Upload 2 files

arena_ui.py

@@ -0,0 +1,35 @@
+import gradio as gr
+import os
+import pandas as pd
+import json
+
+ratings = json.load(open("ratings.json"))
+# df = pd.DataFrame(columns=["Rank", "Name", "Arena Elo"])
+# for i in range(len(ratings)):
+#     df = df.append(
+#         {"Rank": i + 1, "Name": ratings[i][0], "Arena Elo": ratings[i][1]},
+#         ignore_index=True,
+# )
+df = pd.DataFrame(ratings, columns=["Name", "Arena Elo"])
+df["Rank"] = df.index + 1
+df = df[["Rank", "Name", "Arena Elo"]]
+
+
+with gr.Blocks(css="footer {visibility: hidden}") as demo:
+    md = gr.Markdown(
+        """## Internal Leaderboard
+        """
+    )
+
+    md2 = gr.Markdown(
+        """Notes:
+        - This is very unreliable as of now.
+        - The agents have been implemented haphazardly and only evaluated on 10 tasks.
+        """
+    )
+    with gr.Row():
+        table = gr.Dataframe(df)
+
+if __name__ == "__main__":
+    demo.queue()
+    demo.launch(debug=True)

ratings.json

@@ -0,0 +1 @@
+[["vision_agent_reranking", 1013.6523617512925], ["standard_agent", 1003.3564093107665], ["standard_agent_reranking", 997.2581907453138], ["vision_agent", 985.7330381926273]]