Extend to transformers v4.25 (#360)

* Extend to transformers v4.25

* Update serialization test for transformers v4.25

Author: adrianeboyd

requirements.txt

@@ -1,5 +1,5 @@
 spacy>=3.4.0,<4.0.0
-transformers>=3.4.0,<4.25.0
+transformers>=3.4.0,<4.26.0
 torch>=1.6.0
 srsly>=2.4.0,<3.0.0
 dataclasses>=0.6,<1.0; python_version < "3.7"

setup.cfg

@@ -32,7 +32,7 @@ include_package_data = true
 python_requires = >=3.6
 install_requires =
     spacy>=3.4.0,<4.0.0
-    transformers>=3.4.0,<4.25.0
+    transformers>=3.4.0,<4.26.0
     torch>=1.6.0
     srsly>=2.4.0,<3.0.0
     dataclasses>=0.6,<1.0; python_version < "3.7"

spacy_transformers/tests/test_serialize.py

@@ -116,9 +116,12 @@ def test_transformer_pipeline_todisk_settings():
     assert trf.model.transformer.config.output_attentions is False
     assert "attentions" not in nlp("test")._.trf_data.model_output
     # modify model_max_length (note that modifications to
-    # tokenizer.model_max_length are not serialized by save_pretrained
-    # see: https://github.com/explosion/spaCy/discussions/7393)
+    # tokenizer.model_max_length for transformers<4.25 are not serialized by
+    # save_pretrained, see: https://github.com/explosion/spaCy/discussions/7393)
     trf.model.tokenizer.init_kwargs["model_max_length"] = 499
+    # transformer>=4.25, model_max_length is saved and init_kwargs changes are
+    # clobbered, so do both for this test
+    trf.model.tokenizer.model_max_length = 499
     # add attentions on-the-fly
     trf.model.transformer.config.output_attentions = True
     assert nlp("test")._.trf_data.model_output.attentions is not None