New Asynchronous Flow Map Language Models Achieve State-of-the-Art in One-Step Generation

Researchers have introduced Asynchronous Flow Map Language Models (aFMLMs), a novel approach to one-step parallel language modeling that incorporates prediction order. Unlike iterative models, aFMLMs learn flow maps of asynchronous flows, which dictate a tokenwise time schedule to control denoising progress. This method allows models to generate all tokens in a single forward pass by internalizing the learned ordering. In evaluations on Sudoku solving and code generation for GSM8K math problems, aFMLMs achieved new state-of-the-art performance for one-step modeling. AI

IMPACT Introduces a novel one-step parallel generation method that could improve efficiency and performance in language models.

RANK_REASON The cluster contains a research paper detailing a new language model architecture published on arXiv. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• alphaXiv
• arXiv
• Asynchronous Flow Map Language Models
• CatalyzeX Code Finder for Papers
• CORE Recommender
• DagsHub
• Gotit.pub
• GSM8K
• Hugging Face
• Influence Flower
• ScienceCast
• sudoku

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →