New flow-map framework lets generators expand output size on the fly
gottapatchemall · x · 2026-07-29
Expanding Flow Maps lets flow models grow their output size during generation
The paper introduces Expanding Generative Flows (EFlows) and Expanding Flow Maps (EFMs), a framework for few-step flow-based generation that does not require a fixed output canvas upfront.
- Instead of committing to a sequence length or state dimension at initialization, the model can expand the state with new coordinates or tokens as generation proceeds.
- The framework splits generation into two learnable parts: an expand operator that adds new state, and a transport map that denoises and moves the expanded state along the interpolation path.
- The same idea is presented as a unified way to handle variable-length text, graphs, and 3D molecular conformers.
- The authors argue that output size itself can be treated as a learned degree of freedom, rather than a fixed design choice.
Related event: New Flow Map Framework Enables Dynamic Output Expansion During Inference(3 posts)→
More from Research
- Tsinghua professor's ApexIntelligence raises $50M to build self-evolving foundation models — DeryaTR_ · 2026-10-16
- Anders Sandberg on fluid computation: error budgets tighten quadratically over time — anderssandberg · 2026-10-16
- How the fluid computation works: force field advects particle to 0/1 boxes — anderssandberg · 2026-10-16
- FlashAttention creator Tri Dao joins Radical Numerics as scientific advisor — exnx · 2026-10-16
- SAIR Launches Open Math OS, Offering Free Compute and 100+ Tools for Community-Owned Math AI — Thom_Wolf · 2026-10-16
- JapanFold Claims One Tenstorrent Server Beats an Nvidia DGX B200 on Bio Models at a Third of the Price — DavidBennett__ · 2026-10-16