diff --git a/deepmultilingualpunctuation/__init__.py b/deepmultilingualpunctuation/__init__.py
index 5919050..a285e41 100644
--- a/deepmultilingualpunctuation/__init__.py
+++ b/deepmultilingualpunctuation/__init__.py
@@ -1 +1 @@
-from .punctuationmodel import PunctuationModel
\ No newline at end of file
+from .punctuationmodel import PunctuationModel
diff --git a/deepmultilingualpunctuation/punctuationmodel.py b/deepmultilingualpunctuation/punctuationmodel.py
index 14d11fa..e2b811b 100644
--- a/deepmultilingualpunctuation/punctuationmodel.py
+++ b/deepmultilingualpunctuation/punctuationmodel.py
@@ -1,79 +1,90 @@
-from concurrent.futures import process
-from transformers import pipeline
 import re
+
 import torch
+from transformers import pipeline
+
+
+class PunctuationModel:
+    def __init__(
+        self,
+        model="oliverguhr/fullstop-punctuation-multilang-large",
+        **kwargs,
+    ) -> None:
+        if "aggregation_strategy" not in kwargs:
+            kwargs["aggregation_strategy"] = "none"
+        self.pipe = pipeline("ner", model, **kwargs)
 
-class PunctuationModel():
-    def __init__(self, model = "oliverguhr/fullstop-punctuation-multilang-large") -> None:        
-        if torch.cuda.is_available():
-            self.pipe = pipeline("ner",model, aggregation_strategy="none", device=0)
-        else:
-            self.pipe = pipeline("ner",model, aggregation_strategy="none")        
-
-    def preprocess(self,text):
-        #remove markers except for markers in numbers 
-        text = re.sub(r"(?<!\d)[.,;:!?](?!\d)","",text) 
-        #todo: match acronyms https://stackoverflow.com/questions/35076016/regex-to-match-acronyms
+    def preprocess(self, text):
+        # remove punctuation except dots
+        text = re.sub(r"[,;:!?]", "", text)
+        # remove dots that are not in acronyms or decimal points
+        text = re.sub(r"(?<!\b[a-zA-Z])(?<!\d)\.", "", text)
         text = text.split()
         return text
 
-    def restore_punctuation(self, text, chunk_size=230):        
+    def restore_punctuation(self, text, chunk_size=230):
         result = self.predict(self.preprocess(text), chunk_size)
         return self.prediction_to_text(result)
-        
-    def overlap_chunks(self,lst, n, stride=0):
+
+    def overlap_chunks(self, lst, n, stride=0):
         """Yield successive n-sized chunks from lst with stride length of overlap."""
-        for i in range(0, len(lst), n-stride):
-                yield lst[i:i + n]
+        for i in range(0, len(lst), n - stride):
+            yield lst[i : i + n]
 
     def predict(self, words, chunk_size=230):
         overlap = 5
         if len(words) <= chunk_size:
             overlap = 0
 
-        batches = list(self.overlap_chunks(words,chunk_size,overlap))
+        batches = list(self.overlap_chunks(words, chunk_size, overlap))
 
-        # if the last batch is smaller than the overlap, 
+        # if the last batch is smaller than the overlap,
         # we can just remove it
         if len(batches[-1]) <= overlap:
             batches.pop()
 
-        tagged_words = []     
+        tagged_words = []
         for batch in batches:
             # use last batch completely
-            if batch == batches[-1]: 
+            if batch == batches[-1]:
                 overlap = 0
             text = " ".join(batch)
-            result = self.pipe(text)      
-            assert len(text) == result[-1]["end"], "chunk size too large, text got clipped"
-                
+            result = self.pipe(text)
+            assert (
+                len(text) == result[-1]["end"]
+            ), "chunk size too large, text got clipped"
+
             char_index = 0
             result_index = 0
-            for word in batch[:len(batch)-overlap]:                
+            for word in batch[: len(batch) - overlap]:
                 char_index += len(word) + 1
                 # if any subtoken of an word is labled as sentence end
-                # we label the whole word as sentence end        
+                # we label the whole word as sentence end
                 label = "0"
-                while result_index < len(result) and char_index > result[result_index]["end"] :
-                    label = result[result_index]['entity']
-                    score = result[result_index]['score']
-                    result_index += 1                        
-                tagged_words.append([word,label, score])
-        
+                while (
+                    result_index < len(result)
+                    and char_index > result[result_index]["end"]
+                ):
+                    label = result[result_index]["entity"]
+                    score = result[result_index]["score"]
+                    result_index += 1
+                tagged_words.append([word, label, score])
+
         assert len(tagged_words) == len(words)
         return tagged_words
 
-    def prediction_to_text(self,prediction):
+    def prediction_to_text(self, prediction):
         result = ""
         for word, label, _ in prediction:
             result += word
             if label == "0":
                 result += " "
             if label in ".,?-:":
-                result += label+" "
+                result += label + " "
         return result.strip()
 
-if __name__ == "__main__":    
+
+if __name__ == "__main__":
     model = PunctuationModel()
 
     text = "das , ist fies "
diff --git a/setup.py b/setup.py
index f5e01a0..fc0be03 100644
--- a/setup.py
+++ b/setup.py
@@ -12,15 +12,15 @@
     long_description=long_description,
     long_description_content_type="text/markdown",
     url="https://github.com/oliverguhr/deepmultilingualpunctuation",
-    packages=setuptools.find_packages(),    
+    packages=setuptools.find_packages(),
     classifiers=[
         "Programming Language :: Python :: 3",
         "License :: OSI Approved :: MIT License",
         "Operating System :: OS Independent",
     ],
     install_requires=[
-       "transformers",
-       "torch>=1.8.1",
+        "transformers",
+        "torch>=1.8.1",
     ],
-    python_requires='>=3.6',
+    python_requires=">=3.6",
 )