Update app.py

app.py

@@ -1,6 +1,6 @@
-# app.py (final: in-memory ZIP + error handling + Gradio binary file)
+# app.py (final: temp-file ZIP + auto-clean + error handling)
 
-import os, io, json, zipfile, traceback
+import os, glob, json, zipfile, traceback
 import gradio as gr
 import pandas as pd
 from datetime import datetime, timezone
@@ -46,6 +46,13 @@ def run_eval(conversation: str,
         if not conversation or conversation.strip() == "":
             return None, None, None, None, "❌ Please paste a conversation to evaluate."
 
+        # cleanup old ZIPs
+        for f in glob.glob("/tmp/financeeval_*.zip"):
+            try:
+                os.remove(f)
+            except Exception:
+                pass
+
         # normalize weights
         user_weights = {
             "trust": w_trust, "accuracy": w_accuracy, "explain": w_explain,
@@ -107,10 +114,10 @@ def run_eval(conversation: str,
                 avg_row[c] = round(compare_df[c].mean(), 2)
             avg_df = pd.DataFrame([avg_row])
 
-        # ---- In-memory ZIP (privacy-safe) ----
+        # ---- Write ZIP into /tmp (ephemeral, privacy-safe) ----
         ts = datetime.now(timezone.utc).strftime("%Y%m%dT%H%M%SZ")
-        zip_buffer = io.BytesIO()
-        with zipfile.ZipFile(zip_buffer, "w", zipfile.ZIP_DEFLATED) as zf:
+        zip_path = f"/tmp/financeeval_{ts}.zip"
+        with zipfile.ZipFile(zip_path, "w", zipfile.ZIP_DEFLATED) as zf:
             for label, df, total in all_tables:
                 df2 = df.copy()
                 df2.loc[len(df2)] = {
@@ -123,9 +130,8 @@ def run_eval(conversation: str,
                 zf.writestr(f"results_{label}_{ts}.csv", df2.to_csv(index=False).encode("utf-8"))
             zf.writestr(f"comparison_{ts}.csv", compare_df.to_csv(index=False).encode("utf-8"))
             zf.writestr(f"judgments_{ts}.json", json.dumps(json_blobs, indent=2).encode("utf-8"))
-        zip_buffer.seek(0)
-        zip_bytes = zip_buffer.getvalue()  # raw bytes
 
+        # merge tables for UI
         merged_tables = []
         for label, df, total in all_tables:
             merged_tables.append(pd.DataFrame({
@@ -139,7 +145,7 @@ def run_eval(conversation: str,
         merged_df = pd.concat(merged_tables, ignore_index=True)
         usage_text_all = "\n".join(token_usage_blocks)
 
-        return merged_df, compare_df, (avg_df if avg_df is not None else pd.DataFrame()), zip_bytes, usage_text_all
+        return merged_df, compare_df, (avg_df if avg_df is not None else pd.DataFrame()), zip_path, usage_text_all
 
     except Exception as e:
         tb = traceback.format_exc()
@@ -175,7 +181,7 @@ with gr.Blocks(title="FinanceEval – Hybrid Judge (Gradio)") as demo:
         compare_out = gr.Dataframe()
         avg_out = gr.Dataframe()
     with gr.Tab("Downloads & Usage"):
-        zip_file = gr.File(label="Download ZIP (CSVs + JSON)", type="binary")
+        zip_file = gr.File(label="Download ZIP (CSVs + JSON)", type="file")
         usage_text = gr.Textbox(label="Token Usage / Errors", lines=8)
 
     run_btn.click(