[Submitted on 14 Apr 2025 (v1), last revised 18 Nov 2026 (this version, v4)]
Abstract:Reinforcement learning (RL) powers two major lines of large language model (LLM) post-training: reasoning, which elicits deliberate, verifiable problem solving, and alignment, which steers models toward helpful, harmless, and honest behavior. Both hinge on reward design. Reasoning relies on rule-based verification of final answers, whose binary feedback is blind to intermediate derivations, whereas process reward models (PRMs) require step-level annotations too costly to scale. Alignment depends on reward models (RMs) trained on binary preference pairs, making both data collection and model training resource-intensive. We propose RL2R, reinforcement learning with reference-based rewards, as a unified and inexpensive alternative. Using pre-existing or distilled references, RL2R measures semantic similarity with lightweight metrics such as BERTScore between a sampled generation and a reference answer or solution. Similarity preserves ordinal quality signals that binary feedback discards: responses closer to the reference earn higher rewards, and sound derivations receive credit even when the final conclusion is wrong. For mathematical reasoning, augmenting rule-based verification with answer- and step-level similarity provides low-cost process supervision and outperforms answer-only verification. For alignment, a critic-free REINFORCE-style instantiation covers general preference, safety, and confidence tasks on par with RM-based methods. Reference-based rewards thus offer a cost-effective, versatile reward source spanning reasoning and alignment.
Submission history
From: Shuai Zhao [view email]
[v1] Mon, 14 Apr 2025 05:43:21 UTC (414 KB)
[v2] Sun, 22 Jun 2025 13:02:25 UTC (337 KB)
[v3] Mon, 13 Oct 2025 05:00:17 UTC (466 KB)
[v4] Wed, 18 Nov 2026 11:06:19 UTC (2,156 KB)
