Researchers from Soochow University have introduced OpenBA, a groundbreaking 15 billion-parameter bilingual sequence-to-sequence (seq2seq) model for natural language processing, focusing on Chinese language processing. Distinct from predecessors like BLOOM and AlexaTM, OpenBA enhances the encoder-decoder framework and optimally balances English and Chinese tokens. Its unique architecture, favoring a shallow encoder and deep decoder, has shown notable efficacy in denoising tasks. Impressively, OpenBA performs commendably across benchmarks, especially in tasks like translation and summarization.
Read the full story — https://news.superagi.com/2023/09/20/soochow-university-unveils-openba-a-pioneering-15b-bilingual-seq2seq-model-in-nlp/
Top comments (0)