PrismaCoder: Pushing the Boundaries of Open Multimodal Code Intelligence through Decomposed Rubric Synthesis
PrismaCoder releases 9B and 35B open models and a 2M multimodal code dataset trained with rubric rewards.
PrismaCoder introduces PrismaForge, a hierarchical pipeline that synthesizes a rubric covering program correctness, build quality, and visual rendering, then uses it to filter supervised data and reward reinforcement learning. The authors release PrismaCoder-2M, described as the largest and most diverse open multimodal code dataset, plus PrismaCoder-9B and PrismaCoder-35B. They report that both models lead open multimodal code intelligence by a wide margin and rival leading proprietary systems, and that the rubric improves results across backbones. Code and data are released publicly.
- PrismaForge uses a synthesized rubric as both data filter and RL reward.
- PrismaCoder-2M is released as a large open multimodal code dataset.
- PrismaCoder-9B and PrismaCoder-35B weights are released.
- Authors say both lead open models and rival leading proprietary systems.
- Code and data are published on the project site.
- Products
- PrismaCoder-2MPrismaForge
- AI models
- PrismaCoder-9BPrismaCoder-35B
From the original article
1 min read at arxiv.orgMultimodal code intelligence connects the symbolic world of programs with the visual world their execution creates. Interest in this capability is burgeoning, as models are increasingly asked to produce charts, webpages, animations, and semi-structured documents. Recently, proprietary models have pushed far ahead, producing interactive artifacts even from a text description or a screenshot alone.
By Qiushi Sun · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.