DelTA: Discriminative Token Credit Assignment for Reinforcement Learning from Verifiable Rewards
For researchers improving LLM reasoning via RL, DelTA offers a simple fix to a known credit assignment bottleneck, yielding consistent gains.
The paper identifies a limitation in standard sequence-level RLVR for LLMs, where token credit assignment is dominated by high-frequency patterns like formatting tokens, diluting discriminative signals. The proposed DelTA method amplifies side-specific token-gradient directions, achieving 3.26 and 2.62 average point improvements on Qwen3-8B-Base and Qwen3-14B-Base across seven math benchmarks.
Reinforcement learning from verifiable rewards (RLVR) has emerged as a central technique for improving the reasoning capabilities of large language models. Despite its effectiveness, how response-level rewards translate into token-level probability changes remains poorly understood. We introduce a discriminator view of RLVR updates, showing that the policy-gradient update direction implicitly acts as a linear discriminator over token-gradient vectors and thereby determines which token probabilities are increased or decreased during learning. Under standard sequence-level RLVR, this discriminator is constructed from positive- and negative-side centroids formed by advantage-weighted averaging of token-gradient vectors. However, such centroid construction can be dominated by shared high-frequency patterns, such as formatting tokens, diluting sparse yet discriminative directions that better distinguish high-reward responses from low-reward ones. To address this limitation, we propose $\textbf{DelTA}$, a discriminative token credit assignment method that estimates token coefficients to amplify side-specific token-gradient directions and downweight shared or weakly discriminative ones. These coefficients reweight a self-normalized RLVR surrogate, making the effective side-wise centroids more contrastive and thereby reshaping the RLVR update direction. On seven mathematical benchmarks, DelTA outperforms the strongest same-scale baselines by 3.26 and 2.62 average points on Qwen3-8B-Base and Qwen3-14B-Base, respectively. Additional results on code generation, a different backbone, and out-of-domain evaluations further demonstrate the generalization ability of DelTA.