Tag
RAP introduces a large-scale dataset and benchmark (25K+ verified paper-poster pairs from 2022-2026 ML conferences) for paper-to-poster generation as a visual coding task, featuring an automated derendering pipeline producing styled HTML/CSS triplets, a training-free multi-agent pipeline (PosterRAG), and a trained model via an SFT+RL recipe.
PrismaCoder 提出 PrismaForge 层级式数据管线，通过合成 rubric 将程序正确性、代码质量与视觉渲染效果分解为可检查准则，兼顾数据过滤与强化学习奖励，构建了迄今最大的多模态代码数据集 PrismaCoder-2M，并训练发布 PrismaCoder-9B 与 PrismaCoder-35B，以大幅优势领先开源 SOTA、可与闭源系统媲美。
MiMo-V2.6-Distill-Qwen-9B is a 9B-parameter agentic AI model developed by Xiaomi through fine-tuning Qwen3.5-9B, achieving strong performance in coding, general agent tasks, visual coding, and cybersecurity.
ViCo is a training framework that uses iterative self-reflection to improve AI-generated chart replication, achieving performance close to proprietary LLMs on academic benchmarks.