Merge pull request #185 from danmcp/answersunittests

Add reorg answer file test

Author: mergify[bot]

src/instructlab/eval/mt_bench_answers.py

@@ -26,16 +26,19 @@
 def reorg_answer_file(answer_file):
     """Sort by question id and de-duplication"""
     logger.debug(locals())
-    answers = {}
-    with open(answer_file, "r", encoding="utf-8") as fin:
-        for l in fin:
+    with open(answer_file, "r+", encoding="utf-8") as f:
+        answers = {}
+        for l in f:
             qid = json.loads(l)["question_id"]
             answers[qid] = l
 
-    qids = sorted(list(answers.keys()))
-    with open(answer_file, "w", encoding="utf-8") as fout:
+        # Reset to the beginning of the file and clear it
+        f.seek(0)
+        f.truncate()
+
+        qids = sorted(list(answers.keys()))
         for qid in qids:
-            fout.write(answers[qid])
+            f.write(answers[qid])
 
 
 def get_answer(

tests/test_mt_bench_answers.py

@@ -0,0 +1,65 @@
+# SPDX-License-Identifier: Apache-2.0
+
+# Standard
+import json
+import os
+import random
+import shutil
+import tempfile
+
+# First Party
+from instructlab.eval.mt_bench_answers import reorg_answer_file
+
+
+def test_reorg_answer_file():
+    answer_file = os.path.join(
+        os.path.dirname(__file__),
+        "..",
+        "src",
+        "instructlab",
+        "eval",
+        "data",
+        "mt_bench",
+        "reference_answer",
+        "gpt-4.jsonl",
+    )
+
+    # Create a temporary file
+    with tempfile.NamedTemporaryFile(delete=True) as temp_file:
+        temp_answer_file = temp_file.name
+
+        # Copy the original file to the temp file
+        shutil.copy(answer_file, temp_answer_file)
+
+        orig_length = 0
+        with open(temp_answer_file, "r+", encoding="utf-8") as f:
+            answers = {}
+            for l in f:
+                orig_length += 1
+                qid = json.loads(l)["question_id"]
+                answers[qid] = l
+
+            # Reset to the beginning of the file and clear it
+            f.seek(0)
+            f.truncate()
+
+            # Randomize the values
+            qids = sorted(list(answers.keys()), key=lambda answer: random.random())
+            for qid in qids:
+                f.write(answers[qid])
+                # Write each answer twice
+                f.write(answers[qid])
+
+        # Run the reorg which should sort and dedup the file in place
+        reorg_answer_file(temp_answer_file)
+
+        new_length = 0
+        with open(temp_answer_file, "r", encoding="utf-8") as fin:
+            previous_question_id = -1
+            for l in fin:
+                new_length += 1
+                qid = json.loads(l)["question_id"]
+                assert qid > previous_question_id
+                previous_question_id = qid
+
+        assert new_length == orig_length