Rubric-based reinforcement learning (RL) has become a popular approach for fine-tuning large language models (LLMs) by evaluating outputs against explicit criteria. However, in GRPO-style pipelines, these structured judgments are collapsed into a single scalar response-level reward, which is then broadcast uniformly to all generated tokens. This leaves no explicit mechanism for allocating credit within a response, even when different criteria depend on specific spans, formatting decisions, or semantic choices.
To address this, researchers at Hugging Face introduce CoRT (Counterfactual Replay for Token-Level Rubric-Guided Policy Optimization), a token-level credit weighting method for rubric-conditioned GRPO. Instead of training an auxiliary token scoring model, CoRT uses counterfactual replay to rescore the same sampled response under the original rubric-conditioned prompt and a matched criteria-free prompt. The resulting tokenwise log-likelihood contrasts serve as a proxy for dependence on the rubric context.
CoRT maps these contrasts to bounded, response-normalized weights and uses them to redistribute the signed GRPO advantage across tokens. This approach avoids introducing an auxiliary scorer or changing the response-level reward. Experiments across instruction-tuned models and reward granularities show that CoRT improves over matched response-level GRPO in the vast majority of comparisons, with an average gain of 4.4 percentage points.
The method remains competitive with learned token-level credit baselines while avoiding a separate relevance-learning stage. These results suggest that policy-internal counterfactual likelihood contrasts provide an effective training signal for within-response credit allocation while retaining the simplicity and stability of GRPO.