Update library version to 0.4.0. Fix prediction.

requirements-gpu.txt

 torch==1.13.0
 --index-url https://pypi.clarin-pl.eu/simple/ 
 nlp_ws
-winer[torch-gpu]==0.3.1
+winer==0.4.0
 awscli==1.22.57
\ No newline at end of file

requirements.txt

 --index-url https://pypi.clarin-pl.eu/simple/ 
 nlp_ws
-winer==0.3.1
+winer==0.4.0
 awscli==1.22.57
\ No newline at end of file

src/winer_worker.py

 """Implementation of punctuator service"""
 
-from winer.datafiles import read_clarin_json, write_clarin_json
-from winer.document import create_document_from_clarin_json, \
-    create_entities_from_hf_outputs
-from winer.winer import Winer
+from winer import Winer
+from winer import allign_into_entities, allign_into_words
+from winer import create_document_from_clarin_json, create_entities_from_hf_outputs
+from winer import read_clarin_json, write_clarin_json
 
 import logging
 
@@ -16,7 +16,7 @@ class WinerWorker:
     ):
 
         logging.info("Loading models...")
-        self.active_model = Winer(models_location)
+        self._model = Winer(models_location)
 
     def process(
         self,
@@ -24,12 +24,25 @@ class WinerWorker:
         task_options: dict,
         output_path: str
     ) -> None:
+        # Read inputs
         documents = [create_document_from_clarin_json(read_clarin_json(input_path))]
-        outputs = self.active_model.predict(
-            [str(document) for document in documents]
-        )
+        tok_inputs = [document.get_pretokenized_text() for document in documents]
+        plain_inputs = [document.get_text() for document in documents]
+
+        # Predicion
+        encoded_inputs = self._model.tokenize(tok_inputs)
+        predictions_per_token = self._model.predict(encoded_inputs)
+
+        # Aggregation
+        aggregations = [allign_into_words(tokens, labels)
+                        for tokens, labels in zip(
+                            [encoded_inputs.word_ids(index)
+                             for index in range(len(predictions_per_token))],
+                            predictions_per_token)]
+        entities = allign_into_entities(tok_inputs, aggregations,
+                                        inputs_sentences=plain_inputs)
 
         for idx in range(len(documents)):
-            documents[idx].add_entites(create_entities_from_hf_outputs(outputs[idx]))
+            documents[idx].add_entites(create_entities_from_hf_outputs(entities[idx]))
 
         write_clarin_json(documents[0].as_clarin_json(), output_path)