FreedomIntelligence/HuatuoGPT-3-Grader-8B

TEXT GENERATIONPricing:Input $0.468 / Output $1.82Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 17, 2026Architecture:Transformer0.0K Featherless Exclusive Cold

FreedomIntelligence/HuatuoGPT-3-Grader-8B is an 8 billion parameter medical rubric grader model developed by FreedomIntelligence, designed to evaluate open-ended medical responses. Trained on GPT-4.1-annotated samples, it checks multiple criteria in a single generation to provide rewards for medical assessments. This model specializes in automated, objective grading of medical conversations based on predefined rubrics, offering a structured approach to evaluating clinical dialogue. It features a 32768 token context length, enabling comprehensive analysis of medical interactions.

Loading preview...

Overview

FreedomIntelligence/HuatuoGPT-3-Grader-8B is an 8 billion parameter model specifically engineered as a medical rubric grader. Developed by FreedomIntelligence, this model is trained on samples annotated by GPT-4.1 to assess open-ended medical responses against predefined criteria. Its primary function is to provide objective rewards by checking multiple rubric items in a single generation, making it suitable for automated evaluation of medical conversations.

Key Capabilities

  • Automated Rubric Grading: Evaluates medical assistant responses against a set of numbered criteria, each with assigned points.
  • Boolean Output: Returns a JSON list of booleans, indicating whether each criterion is met or not.
  • Negative Criteria Handling: Accurately identifies and flags undesirable behaviors (negative criteria) in responses.
  • OnePO Integration: Designed to integrate with the OnePO framework for calculating raw_score and training_reward based on rubric evaluations.
  • Batch Scoring: Supports efficient batch processing of multiple conversations and their corresponding rubrics for evaluation.

Usage and Integration

This model utilizes a precise grading template, requiring specific formatting for conversation and rubric inputs. It is designed for use with the transformers library, allowing developers to load the model and tokenizer for inference. The output is a clear, machine-readable JSON array, facilitating programmatic assessment and reward calculation in medical dialogue systems.

Finetunes
1 models