Merge branch 'main' of hf.co:spaces/TheFinAI/open_greek_finance_llm_leaderboard into main

backend/app/services/leaderboard.py

@@ -116,6 +116,11 @@ class LeaderboardService:
                     "value": data.get("QA Raw", 0),
                     "normalized_score": data.get("QA", 0)
                 },
+                "fns": {
+                    "name": "FNS",
+                    "value": data.get("FNS Raw", 0),
+                    "normalized_score": data.get("FNS", 0)
+                },
             }
 
             features = {

frontend/src/pages/LeaderboardPage/components/Leaderboard/constants/defaults.js

@@ -137,6 +137,12 @@ const COLUMNS = {
       defaultVisible: true,
       label: "QA",
     },
+    "evaluations.fns.normalized_score": {
+      group: "evaluation",
+      size: COLUMN_SIZES.BENCHMARK,
+      defaultVisible: true,
+      label: "FNS",
+    },
   },
   MODEL_INFO: {
     "metadata.co2_cost": {

frontend/src/pages/LeaderboardPage/components/Leaderboard/constants/tooltips.js

@@ -48,7 +48,7 @@ export const COLUMN_TOOLTIPS = {
       subItems: ["Language Understanding", "Classification"],
     },
     {
-      label: "Scoring: Micro F1",
+      label: "Scoring: ACC Norm",
       description: "Was the correct choice selected among the options.",
     },
   ]),
@@ -61,11 +61,24 @@ export const COLUMN_TOOLTIPS = {
       subItems: ["Language Understanding", "Classification"],
     },
     {
-      label: "Scoring: Micro F1",
+      label: "Scoring: ACC Norm",
       description: "Was the correct choice selected among the options.",
     },
   ]),
 
+  FNS: createTooltipContent("Multilingual Financial NLP (FNS):", [
+    {
+      label: "Purpose",
+      description:
+        "Tests model's ability to summarize real-world financial annual reports",
+      subItems: ["Language Understanding", "Summarization"],
+    },
+    {
+      label: "Scoring: Rouge1",
+      description: "Was the overlap of unigrams (each word) between the predicted and reference summaries.",
+    },
+  ]),
+
   ARCHITECTURE: createTooltipContent("Model Architecture Information:", [
     {
       label: "Definition",

frontend/src/pages/LeaderboardPage/components/Leaderboard/utils/columnUtils.js

@@ -768,6 +768,15 @@ export const createColumns = (
         "evaluations.qa.normalized_score"
       ],
     },
+    {
+      accessorKey: "evaluations.fns.normalized_score",
+      header: createHeaderCell("FNS", COLUMN_TOOLTIPS.FNS),
+      cell: ({ row, getValue }) =>
+        createScoreCell(getValue, row, "evaluations.fns.normalized_score"),
+      size: TABLE_DEFAULTS.COLUMNS.COLUMN_SIZES[
+        "evaluations.fns.normalized_score"
+      ],
+    },
   ];
 
   const optionalColumns = [