Update space
Browse files
app.py
CHANGED
@@ -57,10 +57,11 @@ LEADERBOARD_DF = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS,
|
|
57 |
pending_eval_queue_df,
|
58 |
) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS)
|
59 |
|
|
|
60 |
def init_leaderboard(dataframe):
|
61 |
if dataframe is None or dataframe.empty:
|
62 |
raise ValueError("Leaderboard DataFrame is empty or None.")
|
63 |
-
|
64 |
return Leaderboard(
|
65 |
value=dataframe,
|
66 |
datatype=[c.type for c in fields(AutoEvalColumn)],
|
@@ -105,6 +106,18 @@ with demo:
|
|
105 |
with gr.TabItem("🏅 LLM Benchmark", elem_id="llm-benchmark-tab-table", id=0):
|
106 |
leaderboard = init_leaderboard(LEADERBOARD_DF)
|
107 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
108 |
with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=2):
|
109 |
gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
|
110 |
|
|
|
57 |
pending_eval_queue_df,
|
58 |
) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS)
|
59 |
|
60 |
+
|
61 |
def init_leaderboard(dataframe):
|
62 |
if dataframe is None or dataframe.empty:
|
63 |
raise ValueError("Leaderboard DataFrame is empty or None.")
|
64 |
+
|
65 |
return Leaderboard(
|
66 |
value=dataframe,
|
67 |
datatype=[c.type for c in fields(AutoEvalColumn)],
|
|
|
106 |
with gr.TabItem("🏅 LLM Benchmark", elem_id="llm-benchmark-tab-table", id=0):
|
107 |
leaderboard = init_leaderboard(LEADERBOARD_DF)
|
108 |
|
109 |
+
with gr.TabItem("Math", elem_id="llm-benchmark-tab-table", id=0):
|
110 |
+
leaderboard = init_leaderboard(LEADERBOARD_DF)
|
111 |
+
|
112 |
+
with gr.TabItem("Reasoning", elem_id="llm-benchmark-tab-table", id=0):
|
113 |
+
leaderboard = init_leaderboard(LEADERBOARD_DF)
|
114 |
+
|
115 |
+
with gr.TabItem("Coding", elem_id="llm-benchmark-tab-table", id=0):
|
116 |
+
leaderboard = init_leaderboard(LEADERBOARD_DF)
|
117 |
+
|
118 |
+
with gr.TabItem("Science", elem_id="llm-benchmark-tab-table", id=0):
|
119 |
+
leaderboard = init_leaderboard(LEADERBOARD_DF)
|
120 |
+
|
121 |
with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=2):
|
122 |
gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
|
123 |
|