diff --git a/src/instructlab/eval/mt_bench_answers.py b/src/instructlab/eval/mt_bench_answers.py
index 91333d29..79fc1287 100644
--- a/src/instructlab/eval/mt_bench_answers.py
+++ b/src/instructlab/eval/mt_bench_answers.py
@@ -119,10 +119,10 @@ def generate_answers(
     data_base_dir = bench_dir(data_dir, bench_name, branch)
     output_base_dir = bench_dir(output_dir, bench_name, branch)
 
-    question_file = f"{data_base_dir}/question.jsonl"
+    question_file = os.path.join(data_base_dir, "question.jsonl")
     questions = load_questions(question_file, question_begin, question_end)
 
-    answer_file = f"{output_base_dir}/model_answer/{model_name}.jsonl"
+    answer_file = os.path.join(output_base_dir, "model_answer", f"{model_name}.jsonl")
     if os.path.isfile(answer_file):
         os.remove(answer_file)
         logger.debug("Removing previous answer file: %s", answer_file)
diff --git a/src/instructlab/eval/mt_bench_branch_generator.py b/src/instructlab/eval/mt_bench_branch_generator.py
index ed9a4889..76bec8d0 100644
--- a/src/instructlab/eval/mt_bench_branch_generator.py
+++ b/src/instructlab/eval/mt_bench_branch_generator.py
@@ -112,16 +112,16 @@ def generate(judge_model_name, branch, taxonomy_dir, output_dir):
 
         output_base_dir = bench_dir(output_dir, "mt_bench_branch", branch)
         os.makedirs(output_base_dir, exist_ok=True)
-        question_fn = "question.jsonl"
-        question_file = os.path.join(output_base_dir, question_fn)
+        question_file = os.path.join(output_base_dir, "question.jsonl")
         logger.debug("Generating question file: %s", question_file)
         with open(question_file, "w", encoding="utf-8") as outfile:
             for entry in question_lst:
                 json.dump(entry, outfile)
                 outfile.write("\n")
 
-        answer_file_dir = os.path.join(output_base_dir, "reference_answer")
-        answer_file = os.path.join(answer_file_dir, f"{judge_model_name}.jsonl")
+        answer_file = os.path.join(
+            output_base_dir, "reference_answer", f"{judge_model_name}.jsonl"
+        )
         logger.debug("Generating answer file: %s", answer_file)
         os.makedirs(os.path.dirname(answer_file), exist_ok=True)
         with open(
diff --git a/src/instructlab/eval/mt_bench_common.py b/src/instructlab/eval/mt_bench_common.py
index ee23546d..7c13e2c4 100644
--- a/src/instructlab/eval/mt_bench_common.py
+++ b/src/instructlab/eval/mt_bench_common.py
@@ -71,7 +71,7 @@ class MatchSingle:
 
 
 def bench_dir(output_dir, bench_name, branch) -> str:
-    b_dir = f"{output_dir}/{bench_name}"
+    b_dir = os.path.join(output_dir, bench_name)
     if branch is not None:
         b_dir = os.path.join(b_dir, branch)
     return b_dir
@@ -283,7 +283,7 @@ def _get_messages(
         and messages[0]["role"] == "system"
         and messages[1]["role"] == "user"
     ):
-        messages[1]["content"] = messages[0]["content"] + "\n" + messages[1]["content"]
+        messages[1]["content"] = f"{messages[0]['content']}\n{messages[1]['content']}"
         return messages[1:]
     return messages
 
diff --git a/src/instructlab/eval/mt_bench_judgment.py b/src/instructlab/eval/mt_bench_judgment.py
index f7b40b89..a4893c34 100644
--- a/src/instructlab/eval/mt_bench_judgment.py
+++ b/src/instructlab/eval/mt_bench_judgment.py
@@ -169,12 +169,12 @@ def judge_model(
     data_base_dir = bench_dir(data_dir, bench_name, branch)
     output_base_dir = bench_dir(output_dir, bench_name, branch)
 
-    judge_file = f"{package_data_dir}/{bench_name}/judge_prompts.jsonl"
+    judge_file = os.path.join(package_data_dir, bench_name, "judge_prompts.jsonl")
 
-    question_file = f"{data_base_dir}/question.jsonl"
-    answer_file = f"{output_base_dir}/model_answer/{model_name}.jsonl"
+    question_file = os.path.join(data_base_dir, "question.jsonl")
+    answer_file = os.path.join(output_base_dir, "model_answer", f"{model_name}.jsonl")
     answer_dir = os.path.dirname(answer_file)
-    ref_answer_dir = f"{data_base_dir}/reference_answer"
+    ref_answer_dir = os.path.join(data_base_dir, "reference_answer")
 
     # Load questions
     questions = load_questions(question_file, None, None)
@@ -195,7 +195,9 @@ def judge_model(
         models = model_list
 
     judges = make_judge_single(judge_model_name, judge_prompts)
-    output_file = f"{output_base_dir}/model_judgment/{judge_model_name}_single.jsonl"
+    output_file = os.path.join(
+        output_base_dir, "model_judgment", f"{judge_model_name}_single.jsonl"
+    )
     if os.path.isfile(output_file):
         os.remove(output_file)
         logger.debug("Removing previous judgment file: %s", output_file)