让生成模型在推理时动态调整参数，实现每张输入的个性化生成。
Test-Time Instance-Specific Parameter Composition: A New Paradigm for Adaptive Generative Modeling
- 推理时根据输入生成特定参数，注入预训练模型提升适应性。
- 无需微调或重训练，一次调整即可支持多步生成，效果显著提升。
- 适合轻量化模型、量化模型及测试时扩展场景，计算开销极低。
现有生成模型（如扩散模型和自回归网络）本质上是静态的，依赖固定预训练参数处理所有输入。而人类能灵活适应不同感知或想象情境。受此启发，我们提出Composer——一种基于测试时实例特定参数组合的新范式。Composer在推理时生成输入相关的参数调整，并注入预训练模型权重中，实现无需微调或重训练的逐输入定制化。该调整仅需一次，在多步生成前完成，带来更高品质、上下文感知的输出，且计算与内存开销极小。实验表明，Composer在多种生成模型和应用场景中均显著提升性能，包括轻量级/量化模型和测试时扩展。通过输入感知的参数组合，Composer建立了一种动态适应每个输入的生成模型新范式，突破了静态参数化的局限。
原文摘要 · Abstract (English)
Existing generative models, such as diffusion and auto-regressive networks, are inherently static, relying on a fixed set of pretrained parameters to handle all inputs. In contrast, humans flexibly adapt their internal generative representations to each perceptual or imaginative context. Inspired by this capability, we introduce Composer, a new paradigm for adaptive generative modeling based on test-time instance-specific parameter composition. Composer generates input-conditioned parameter adaptations at inference time, which are injected into the pretrained model's weights, enabling per-input specialization without fine-tuning or retraining. Adaptation occurs once prior to multi-step generation, yielding higher-quality, context-aware outputs with minimal computational and memory overhead. Experiments show that Composer substantially improves performance across diverse generative models and use cases, including lightweight/quantized models and test-time scaling. By leveraging input-aware parameter composition, Composer establishes a new paradigm for designing generative models that dynamically adapt to each input, moving beyond static parameterization.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。