Remove spacy download step; use pip-installed model

categories/accuracy.py

@@ -10,7 +10,7 @@ from transformers import AutoModel, AutoTokenizer
 # setup global variables on import (bad practice, but whatever)
 # --------------------------------------------------------------
 
-aligner = SentenceAligner(model="bert-base-multilingual-cased", layer=6)
+aligner = SentenceAligner(model="xlm-roberta-base", layer=6)
 
 de_encoder = LaserEncoderPipeline(lang="deu_Latn")
 en_encoder = LaserEncoderPipeline(lang="eng_Latn")

categories/fluency.py

@@ -12,7 +12,7 @@ from transformers import AutoModelForMaskedLM, AutoTokenizer
 tool = language_tool_python.LanguageTool("en-US")
 
 # masked language model and tokenizer from huggingface
-model_name = "distilbert-base-multilingual-cased"
+model_name = "xlm-roberta-base"
 model = AutoModelForMaskedLM.from_pretrained(model_name)
 model.eval()
 tokenizer = AutoTokenizer.from_pretrained(model_name)  # tokenizer