Removing model_types from utils.py

[dougian]

Hey, looks like the library is currently broken when installed with the latest versions of everything.

Since the latest release in transformers (more specifically, since this commit huggingface/transformers@d4c2cb4) the config maps (pretrained_config_archive_map) do not exist.

This means that bertscore right now fails when used with the latest version of transformers, which it will pull when installing locally.

e.g, when running the unittests:

ERROR: bert_score (unittest.loader._FailedTest)

ImportError: Failed to import test module: bert_score
Traceback (most recent call last):
File "/home/ec2-user/anaconda3/lib/python3.6/unittest/loader.py", line 462, in _find_test_path
package = self._get_module_from_name(name)
File "/home/ec2-user/anaconda3/lib/python3.6/unittest/loader.py", line 369, in _get_module_from_name
import(name)
File "/home/ec2-user/useful_code/bert_score/bert_score/init.py", line 2, in
from .utils import *
File "/home/ec2-user/useful_code/bert_score/bert_score/utils.py", line 32, in
+ list(SCIBERT_URL_DICT.keys())
AttributeError: type object 'BertConfig' has no attribute 'pretrained_config_archive_map'

Solution

SInce this attribute is no longer being used anywhere, the easy fix here is to remove the map from utils.py as everything is migrated to AutoTokenizer anyway.

[Tiiiger]

Good catch. Looks good to me.