diff --git a/src/lighteval/evaluator.py b/src/lighteval/evaluator.py
index c547bc1ad..0b09ff23f 100644
--- a/src/lighteval/evaluator.py
+++ b/src/lighteval/evaluator.py
@@ -5,6 +5,8 @@
 import copy
 from typing import Dict, Union
 
+from pytablewriter import LatexTableWriter, MarkdownTableWriter
+
 from lighteval.logging.evaluation_tracker import EvaluationTracker
 from lighteval.logging.hierarchical_logger import hlog
 from lighteval.models.base_model import BaseModel
@@ -99,8 +101,6 @@ def evaluate(  # noqa: C901
 
 def make_results_table(result_dict):
     """Generate table of results."""
-    from pytablewriter import LatexTableWriter, MarkdownTableWriter
-
     md_writer = MarkdownTableWriter()
     latex_writer = LatexTableWriter()
     md_writer.headers = ["Task", "Version", "Metric", "Value", "", "Stderr"]
diff --git a/src/lighteval/models/nanotron_model.py b/src/lighteval/models/nanotron_model.py
index 51e682eb6..e6a3223d7 100644
--- a/src/lighteval/models/nanotron_model.py
+++ b/src/lighteval/models/nanotron_model.py
@@ -32,6 +32,7 @@
     LoglikelihoodDataset,
     LoglikelihoodSingleTokenDataset,
 )
+from lighteval.models.base_model import LightevalModel
 from lighteval.models.model_output import Batch, GenerateReturn, LoglikelihoodReturn, LoglikelihoodSingleTokenReturn
 from lighteval.tasks.requests import (
     GreedyUntilRequest,
@@ -51,7 +52,7 @@
 # _DeviceMapping = NewType("DeviceMapping", Mapping[str, Union[int, str, torch.device]])
 
 
-class NanotronLightevalModel:
+class NanotronLightevalModel(LightevalModel):
     # Default max sequence length setting for when no `max_length` is provided
     # or no max length config setting is found in the model or tokenizer.
     _DEFAULT_MAX_LENGTH: int = 2048
@@ -77,7 +78,6 @@ def __init__(
         """Initializes a nanotron model for evaluation.
         Args:
         """
-        super().__init__()
 
         self._batch_size = batch_size
         self._max_gen_toks = max_gen_toks
@@ -117,12 +117,12 @@ def __init__(
             self.model_config.num_hidden_layers = 1
 
         self._add_special_tokens = add_special_tokens
-        self.tokenizer = self._create_auto_tokenizer(
+        self._tokenizer = self._create_auto_tokenizer(
             pretrained=tokenizer.tokenizer_name_or_path,
             cache_dir=cache_dir,
             trust_remote_code=trust_remote_code,
         )
-        self.tokenizer.model_max_length = self.max_length
+        self._tokenizer.model_max_length = self.max_length
 
         model_config_cls = self.model_config.__class__.__name__
         if model_class is not None:
@@ -197,6 +197,10 @@ def __init__(
 
         self.multichoice_continuations_start_space = multichoice_continuations_start_space
 
+    @property
+    def tokenizer(self):
+        return self._tokenizer
+
     def _create_auto_tokenizer(
         self,
         *,
diff --git a/tasks_examples/custom_tasks/lighteval_config_override_template.yaml b/tasks_examples/custom_tasks/lighteval_config_override_template.yaml
index da9258ff9..6544a88af 100644
--- a/tasks_examples/custom_tasks/lighteval_config_override_template.yaml
+++ b/tasks_examples/custom_tasks/lighteval_config_override_template.yaml
@@ -12,7 +12,7 @@ lighteval:
     push_results_to_tensorboard: true
     tensorboard_metric_prefix: e
   parallelism:
-    dp: 8
+    dp: 1
     pp: 1
     pp_engine: 1f1b
     recompute_granularity: null
@@ -20,7 +20,7 @@ lighteval:
     tp_linear_async_communication: false
     tp_mode: ALL_REDUCE
   tasks:
-    custom_tasks_file: /fsx/thomwolf/github/lighteval-harness/tasks_examples/custom_evaluation_tasks.py
+    custom_tasks_file: /fsx/thomwolf/github/lighteval/tasks_examples/custom_tasks/custom_evaluation_tasks.py
     dataset_loading_processes: 8
     max_samples: 1000
     multichoice_continuations_start_space: null