Published:
CENIA released LatamGPT’s weights without benchmarks, and the technical report isn’t out yet—so here’s an early, independent look at how it does on Spanish tasks and on CENIA’s new cultural benchmarks.
Published:
Making Spanish NLP accessible: lightweight language models (ALBETO, DistilBETO), evaluation benchmarks (Spanish SentEval and DiscoEval), and seq2seq models (BARTO, T5S)—all open.
Published:
Recommending in-game items in MOBA games like League of Legends—from a data-mining baseline to a Transformer (TTIR) that uses match context and attention weights as explanations. Two RecSys papers.