning) on the TSAC dataset. We use TunBERT and Multilingual BERT as backbones to enable 1-shot learning in a low-resource language context.
Technical Requirement: > Meta-learning requires computing higher-order gradients (via the higher library), which creates a massive computation graph. My current implementation leads to Out-of-Memory (OOM) errors and crashes on standard CPU/Free-GPU setups during the meta-update phase. A GPU Grant is essential to handle the VRAM overhead required for these second-order derivatives.
Plan for the Space: The Space will serve as a research demo where users can input Tunisian dialect text and see the model adapt in real-time to specific sentiment nuances using few-shot adaptation.