feat: support gpu

processing/ner/name_model.py

@@ -29,6 +29,15 @@ class NameModel:
         """Create a blank spaCy model with NER pipeline"""
         logging.info(f"Creating blank {language} model for NER training")
 
+        # Prefer GPU for spaCy if available (falls back to CPU automatically)
+        try:
+            if spacy.prefer_gpu():
+                logging.info("spaCy GPU enabled (cupy) for NER training")
+            else:
+                logging.info("spaCy running on CPU")
+        except Exception as e:
+            logging.debug(f"spaCy GPU selection skipped: {e}")
+
         # Create blank model - French tokenizer works well for DRC names
         self.nlp = spacy.blank(language)
 

processing/steps/data_selection_step.py

@@ -20,11 +20,15 @@ class DataSelectionStep(PipelineStep):
         # Remove rows where region == "global" only for specific years
         if "region" in batch.columns and "year" in batch.columns:
             target_years = {2015, 2021, 2022}
-            mask_remove = batch["region"].str.lower().eq("global") & batch["year"].isin(target_years)
+            mask_remove = batch["region"].str.lower().eq("global") & batch["year"].isin(
+                target_years
+            )
             removed = int(mask_remove.sum())
             if removed:
                 batch = batch[~mask_remove]
-                logging.info(f"Removed {removed} rows with region == 'global' for years {sorted(target_years)} in batch {batch_id}")
+                logging.info(
+                    f"Removed {removed} rows with region == 'global' for years {sorted(target_years)} in batch {batch_id}"
+                )
 
         # Check which columns exist in the batch
         available_columns = [col for col in self.selected_columns if col in batch.columns]