PrismaCoder: Pushing the Boundaries of Open Multimodal Code Intelligence through Decomposed Rubric Synthesis
PrismaForge uses the same synthesized rubric to filter training data and reward models for visual coding.
The rubric breaks program correctness, construction quality and rendered appearance into checkable criteria. The authors use that pipeline to build PrismaCoder-2M, which they describe as the largest and most diverse open dataset for multimodal code intelligence. They release two models, PrismaCoder-9B and PrismaCoder-35B, and report that both outperform existing open models and rival leading proprietary systems. The paper is marked as work in progress.
ArXiv · AI/CL/LG's note
The rubric breaks program correctness, construction quality and rendered appearance into checkable criteria. The authors use that pipeline to build PrismaCoder-2M, which they describe as the largest and most diverse open dataset for multimodal code intelligence. They release two models, PrismaCoder-9B and PrismaCoder-35B, and report that both outperform existing open models and rival leading proprietary systems. The paper is marked as work in progress.
ArXiv · AI/CL/LG's note
score 5