TIGER-Lab/TIGERScore-13B
TEXT GENERATIONConcurrency Cost:1Model Size:13BQuant:FP8Ctx Length:4kPublished:Nov 26, 2023License:mitArchitecture:Transformer0.0K Open Weights Cold

TIGER-Lab/TIGERScore-13B is a 13 billion parameter LLaMA-2 based model developed by TIGER-Lab, fine-tuned as an instruction-guided, reference-free metric for evaluating text generation tasks. It provides explainable error analysis, pinpointing mistakes with location, aspect, explanation, and penalty scores. This model excels at correlating with human ratings across various text generation tasks, surpassing many existing reference-based and reference-free metrics.

Loading preview...