Tiiiger/bert_score

Health v2 · maintenance only

Adoption not part of the score

1916 stars · 240 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

BERTScore is a PyTorch implementation of the BERTScore automatic evaluation metric for text generation, which scores generated text against references using contextual embeddings from pretrained models like BERT and RoBERTa. It supports ~130 pretrained models and correlates well with human evaluation.

Use cases

• evaluate machine translation output quality
• score text summarization against references
• compare generated text with human judgments
• compute semantic similarity between sentences
• evaluate LLM text generation quality
• benchmark NLP generation models

When to choose

• you need an automatic evaluation metric that correlates better with human judgment than BLEU or ROUGE
• you are evaluating text generation tasks like summarization or translation
• you want a PyTorch-based metric with broad pretrained model support

When to avoid

• you need a lightweight metric without GPU or large model downloads
• you need exact n-gram overlap scores like BLEU for reporting
• you need a metric actively adding new features rather than maintenance fixes

Facets

library · maturity maintenance

nlp machine-learning benchmarking machine-learning python bert evaluation-metric text-generation pytorch transformers natural-language-processing

Member repositories

For agents

markdown · JSON · MCP: product_card(name="Tiiiger/bert_score")