External models: Sabia-3 e Llama-3.1-405b

external_models_results.json

@@ -241,5 +241,50 @@
         },
         "result_metrics_average": 0.7777870380406591,
         "result_metrics_npm": 0.6740728488043128
+    },
+    {
+        "model": "llama_405b_instruct",
+        "name": "meta-llama/Meta-Llama-3.1-405B-Instruct (Vertex AI)",
+        "link": "https://cloud.google.com/vertex-ai",
+        "date": "2024-08-20",
+        "status": "full",
+        "main_language": "English",
+        "model_type": "chat",
+        "params": 406.0,
+        "result_metrics": {
+            "enem_challenge": 0.8523442967109867,
+            "bluex": 0.8011126564673157,
+            "oab_exams": 0.7640091116173121,
+            "assin2_sts": 0.7888441732870783,
+            "assin2_rte": 0.6317630318610981,
+            "faquad_nli": 0.825063276593557,
+            "hatebr_offensive": 0.9073940659389119,
+            "portuguese_hate_speech": 0.7191480935512969,
+            "tweetsentbr": 0.7821434639106575
+        },
+        "result_metrics_average": 0.7857580188820238,
+        "result_metrics_npm": 0.6584973442501938
+    },
+    {
+        "model": "sabia-3",
+        "name": "Sabiá-3",
+        "link": "https://www.maritaca.ai/",
+        "date": "2024-08-20",
+        "status": "full",
+        "main_language": "Portuguese",
+        "model_type": "proprietary",
+        "result_metrics": {
+            "enem_challenge": 0.8789363191042687,
+            "bluex": 0.7899860917941586,
+            "oab_exams": 0.8391799544419134,
+            "assin2_sts": 0.8253863689009022,
+            "assin2_rte": 0.9477034821619312,
+            "faquad_nli": 0.8243848812618203,
+            "hatebr_offensive": 0.5519158516393349,
+            "portuguese_hate_speech": 0.48273809523809524,
+            "tweetsentbr": 0.5632959814986498
+        },
+        "result_metrics_average": 0.744836336226786,
+        "result_metrics_npm": 0.5802643096708316
     }
 ]