OpenRubrics/RubricRM-8B-Rubric-v2 is an 8 billion parameter RubricRM-Rubric model, fine-tuned from Qwen3/Qwen3-8B by OpenRubrics. This model is specifically designed to extract rubric-style instructions from user requests, generating evaluation criteria for assessing response quality. It excels at creating universal, domain-agnostic principles and hard rules for comprehensive and concise evaluation. The model is optimized for generating structured rubrics for LLM alignment and reward modeling.
Loading preview...
OpenRubrics/RubricRM-8B-Rubric-v2 Overview
OpenRubrics/RubricRM-8B-Rubric-v2 is an 8 billion parameter model developed by OpenRubrics, fine-tuned from the Qwen3/Qwen3-8B architecture. Its core function is to generate structured, rubric-style evaluation criteria from user prompts, intended for assessing the quality and adherence of responses to specific requests. This model is particularly useful for tasks requiring precise and objective evaluation frameworks.
Key Capabilities
- Rubric Extraction: Automatically extracts and formats evaluation rubrics from natural language requests.
- Categorization: Distinguishes between "Hard Rules" (explicit requirements) and "Principles" (abstracted, domain-agnostic quality criteria).
- Comprehensiveness: Ensures rubrics cover all critical aspects, including explicit requirements and implicit quality standards.
- Conciseness & Uniqueness: Generates distinct, non-redundant evaluation criteria, merging overlapping items for precision.
- Structured Output: Formats rubrics as numbered lists, with each item starting "The response" and appending
[Hard Rule]or[Principle].
Good For
- LLM Alignment: Generating synthetic rubrics for reward modeling and aligning large language models.
- Automated Evaluation: Creating clear, objective criteria for automated assessment of generated text.
- Instruction Following: Developing detailed guidelines to ensure LLM outputs strictly adhere to user instructions.
- Research: Supporting research in scalable synthetic rubric generation, as detailed in the associated paper "OpenRubrics: Towards Scalable Synthetic Rubric Generation for Reward Modeling and LLM Alignment".