diff --git a/.github/workflows/staging-model.yml b/.github/workflows/staging-model.yml
index a02cd77c..6b005004 100644
--- a/.github/workflows/staging-model.yml
+++ b/.github/workflows/staging-model.yml
@@ -35,8 +35,6 @@ jobs:
         predict.src_las=/data/val/*.las
         predict.output_dir=/lidar_prod_optimization/${model_id}/preds-valset/
         predict.interpolator.probas_to_save=[building]
-        datamodule.prefetch_factor=1
-        datamodule.batch_size=2
 
     - name: Inference on test set of 20211001_building_val-lidar-prod-optimization
       run: >
@@ -51,8 +49,6 @@ jobs:
         predict.src_las=/data/test/*.las
         predict.output_dir=/lidar_prod_optimization/${model_id}/preds-testset/
         predict.interpolator.probas_to_save=[building]
-        datamodule.prefetch_factor=1
-        datamodule.batch_size=2
 
     # NB: we run the new, tagged image at least once so that is it not prunned
     # by mistake when using docker system prune.
diff --git a/trained_model_assets/20230930_60k_basic_targetted_epoch37_Myria3DV3.4.0_predict_config_V3.5.0.yaml b/trained_model_assets/20230930_60k_basic_targetted_epoch37_Myria3DV3.4.0_predict_config_V3.5.0.yaml
index 53618fb4..c7b94791 100644
--- a/trained_model_assets/20230930_60k_basic_targetted_epoch37_Myria3DV3.4.0_predict_config_V3.5.0.yaml
+++ b/trained_model_assets/20230930_60k_basic_targetted_epoch37_Myria3DV3.4.0_predict_config_V3.5.0.yaml
@@ -125,8 +125,8 @@ datamodule:
     _args_:
     - ${get_method:myria3d.pctl.dataset.utils.pre_filter_below_n_points}
     min_num_nodes: 1
-  tile_width: 50
-  subtile_width: 1000
+  tile_width: 5000
+  subtile_width: 50
   subtile_overlap_train: 0
   subtile_overlap_predict: ${predict.subtile_overlap}
   batch_size: 10