dongboklee

/

gPRM-8B

like 0

Hi! Great work on this model and the detailed model card. This PR introduces a couple of improvements:

• Adds the direct Hugging Face paper link for "Rethinking Reward Models for Multi-Domain Test-Time Scaling" for improved discoverability, while retaining the arXiv link.
• Fixes logical errors in the "Direct Use" code snippet related to prompt formatting (steps variable definition and prefix assignment) and how the model's output scores are accessed. The updated snippet should now run without errors, ensuring users can correctly try out the model.

Thanks for sharing your work with the community!