Add results file for gpt-oss-120b

results/openai/gpt-oss-120b_results_2025-08-07_13-48-14.json

@@ -0,0 +1,153 @@
+{
+  "results": {
+    "average_score": 7.819148936170213,
+    "speed": 7.714403817867055,
+    "contamination_score": 0,
+    "execution_time": 11656.247472,
+    "errors": [],
+    "scores_by_category": [
+      {
+        "category": "Paraphrasing",
+        "average_score": 10.0,
+        "count": 6
+      },
+      {
+        "category": "Coding",
+        "average_score": 10.0,
+        "count": 3
+      },
+      {
+        "category": "Reading Comprehension",
+        "average_score": 9.411764705882353,
+        "count": 17
+      },
+      {
+        "category": "MMLU",
+        "average_score": 9.256198347107437,
+        "count": 121
+      },
+      {
+        "category": "Entity Extraction",
+        "average_score": 9.0,
+        "count": 5
+      },
+      {
+        "category": "Sentiment Analysis",
+        "average_score": 8.88888888888889,
+        "count": 9
+      },
+      {
+        "category": "Instruction Following",
+        "average_score": 8.714285714285714,
+        "count": 7
+      },
+      {
+        "category": "General Knowledge",
+        "average_score": 8.46031746031746,
+        "count": 63
+      },
+      {
+        "category": "Function Calling",
+        "average_score": 8.333333333333334,
+        "count": 3
+      },
+      {
+        "category": "Reasoning & Math",
+        "average_score": 8.162790697674419,
+        "count": 43
+      },
+      {
+        "category": "Writing (incl Dialects)",
+        "average_score": 7.909090909090909,
+        "count": 22
+      },
+      {
+        "category": "Long Context",
+        "average_score": 7.5,
+        "count": 4
+      },
+      {
+        "category": "Translation (incl Dialects)",
+        "average_score": 7.083333333333333,
+        "count": 36
+      },
+      {
+        "category": "Hallucination",
+        "average_score": 6.666666666666667,
+        "count": 3
+      },
+      {
+        "category": "Trust & Safety",
+        "average_score": 6.666666666666667,
+        "count": 30
+      },
+      {
+        "category": "Summarization",
+        "average_score": 6.625,
+        "count": 8
+      },
+      {
+        "category": "RAG QA",
+        "average_score": 5.975609756097561,
+        "count": 41
+      },
+      {
+        "category": "Dialect Detection",
+        "average_score": 5.818181818181818,
+        "count": 11
+      },
+      {
+        "category": "Transliteration",
+        "average_score": 5.666666666666667,
+        "count": 6
+      },
+      {
+        "category": "Structuring",
+        "average_score": 5.333333333333333,
+        "count": 3
+      },
+      {
+        "category": "Arabic Language & Grammar",
+        "average_score": 5.294117647058823,
+        "count": 17
+      },
+      {
+        "category": "Diacritization",
+        "average_score": 2.4166666666666665,
+        "count": 12
+      }
+    ],
+    "scores_by_format": [
+      {
+        "format": "Short Answer",
+        "average_score": 10.0,
+        "count": 5
+      },
+      {
+        "format": "MCQ",
+        "average_score": 8.724890829694322,
+        "count": 229
+      },
+      {
+        "format": "Generation",
+        "average_score": 6.916666666666667,
+        "count": 228
+      },
+      {
+        "format": "Fill-in-the-blank",
+        "average_score": 6.25,
+        "count": 8
+      }
+    ]
+  },
+  "config": {
+    "model": "openai/gpt-oss-120b",
+    "model_sha": "eabf0c518da7584a2e7dab4ab272709785a72126",
+    "submitted_time": "2025-08-07T09:27:04Z",
+    "likes": 2639,
+    "params": 120.412,
+    "license": "apache-2.0",
+    "model_source": "Hugging Face",
+    "model_category": "Large"
+  }
+}