opencompass

/

CompassJudger-2-7B-Instruct

like 2

Follow

OpenCompass 133

This PR updates the model card to improve its accuracy and usability:

1. Metadata Update: The pipeline_tag has been changed from text-generation to text-ranking. This more precisely reflects the model's primary function as a judge model designed for evaluating and comparing (ranking) responses from other language models. This update will enhance the model's discoverability on the Hugging Face Hub, allowing users to easily find it when filtering for text-ranking capabilities (e.g., via https://huggingface.co/models?pipeline_tag=text-ranking).

2. Quickstart Code Fix: A minor bug in the Python quickstart example has been resolved. The variables model_name were replaced with model_path in the AutoModelForCausalLM.from_pretrained and AutoTokenizer.from_pretrained calls. This ensures the sample code is directly runnable and correctly loads the model, preventing a NameError for users trying to replicate the example.

These changes contribute to a more accurate and user-friendly model card.