Let's (not) just put things in Context: Test-time Training for Long-context LLMs

Rachit Bansal, Aston Zhang, Rishabh Tiwari, Lovish Madaan, Sai Surya Duvvuri, Fnu Devvrit, David Brandfonbrener, David Alvarez-Melis, Prajjwal Bhargava, Mihir Kale, Samy Jelassi

Keywords: long-context language models, test-time training, inference-time scaling

TL;DR: We study the limitations of vanilla inference-time scaling approaches for long-context large language models and study test-time training as a promising alternative.

Abstract: Progress on training and architecture strategies has enabled LLMs with millions of tokens in context length. However, empirical evidence suggests that such long-context LLMs can consume far more text than they can reliably use. On the other hand, it has been shown that inference-time compute can be used to scale performance of LLMs, often by generating thinking tokens, on challenging tasks involving multi-step reasoning. Through controlled experiments on sandbox long-context tasks, we find that such inference-time strategies show rapidly diminishing returns and fail at long context. We attribute these failures to score dilution, a phenomenon inherent to static self-attention. Further, we show that current inference-time strategies cannot retrieve relevant long-context signals under certain conditions. We propose a simple method that, through targeted gradient updates on the given context, provably overcomes limitations of static self-attention. We find that this shift in how inference-time compute is spent leads to consistently large performance improvements across models and long-context benchmarks. Our method leads to large 12.6 and 14.1 percentage point improvements for Qwen3-4B on average across subsets of LongBench-v2 and ZeroScrolls benchmarks. The takeaway is practical: for long context, a small amount of context-specific training is a better use of inference compute than current inference-time scaling strategies like producing more thinking tokens.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Submission Number: 8016