A method called ATEST, which adaptively allocates test-time scaling resources for large language model inference, reports measuring up to a 4.1-fold increase in energy for the same four-sample query under GPU contention. The approach targets the gap left by fixed sampling budgets, which ignore both query difficulty and the changing energy cost of inference. The work is framed as a resource-aware alternative to uniform test-time scaling for LLM reasoning.
Test-time scaling (TTS) improves large language model (LLM) reasoning by sampling multiple answers, but fixed sampling budgets ignore both query difficulty and the changing energy cost of inference under GPU contention. We measure up to a 4.1-fold increase in energy for the same four-sample query as
EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.