RoleMRC: A Fine-Grained Composite Benchmark for Role-Playing and Instruction-Following

TL;DR

RoleMRC introduces a large-scale, fine-grained benchmark that couples role-playing with instruction-following in LLMs. It builds a meta-pool of 10k role profiles, 38k role-playing instructions, and 1.4k test samples, spanning free chats, on-scene dialogues, and ruled chats. The authors propose a comprehensive evaluation pipeline using reference-based metrics and a reference-free LLM-as-a-judge, and show that fine-tuning on RoleMRC improves instruction-following without sacrificing general abilities; cross-benchmark evaluations confirm benefits and reveal an alignment tax that can be mitigated via targeted neuron-level restraints. The work also provides neural activation analyses to understand how post-tuning reshapes model behavior and offers a public resource for replication.

Abstract

Role-playing is important for Large Language Models (LLMs) to follow diverse instructions while maintaining role identity and the role's pre-defined ability limits. Existing role-playing datasets mostly contribute to controlling role style and knowledge boundaries, but overlook role-playing in instruction-following scenarios. We introduce a fine-grained role-playing and instruction-following composite benchmark, named RoleMRC, including: (1) Multi-turn dialogues between ideal roles and humans, including free chats or discussions upon given passages; (2) Role-playing machine reading comprehension, involving response, refusal, and attempts according to passage answerability and role ability; (3) More complex scenarios with nested, multi-turn and prioritized instructions. The final RoleMRC features a 10.2k role profile meta-pool, 37.9k well-synthesized role-playing instructions, and 1.4k testing samples. We develop a pipeline to quantitatively evaluate the fine-grained role-playing and instruction-following capabilities of several mainstream LLMs, as well as models that are fine-tuned on our data. Moreover, cross-evaluation on external role-playing datasets confirms that models fine-tuned on RoleMRC enhances instruction-following without compromising general role-playing and reasoning capabilities. We also probe the neural-level activation maps of different capabilities over post-tuned LLMs. Access to our RoleMRC, RoleMRC-mix and Codes: https://github.com/LuJunru/RoleMRC.

Figures (15)

• Figure 1: Example of instructional requests from human user, answered by role-playing LLMs in different ways.
• Figure 2: Schematic overview of RoleMRC's construction, which consists of persona sampling, role profile standardization and multi-stage dialogue synthesis. Partial icons are copyrighted by PersonHub ge2024scaling.
• Figure 3: The strategy of gradually synthesizing finer role-playing instructions in step 3 of Figure \ref{['fig:method']}.
• Figure 4: Visualization of reference-free LLM-as-a-judge results. We provide numerical result in Table \ref{['tab:llm_as_judge']}.
• Figure 5: Discrepancies between SFT and DPO neuron activations (top-20% active neurons) in LLaMA3.1-8B for multi-turn instructions. Layers 3-11 show minimal changes (green), while layers 12--31 exhibit larger shifts (red).