diff --git a/src/cerebras/modelzoo/data_preparation/data_preprocessing/finetuning_token_generator.py b/src/cerebras/modelzoo/data_preparation/data_preprocessing/finetuning_token_generator.py
index 4d66c287..7495397d 100644
--- a/src/cerebras/modelzoo/data_preparation/data_preprocessing/finetuning_token_generator.py
+++ b/src/cerebras/modelzoo/data_preparation/data_preprocessing/finetuning_token_generator.py
@@ -70,13 +70,15 @@ def __init__(self, params, tokenizer, eos_id, pad_id):
             )
             self.tokenizer.chat_template = default_chat_template()
 
-        self.eos_id = eos_id
-        self.eos_token = (
-            self.tokenizer.convert_ids_to_tokens(self.pad_id)
-            if self.eos_id is None
-            else self.tokenizer.convert_ids_to_tokens(self.eos_id)
-        )
-        self.pad_id = pad_id
+            self.pad_id = pad_id
+            # Assign pad_id first so it's available for use below
+            self.eos_id = eos_id
+            # Set EOS token, using pad_id as fallback if eos_id is None
+            self.eos_token = (
+                self.tokenizer.convert_ids_to_tokens(self.pad_id)
+                if self.eos_id is None
+                else self.tokenizer.convert_ids_to_tokens(self.eos_id)
+            )
         self.features = ["input_ids", "attention_mask", "labels"]
         self.semantic_loss_weight = processing_params.pop(
             "semantic_loss_weight", {}
diff --git a/src/cerebras/modelzoo/data_preparation/vision/dit/vae_hf_cs.py b/src/cerebras/modelzoo/data_preparation/vision/dit/vae_hf_cs.py
index ddf366ee..cb249ca5 100644
--- a/src/cerebras/modelzoo/data_preparation/vision/dit/vae_hf_cs.py
+++ b/src/cerebras/modelzoo/data_preparation/vision/dit/vae_hf_cs.py
@@ -14,6 +14,8 @@
 
 # isort: off
 import sys
+# ensure os is available before it's used below
+import os
 
 sys.path.append(os.path.join(os.path.dirname(__file__), "../../../"))
 # isort: on