update

app.py

@@ -79,34 +79,34 @@ MMLU_COL = "MMLU (25-shot)"
 TRUTHFULQA_COL = "TruthfulQA (0-shot)"
 NOTES_COL = "Notes"  # For search only
 
-COLS = [MODEL_COL, LANG_COL, CODE_COL, AVERAGE_COL, ARC_COL, HELLASWAG_COL, MMLU_COL, TRUTHFULQA_COL, NOTES_COL]
-TYPES = ["str", "str", "str", "number", "number", "number", "number", "number", "str"]
+# COLS = [MODEL_COL, LANG_COL, CODE_COL, AVERAGE_COL, ARC_COL, HELLASWAG_COL, MMLU_COL, TRUTHFULQA_COL, NOTES_COL]
+# TYPES = ["str", "str", "str", "number", "number", "number", "number", "number", "str"]
 
 
 
 COLS = [MODEL_COL, MSVAMP_COL, MGSM_COL, MNUM_COL,NOTES_COL]
-TYPES = ["str", "number", "number", "number","str"]
+TYPES = ["str", "number", "number", "number","html"]
 
 
 
 def get_leaderboard_df():
     df = list()
     results = [
-        ["GPT-3.5-Turbo", 46.6, 42.2, 49.4],
-        ["MAmmoTH 7B", 26.3, 21.3, 24.2],
-        ["WizardMath 7B", 32.5, 23.0, 28.7],
-        ["MetaMath 7B", 46.2, 37.0, 43.2],
-        ["QAlign 7B", 57.2, 49.6, None],
-        ["MathOctopus 7B", 41.2, 39.5, 37.1],
-        ["MathOctopus-MAPO-DPO 7B(ours)🔥", 57.4, 41.6, 50.4],
-        ["MetaMathOctopus 7B", 53.0, 45.5, 39.2],
-        ["MetaMathOctopus-MAPO-DPO 7B(ours) 👑", 64.7, 51.6, 52.9],
-        ["MistralMathOctopus 7B", 59.0, 58.0, 56.8],
-        ["MistralMathOctopus-MAPO-DPO 7B(ours) 👑", 74.6, 67.3, 70.0]
+        ["GPT-3.5-Turbo", 46.6, 42.2, 49.4,'GPT-3.5-Turbo'],
+        ["MAmmoTH 7B", 26.3, 21.3, 24.2,'<a href="https://arxiv.org/abs/2309.05653" target="_blank">MAmmoTH</a>'],
+        ["WizardMath 7B", 32.5, 23.0, 28.7,'<a href="https://arxiv.org/abs/2308.09583" target="_blank">WizardMath</a>'],
+        ["MetaMath 7B", 46.2, 37.0, 43.2,'<a href="https://arxiv.org/abs/2309.12284" target="_blank">MetaMath</a>'],
+        ["MetaMath-LB-9B",None,50.2,None,'<a href="https://arxiv.org/abs/2401.10695" target="_blank">LangBridge</a>'],
+        ["XCoT 7B",42.9,41.5,None,'<a href="https://arxiv.org/abs/2401.07037" target="_blank">XCoT</a>'],
+        ["QAlign 7B", 57.2, 49.6, None,'<a href="https://arxiv.org/abs/2401.07817" target="_blank">QAlign</a>'],
+        ["MathOctopus 7B", 41.2, 39.5, 37.1,'<a href="https://arxiv.org/abs/2310.20246" target="_blank">MathOctopus</a>'],
+        ["MathOctopus-MAPO-DPO 7B", 57.4, 41.6, 50.4,'<a href="https://arxiv.org/abs/2401.06838" target="_blank">MAPO</a>'],
+        ["MetaMathOctopus 7B", 53.0, 45.5, 39.2,'<a href="https://arxiv.org/abs/2401.06838" target="_blank">MAPO</a>'],
+        ["MetaMathOctopus-MAPO-DPO 7B 👑", 64.7, 51.6, 52.9,'<a href="https://arxiv.org/abs/2401.06838" target="_blank">MAPO</a>'],
+        ["MistralMathOctopus 7B", 59.0, 58.0, 56.8,'<a href="https://arxiv.org/abs/2401.06838" target="_blank">MAPO</a>'],
+        ["MistralMathOctopus-MAPO-DPO 7B 👑", 74.6, 67.3, 70.0,'<a href="https://arxiv.org/abs/2401.06838" target="_blank">MAPO</a>'],
     ]
 
-    for i in results:
-        i.append(i[0])
     df = pd.DataFrame.from_records(results, columns=COLS)
     df = df.sort_values(by=[ MSVAMP_COL], ascending=False)
     df = df[COLS]
@@ -116,18 +116,20 @@ def get_leaderboard_df():
 def get_leaderboard_13Bdf():
     df = list()
     results = [
-        ["GPT-3.5-Turbo", 46.6, 42.2, 49.4],
-        ["MAmmoTH 13B", 38.6, 28.9, 29.5],
-        ["WizardMath 13B", 35.7, 28.3, 29.0],
-        ["MetaMath 13B", 46.2, 43.9, 43.3],
-        ["QAlign 13B", 62.6, 57.1, None],
-        ["MathOctopus 13B", 51.8, 46.0, 40.3],
-        ["MathOctopus-MAPO-DPO 13B (ours) 🔥", 60.1, 48.5, 53.8],
-        ["MetaMathOctopus 13B", 56.3, 51.4, 49.5],
-        ["MetaMathOctopus-MAPO-DPO 13B (ours) 👑", 67.0, 58.0, 59.8]
+        ["GPT-3.5-Turbo", 46.6, 42.2, 49.4,'GPT-3.5-Turbo'],
+        ["MAmmoTH 13B", 38.6, 28.9, 29.5,'<a href="https://arxiv.org/abs/2309.05653" target="_blank">MAmmoTH</a>'],
+        ["WizardMath 13B", 35.7, 28.3, 29.0,'<a href="https://arxiv.org/abs/2308.09583" target="_blank">WizardMath</a>'],
+        ["MetaMath 13B", 46.2, 43.9, 43.3,'<a href="https://arxiv.org/abs/2309.12284" target="_blank">MetaMath</a>'],
+        ["QAlign 13B", 62.6, 57.1, None,'<a href="https://arxiv.org/abs/2401.07817" target="_blank">QAlign</a>'],
+        ["MathOctopus 13B", 51.8, 46.0, 40.3,'<a href="https://arxiv.org/abs/2310.20246" target="_blank">MathOctopus</a>'],
+        ["MetaMath-LB-15B",None,55.2,None,'<a href="https://arxiv.org/abs/2401.10695" target="_blank">LangBridge</a>'],
+        ["MetaMath-LB-20B",None,56.7,None,'<a href="https://arxiv.org/abs/2401.10695" target="_blank">LangBridge</a>'],
+        ["MathOctopus-MAPO-DPO 13B ", 60.1, 48.5, 53.8,'<a href="https://arxiv.org/abs/2401.06838" target="_blank">MAPO</a>'],
+        ["MetaMathOctopus 13B", 56.3, 51.4, 49.5,'<a href="https://arxiv.org/abs/2401.06838" target="_blank">MAPO</a>'],
+        ["MetaMathOctopus-MAPO-DPO 13B  👑", 67.0, 58.0, 59.8,'<a href="https://arxiv.org/abs/2401.06838" target="_blank">MAPO</a>'],
     ]
-    for i in results:
-        i.append(i[0])
+    # for i in results:
+    #     i.append(i[0])
     df = pd.DataFrame.from_records(results, columns=COLS)
     df = df.sort_values(by=[ MSVAMP_COL], ascending=False)
     df = df[COLS]
@@ -149,7 +151,7 @@ with demo:
     gr.Markdown(INTRO_TEXT, elem_classes="markdown-text")
     #gr.Markdown(HOW_TO, elem_classes="markdown-text")
 
-    with gr.Box():
+    with gr.Group():
         search_bar = gr.Textbox(
             placeholder="Search models and languages...", show_label=False, elem_id="search-bar"
         )
@@ -158,13 +160,12 @@ with demo:
             value=original_df,
             headers=COLS,
             datatype=TYPES,
-            max_rows=5,
             elem_id="leaderboard-table",
         )
 
         # # Dummy leaderboard for handling the case when the user uses backspace key
         hidden_leaderboard_table_for_search = gr.components.Dataframe(
-            value=original_df, headers=COLS, datatype=TYPES, max_rows=5, visible=False
+            value=original_df, headers=COLS, datatype=TYPES, visible=False
         )
 
         search_bar.change(
@@ -173,7 +174,7 @@ with demo:
             leaderboard_table,
         )
 
-    with gr.Box():
+    with gr.Group():
         search_bar = gr.Textbox(
             placeholder="Search models and languages...", show_label=False, elem_id="search-bar"
         )
@@ -182,13 +183,12 @@ with demo:
             value=original_13Bdf,
             headers=COLS,
             datatype=TYPES,
-            max_rows=5,
             elem_id="leaderboard-table",
         )
 
         # # Dummy leaderboard for handling the case when the user uses backspace key
         hidden_leaderboard_table_for_search_13B = gr.components.Dataframe(
-            value=original_13Bdf, headers=COLS, datatype=TYPES, max_rows=5, visible=False
+            value=original_13Bdf, headers=COLS, datatype=TYPES,  visible=False
         )
 
         search_bar.change(
@@ -197,7 +197,6 @@ with demo:
             leaderboard_table_13B,
         )
 
-    #gr.Markdown(CREDIT, elem_classes="markdown-text")
     gr.Markdown(CITATION, elem_classes="markdown-text")
 
 demo.launch()

css.py

@@ -1,15 +1,27 @@
+# css.py 文件
 CUSTOM_CSS = """
-/* Hides the final column */
-table td:last-child,
-table th:last-child {
-    display: none;
-}
-
-/* 控制第一列的宽度 */
-table td:first-child,
-table th:first-child {
-    max-width: 200px;
-    overflow: auto;
-    white-space: nowrap;
+#leaderboard-table .dataframe th:nth-child(1),
+#leaderboard-table .dataframe td:nth-child(1) {
+    width: 150px; /* Model column width */
+}
+
+#leaderboard-table .dataframe th:nth-child(2),
+#leaderboard-table .dataframe td:nth-child(2) {
+    width: 100px; /* MSVAMP column width */
+}
+
+#leaderboard-table .dataframe th:nth-child(3),
+#leaderboard-table .dataframe td:nth-child(3) {
+    width: 100px; /* MGSM column width */
+}
+
+#leaderboard-table .dataframe th:nth-child(4),
+#leaderboard-table .dataframe td:nth-child(4) {
+    width: 100px; /* MNum column width */
+}
+
+#leaderboard-table .dataframe th:nth-child(5),
+#leaderboard-table .dataframe td:nth-child(5) {
+    width: 200px; /* Notes column width */
 }
 """