Lafa Scales Continuous Flow Language Models to 8B with 2.54x Throughput

Lafa, a continuous flow language model from UC Berkeley, Amazon, KAIST and CMU, scales to 8B parameters for the first time, achieving 2.54x the throughput of autoregressive models and showing flow-based modeling as a scalable paradigm.

2026-10-24 ~ 2026-10-26 · 2 related posts

• Lafa: 8B flow language model hits 2.54x AR throughput with under 30B training tokens — LucaAmb · 2026-10-24
• Lafa paper scales flow models to 8B by adapting AR weights, 2.54x AR throughput — burny_tech · 2026-10-26