alphaone-test-time-reasoning

Dynamically modulate reasoning depth at test time using alpha moments and Bernoulli scheduling to optimize inference speed-quality tradeoffs without retraining.

Independent GitHub snapshot · Automated classification · Original documentation governs