arXiv cs.AI· Michal Chudoba, Sergey Alyaev, Petra Galuscakova, Tomasz Wiktorski·· 14 小时前AI 评分32
ART：用一张输入图像微调多模态 LLM 的强化训练方法
Fine-tuning Multi-modal LLMs with ART: Art-based Reinforcement Training
AI 导读
研究提出 Art-Based Reinforcement Training（ART），通过优化一张辅助输入图像，让冻结的多模态 LLM 适配特定任务，推理时只需把该图像放进标准多模态提示词，无需改动服务引擎。
正文 · 原文
Abstract:Adapting Large Language Models (LLMs) to specific tasks usually involves fine-tuning: Low-Rank Adaptation (LoRA) edits the weights, while soft prompting prepends trainable virtual tokens. Both change how an already-optimized model is served, so serving many task-specific adapters at high throughput requires adapter-aware infrastructure. We propose Art-Based Reinforcement Training (ART), which optimizes a single auxiliary input image to specialize a frozen Multimodal Large Language Model (MLLM) for a specific task. At inference time, ART simply includes this image in each standard multimodal prompt, requiring no changes to the serving engine. We compare ART with LoRA and soft prompting across five reasoning benchmarks for four small MLLMs from three architectures. While different methods perform best for different tasks, ART ranks first or second in 10 of 20 benchmark-MLLM pairs, with its best results on the two math word problem benchmarks (GSM8k and SVAMP). It is significantly ahead of the next best adapter in 2 cases, within the noise of the next best adapter in 6, and significantly above the text-only baseline in 18. ART therefore reaches accuracy close to LoRA and soft prompts while serving faster than LoRA. In the vLLM serving stack, ART reduces throughput by 6% to 14% relative to text-only requests, compared to LoRA's 13% to 93% slowdown. Beyond efficient serving, ART artifacts are plain images, making them easy to store, distribute, and share. This approach keep task-specific adaptation outside the model weights, so the same frozen model can support different tasks without separate model variants.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2606.11854 [cs.LG] |
| (or arXiv:2606.11854v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2606.11854 arXiv-issued DOI via DataCite |
Submission history
From: Sergey Alyaev [view email]
[v1]
Wed, 10 Jun 2026 09:30:37 UTC (611 KB)
[v2]
Thu, 29 Oct 2026 17:56:43 UTC (2,111 KB)
来源：arXiv cs.AI · arxiv.org