Research map / Large language models / Language, safety and interpretability
Persona-Pruner: Sculpting Lightweight Models for Role-Playing
Jinsu Kim, Jihoon Tack, Noah Lee, Jongheon Jeong
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add persona-sensitivity filtering of queries (ablation) | Persona-Pruner · Alpaca-P, Llama-3.2-3B 25% sparsity | LLM-judge persona score: 78.80 -> 80.19 | Improved |
| add persona-conditioned answer rewriting (ablation) | Persona-Pruner · Alpaca-P, Llama-3.2-3B 25% sparsity | LLM-judge persona score: 39.36 -> 80.19 | Improved |
| add both filtering and rewriting (ablation) | Persona-Pruner · Alpaca-P, Llama-3.2-3B 25% sparsity | LLM-judge persona score: 32.50 -> 80.19 | Improved |
| use straight-through estimator instead of Gumbel-Softmax (ablation) | Persona-Pruner · Alpaca-P, Llama-3.2-3B 25% sparsity | LLM-judge persona score: 75.77 -> 80.19 | Improved |
| use straight-through estimator instead of Soft-TopK (ablation) | Persona-Pruner · Alpaca-P, Llama-3.2-3B 25% sparsity | LLM-judge persona score: 68.55 -> 80.19 | Improved |
| use straight-through estimator instead of Soft-TopK (ablation) | Persona-Pruner · PIQA, Llama-3.2-3B 25% sparsity | PIQA accuracy: 0.59 -> 0.71 | Improved |
| add answer rewriting (with filtering) (ablation) | Persona-Pruner · PIQA, Llama-3.2-3B 25% sparsity | PIQA accuracy: 0.74 -> 0.71 | Hurt |
| add persona-specific masks instead of one global mask learned from all personas (ablation) | Persona-Pruner · Alpaca-P, 10 personas | LLM-judge persona score | Improved |
| use persona-sensitivity filtered Alpaca-P queries versus random Alpaca queries | Llama-3.2-3B-Instruct dense · Alpaca-P vs Alpaca | LLM-judge persona score: 60.00 -> 84.86 (+24.86) | Improved |
| add recovery finetuning (ablation) | Persona-Pruner · RoleBench, Llama-3.2-3B 25% sparsity | LLM-judge persona score: 81.20 -> 84.26 | Improved |
| add recovery finetuning on general Alpaca (ablation) | LLM-Pruner · RoleBench, Llama-3.2-3B 25% sparsity | LLM-judge persona score: 48.43 -> 30.00 | Hurt |
| add multi-source data mixture (Alpaca, Open-Orca, Dolly) instead of single-source, scaling 10k to 40k (ablation) | Persona-Pruner · Alpaca-P, one persona | LLM-judge persona score | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- LLM-Pruner — LLM-Pruner: On the Structural Pruning of Large Language Models (NeurIPS 2023) · better on 16 of 16 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Quantifying and Optimizing Global Faithfulness in Persona-driven Role-playing (NeurIPS 2024 · similarity 0.89)
- CoSER: Coordinating LLM-Based Persona Simulation of Established Roles (ICML 2025 · similarity 0.88)
- Consistently Simulating Human Personas with Multi-Turn Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- PersonaManifold: Revealing and Exploiting Curved Geometry in LLM Persona Representations (NeurIPS 2026 · similarity 0.87)
- CharacterBench: Benchmarking Character Customization of Large Language Models (AAAI 2025 · similarity 0.87)
- Persona Generators: Generating Diverse Synthetic Personas for Arbitrary Contexts (NeurIPS 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
