PrismaCoder: Pushing the Boundaries of Open Multimodal Code Intelligence through Decomposed Rubric Synthesis

Key points

• The same rubric serves as both filter and reward: it curates the supervised corpus, then drives reinforcement learning on visual-coding tasks.1
• Leveraging PrismaForge for data synthesis and filtering, we build PrismaCoder-2M, the most diverse open dataset for multimodal code intelligence and the largest to date.1
• On the model side, we train PrismaCoder-9B and PrismaCoder-35B on this data and release both.1
• Our analyses further show that the proposed rubric-based pipeline improves multimodal coding performance across model backbones.1

Sources (2)

• [1]PrismaCoder: Pushing the Boundaries of Open Multimodal Code Intelligence through Decomposed Rubric Synthesis
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 05:32 PM

  Multimodal code intelligence connects the symbolic world of programs with the visual world their execution creates.

  The same rubric serves as both filter and reward: it curates the supervised corpus, then drives reinforcement learning on visual-coding tasks.

• [2]PrismaCoder: Pushing the Boundaries of Open Multimodal Code Intelligence through Decomposed Rubric Synthesis
  Hugging Face Daily Papers · Oct 14, 12:00 AM · same content

Extractive summary: sentences quoted from the sources.