Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -825,7 +825,19 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
|
|
| 825 |
results_df = pd.DataFrame(results_log)
|
| 826 |
return status_message, results_df
|
| 827 |
|
| 828 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 829 |
# --- Build Gradio Interface using Blocks ---
|
| 830 |
with gr.Blocks() as demo:
|
| 831 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
|
@@ -849,14 +861,22 @@ with gr.Blocks() as demo:
|
|
| 849 |
run_button = gr.Button("Run Evaluation & Submit All Answers")
|
| 850 |
|
| 851 |
status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
|
| 852 |
-
# Removed max_rows=10 from DataFrame constructor
|
| 853 |
results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
|
| 854 |
-
|
|
|
|
|
|
|
| 855 |
run_button.click(
|
| 856 |
fn=run_and_submit_all,
|
| 857 |
-
outputs=[status_output, results_table]
|
| 858 |
)
|
| 859 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 860 |
if __name__ == "__main__":
|
| 861 |
print("\n" + "-"*30 + " App Starting " + "-"*30)
|
| 862 |
# Check for SPACE_HOST and SPACE_ID at startup for information
|
|
|
|
| 825 |
results_df = pd.DataFrame(results_log)
|
| 826 |
return status_message, results_df
|
| 827 |
|
| 828 |
+
answers_state = gr.State([])
|
| 829 |
+
|
| 830 |
+
def build_jsonl(answers):
|
| 831 |
+
import json, os
|
| 832 |
+
path = "/tmp/submission.jsonl"
|
| 833 |
+
with open(path, "w", encoding="utf-8") as f:
|
| 834 |
+
for a in answers:
|
| 835 |
+
f.write(json.dumps(
|
| 836 |
+
{"task_id": a["task_id"], "model_answer": a["submitted_answer"]},
|
| 837 |
+
ensure_ascii=False
|
| 838 |
+
) + "\n")
|
| 839 |
+
return path # gr.File/DownloadButton will serve this file
|
| 840 |
+
|
| 841 |
# --- Build Gradio Interface using Blocks ---
|
| 842 |
with gr.Blocks() as demo:
|
| 843 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
|
|
|
| 861 |
run_button = gr.Button("Run Evaluation & Submit All Answers")
|
| 862 |
|
| 863 |
status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
|
|
|
|
| 864 |
results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
|
| 865 |
+
jsonl_file = gr.File(label="submission.jsonl", interactive=False)
|
| 866 |
+
save_btn = gr.Button("Build JSONL for Leaderboard")
|
| 867 |
+
|
| 868 |
run_button.click(
|
| 869 |
fn=run_and_submit_all,
|
| 870 |
+
outputs=[status_output, results_table, answers_state]
|
| 871 |
)
|
| 872 |
|
| 873 |
+
save_btn.click(
|
| 874 |
+
fn=build_jsonl,
|
| 875 |
+
inputs=[answers_state],
|
| 876 |
+
outputs=[jsonl_file]
|
| 877 |
+
)
|
| 878 |
+
|
| 879 |
+
|
| 880 |
if __name__ == "__main__":
|
| 881 |
print("\n" + "-"*30 + " App Starting " + "-"*30)
|
| 882 |
# Check for SPACE_HOST and SPACE_ID at startup for information
|