When to Ponder: Adaptive Compute Allocation for Code Generation via Test-Time Training

arXiv:2601.00894v2 Announce Type: replace-cross Abstract: Large language models apply uniform computation to all inputs, regardless of difficulty. We propose PonderTTT, a gating strategy using the TTT layer's self-supervised reconstruction loss to selectively trigger Test-Time Training (TTT) updates. The gating decision itself is training-free--requiring no learned classifier or auxiliary networks; only a single scalar threshold is initially calibrated on unlabeled data and continuously adapted via EMA to maintain target update rates. Our experiments with GPT-2 models (124M to 1.5B) on code language modeling (The Stack v2, teacher-forced perplexity) demonstrate that this signal is inference-compatible, requiring no ground-truth labels. Our Reconstruction Gating achieves 82-89% Oracle Recovery while being fully training-free, significantly outperforming Random Skip baselines (up to 16% lower loss on OOD languages).

Related

• Selective Left-Shift: Turning Test-Time Compute and Difficulty-based Curation into Training Data for Low-Resource Code Generation
• Uncertainty-Aware Budget Allocation for Adaptive Test-Time Reasoning
• Do Not Waste Your Rollouts: Recycling Search Experience for Efficient Test-Time Scaling
• Small RL Controller, Large Language Model: RL-Guided Adaptive Sampling for Test-Time Scaling
• CodeScaler: Scaling Code LLM Training and Test-Time Inference via Reward Models

Source: arXiv cs.CL | 2026-08-24