Skip to main content

PyTorch implementation of BERT score

Project description


Automatic Evaluation Metric described in the paper BERTScore: Evaluating Text Generation with BERT.


  • Updated to version 0.2.1
    • SciBERT (Beltagy et al.) models are now included. Thanks to AI2 for sharing the models. By default, we use the 9th layer (the same as BERT-base), but this is not tuned.
  • Our arXiv paper has been updated to v2 with more experiments and analysis.
  • Updated to version 0.2.0
    • Supporting BERT, XLM, XLNet, and RoBERTa models using huggingface's Transformers library
    • Automatically picking the best model for a given language
    • Automatically picking the layer based a model
    • IDF is not set as default as we show in the new version that the improvement brought by importance weighting is not consistent


*: Equal Contribution


BERTScore leverages the pre-trained contextual embeddings from BERT and matches words in candidate and reference sentences by cosine similarity. It has been shown to correlate with human judgment on setence-level and system-level evaluation. Moreover, BERTScore computes precision, recall, and F1 measure, which can be useful for evaluating different language generation tasks.

For an illustration, BERTScore precision can be computed as

If you find this repo useful, please cite:

  title={BERTScore: Evaluating Text Generation with BERT},
  author={Zhang, Tianyi and Kishore, Varsha and Wu, Felix and Weinberger, Kilian Q. and Artzi, Yoav.},
  journal={arXiv preprint arXiv:1904.09675},


  • Python version >= 3.6
  • PyTorch version >= 1.0.0

Install from pip by

pip install bert-score

Install it from the source by:

git clone
cd bert_score
pip install .

and you may test your installation by:

python -m unittest discover


Command Line Interface (CLI)

We provide a command line interface (CLI) of BERTScore as well as a python module. For the CLI, you can use it as follows:

  1. To evaluate English text files:

We provide example inputs under ./example.

bert-score -r example/refs.txt -c example/hyps.txt --lang en

You will get the following output at the end:

roberta-large_L17_no-idf_version=0.2.1 BERT-P: 0.950530 BERT-R: 0.949223 BERT-F1: 0.949839

where "roberta-large_L17_no-idf_version=0.2.1" is the hashcode.

  1. To evaluate text files in other languages:

We currently support the 104 languages in multilingual BERT (full list).

Please specify the two-letter abbrevation of the language. For instance, using --lang zh for Chinese text.

See more options by bert-score -h.

  1. To visualize matching scores:
bert-score-show --lang en -r "There are two bananas on the table." -c "On the table are two apples." -f out.png

The figure will be saved to out.png.

Python Function

For the python module, we provide a demo. Please refer to bert_score/ for more details.

Running BERTScore can be computationally intensive (because it uses BERT :p). Therefore, a GPU is usually necessary. If you don't have access to a GPU, you can try our demo on Google Colab

Practical Tips

  • Report the hash code (e.g., roberta-large_L17_no-idf_version=0.2.1) in your paper so that people know what setting you use. This is inspired by sacreBLEU.
  • Unlike BERT, RoBERTa uses GPT2-style tokenizer which creates addition " " tokens when there are multiple spaces appearing together. It is recommended to remove addition spaces by sent = re.sub(r' +', ' ', sent) or sent = re.sub(r'\s+', ' ', sent).
  • Using inverse document frequency (idf) on the reference sentences to weigh word importance may correlate better with human judgment. However, when the set of reference sentences become too small, the idf score would become inaccurate/invalid. We now make it optional. To use idf, please set --idf when using the CLI tool or idf=True when calling bert_score.score function.
  • When you are low on GPU memory, consider setting batch_size when calling bert_score.score function.
  • To use a particular model please set -m MODEL_TYPE when using the CLI tool or model_type=MODEL_TYPE when calling bert_score.score function.
  • We tune layer to use based on WMT16 metric evaluation dataset. You may use a different layer by setting -l LAYER or num_layers=LAYER
  • Limitation: Because BERT, RoBERTa, and XLM with learned positional embeddings are pre-trained on sentences with max length 512, BERTScore is undefined between sentences longer than 510 (512 after adding [CLS] and [SEP] tokens). The sentences longer than this will be truncated. Please consider using XLNet which can support much longer inputs.

Default Behavior

Default Model

Language Model
en roberta-large
en-sci scibert-scivocab-uncased
zh bert-base-chinese
others bert-base-multilingual-cased

Default Layers

Model Best Layer Max Length
bert-base-uncased 9 512
bert-large-uncased 18 512
bert-base-cased-finetuned-mrpc 9 512
bert-base-multilingual-cased 9 512
bert-base-chinese 8 512
roberta-base 10 512
roberta-large 17 512
roberta-large-mnli 19 512
xlnet-base-cased 5 1000000000000
xlnet-large-cased 7 1000000000000
xlm-mlm-en-2048 7 512
xlm-mlm-100-1280 11 512
scibert-scivocab-uncased 9* 512
scibert-scivocab-cased 9* 512
scibert-basevocab-uncased 9* 512
scibert-basevocab-cased 9* 512

*: Not tuned


This repo wouldn't be possible without the awesome bert, fairseq, and transformers.

Project details

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page