Width Scaling
A form of test-time scaling in which a model generates multiple candidate paths at each step and uses a selection mechanism to choose the best branch. The central challenge lies in the design of the selection mechanism.
Timeline (last 90 days)
Width scaling means that the model generates multiple candidates at each step and then uses some selection mechanism to filter the optimal branch; common selection mechanisms include self-consistency voting and process reward model scoring.
In width scaling, the larger the width, the higher the probability of finding the correct path, but the computational cost also grows linearly.
The three dimensions of depth, width, and feedback share the same reasoning budget pool; increasing depth means reducing opportunities for width exploration, and investing in feedback iterations may sacrifice the depth of a single reasoning pass.
For mathematical proof tasks, depth scaling is usually more effective because the logical chain needs to remain coherent; for open-domain question answering, width scaling has more advantages; for code generation and debugging, feedback mechanisms are most valuable because compilers and test cases provide reliable external verification signals.
All Facts (4)
Width scaling means that the model generates multiple candidates at each step and then uses some selection mechanism to filter the optimal branch; common selection mechanisms include self-consistency voting and process reward model scoring.
50%UnverifiedIn width scaling, the larger the width, the higher the probability of finding the correct path, but the computational cost also grows linearly.
50%UnverifiedThe three dimensions of depth, width, and feedback share the same reasoning budget pool; increasing depth means reducing opportunities for width exploration, and investing in feedback iterations may sacrifice the depth of a single reasoning pass.
50%UnverifiedFor mathematical proof tasks, depth scaling is usually more effective because the logical chain needs to remain coherent; for open-domain question answering, width scaling has more advantages; for code generation and debugging, feedback mechanisms are most valuable because compilers and test cases provide reliable external verification signals.
50%