Antidoom: Stopping AI Doom Loops with Final Token Preference Optimization

Reducing Doom Loops with Final Token Preference Optimization

I developed Antidoom to stop reasoning models from getting stuck in repetitive loops that waste context windows. Instead of generic fixes, we use Final Token Preference Optimization to retrain the model at the exact moment it starts looping. This targeted approach slashed failure rates from 10% to 1.4% on LFM2.5-2.6B and Qwen3.5-4B, significantly boosting performance without degrading other capabilities.

"There has been a prevailing wisdom that higher temperatures may be preferable for reasoning models, allowing them to explore the solution space. However, this intuition may be misplaced, being conflated with the dominant effect of doom-looping."