Paper reframes LLM decoding as optimization on a simplex, better via mirror descent
hbouammar · x · 2026-09-29
A new paper formalizes LLM decoding—long treated as folklore—as an optimization problem over a probability simplex. The authors implement the idea with mirror descent, enabling mix-and-match composition of decoding strategies. They release an open-source library and show that compositional decoders often outperform standard off-the-shelf ones. Paper and code are both available.
More from Research
- KAN author Ziming Liu: LLMs learn alchemy, not the science of AI — ZimingLiu11 · 2026-10-24
- Training-Physics Intuition Compressed Into a Decision Tree Hits 78.2% Out-of-Fold — ZimingLiu11 · 2026-10-24
- LLMs Rank Optimizers at 85% but Are Data-Blind on Architectures, New Benchmark Shows — ZimingLiu11 · 2026-10-24
- NVIDIA releases 43K instruction-hierarchy rubrics for judging system vs user prompts — kalyan_kpl · 2026-10-24
- Architecture IQ benchmark shows LLMs still lag humans in training intuition, ResNet-style — ZimingLiu11 · 2026-10-24
- François Chollet declares "Symbolic learning works" in terse but weighty statement — fchollet · 2026-10-24