use_custom_baseline still produces errors for bertscore

[penatbater]

metric = load_metric('bertscore')
a1 = "random sentences"
b1 = "random sentences"
metric.compute(predictions = [a1], references = [b1], lang = 'en')

Traceback (most recent call last): File "<stdin>", line 1, in <module> File "/home/stephen_chan/.local/lib/python3.6/site-packages/datasets/metric.py", line 393, in compute output = self._compute(predictions=predictions, references=references, **kwargs) File "/home/stephen_chan/.cache/huggingface/modules/datasets_modules/metrics/bertscore/361e597a01a41d6cf95d94bbfb01dea16261687abc0c6c74cc9930f80488f363/bertscore.py", line 108, in _compute hashcode = bert_score.utils.get_hash(model_type, num_layers, idf, rescale_with_baseline) TypeError: get_hash() missing 1 required positional argument: 'use_custom_baseline'

Adding 'use_custom_baseline = False' as an argument produces this error

Traceback (most recent call last): File "<stdin>", line 1, in <module> File "/home/stephen_chan/.local/lib/python3.6/site-packages/datasets/metric.py", line 393, in compute output = self._compute(predictions=predictions, references=references, **kwargs) TypeError: _compute() got an unexpected keyword argument 'use_custom_baseline'

This is on Ubuntu 18.04, Python 3.6.9, datasets version 1.1.2

[lhoestq]

Thanks for reporting ! That's a bug indeed
If you want to contribute, feel free to fix this issue and open a PR :)

[pvl]

This error is because of a mismatch between datasets and bert_score. With datasets=1.1.2 and bert_score>=0.3.6 it works ok. So pip install -U bert_score should fix the problem.

[lhoestq]

Thanks for the heads up @pvl and for the PR as well :)

[SkanderHellal]

Hello everyone,

I think the problem is not solved:

from datasets import load_metric
metric=load_metric('bertscore')
metric.compute(
    predictions=predictions,
    references=references,
    lang='fr',
    rescale_with_baseline=True
)
TypeError: get_hash() missing 2 required positional arguments: 'use_custom_baseline' and 'use_fast_tokenizer'

This code is produced using Python 3.6.9 datasets==1.1.2 and bert_score==0.3.10

[lhoestq]

Hi ! This has been fixed by #2770, we'll do a new release soon to make the fix available :)

In the meantime please use an older version of bert_score