Gacha Decoding: Eliciting Diverse Generations Through Instruction Following
cs.AI, cs.CL
Submitted: 2026-10-01
Updated: 2026-10-01
License: http://creativecommons.org/licenses/by/4.0/
The gist: We introduce Gacha Decoding, an inference-time method for eliciting diverse language model generations that scales with model capability.
Terminology
Abstract
We introduce Gacha Decoding, an inference-time method for eliciting diverse language model generations that scales with model capability. Across open-ended domains (in-the-wild chat, creative writing, planning for image generation, and protein design), Gacha Decoding significantly outperforms existing generation diversity approaches at equal quality (up to 2.4x Vendi over the next-best prior approach), reaching the same number of high-quality modes with over an order of magnitude fewer samples (11.0x) and discovering novel modes that no other approach surfaces. Our key insight is to treat diversity as an instruction-following problem: rather than relying on the LM's token entropy, we combine its instruction-following capability with randomness from an external RNG tool to scalably identify and realize distinct modes of the response space. This approach of "planning with dice" enables Gacha to invert the long-observed tension between diversity and model capability. As the underlying LM becomes a better instruction follower, diversity under Gacha Decoding consistently improves--even as its token entropy and diversity under prior approaches decline. Together, our results highlight that instruction following, rather than token entropy alone, can drive generation diversity.
Sources
- How LLMs Distort Our Written Language
- gpt-oss-120b & gpt-oss-20b Model Card
- Intent Factored Generation: Unleashing the Diversity in Your Language Model
- The Entropy Mechanism of Reinforcement Learning for Reasoning Language Models
- The Vendi Score: A Diversity Evaluation Metric for Machine Learning
- DreamSim: Learning New Dimensions of Human Visual Similarity using Synthetic Data
- Scaling Synthetic Data Creation with 1,000,000,000 Personas
- The Curious Case of Neural Text Degeneration
- Where does output diversity collapse in post-training?
- Understanding the Effects of RLHF on LLM Generalisation and Diversity
- Jointly Reinforcing Diversity and Quality in Language Model Generations
- The Basic B*** Effect: The Use of LLM-based Agents Reduces the Distinctiveness and Diversity of People's Choices
- EQ-Bench: An Emotional Intelligence Benchmark for Large Language Models
- Toward Diverse Text Generation with Inverse Reinforcement Learning
- Evaluating the Diversity and Quality of LLM Generated Content
- Annotations Mitigate Post-Training Mode Collapse
- Gemma 4 Technical Report
- Base Models Beat Aligned Models at Randomness and Creativity
- Mode-Conditioning Unlocks Superior Test-Time Scaling
- Qwen-Image Technical Report
Related papers
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Agents' Last Exam
- The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
- MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
- Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems