COFFEE operates at each diffusion step: a target-free graphical model (the dependence carrier) absorbs marginal token distributions from the denoiser to construct a joint model over unresolved positions.
Framework guides discrete diffusion with sequence-level objectives without exponential enumeration
COFFEE couples unresolved token predictions and compiles preferences into tractable graphical models for polynomial-time inference-time guidance.
Top University
Hua (Edward) · Xu · Dongxin Li · Gwen Yidou-Weng · Guy Van den Broeck · Wei Wang · +1 more
Research Digest··2 min read
The authors introduce COFFEE, a plug-and-play framework that avoids exponentially enumerating all possible completions when guiding discrete diffusion models with sequence-level objectives.
Why this paper
From National University of Singapore and 2 others
In one line
COFFEE enables future-aware guidance for discrete diffusion without retraining by compiling sequence-level objectives into graphical models.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§