Update app.py

app.py

@@ -8,13 +8,30 @@ import os
 @spaces.GPU
 def run_peft_eval():
     print("Starting evaluation...", flush=True)
+
     try:
-        result = subprocess.run(["python3", "run_eval.py"], capture_output=True, text=True, check=True)
-        print(result.stdout, flush=True)
-    except subprocess.CalledProcessError as e:
-        print("Eval script failed:\n", e.stderr, flush=True)
+        process = subprocess.Popen(
+            ["python3", "run_eval.py"],
+            stdout=subprocess.PIPE,
+            stderr=subprocess.STDOUT,
+            universal_newlines=True,
+        )
+
+        log_lines = []
+        for line in process.stdout:
+            print(line, end='', flush=True)
+            log_lines.append(line)
+
+        process.wait()
+
+        if process.returncode != 0:
+            return "Evaluation failed. See logs above."
+
+    except Exception as e:
+        print("Eval script failed:", e, flush=True)
         return "Evaluation failed. Check logs."
 
+    # Optional shutdown trigger (Spaces-only feature)
     try:
         shutdown_url = os.environ.get("HF_ENDPOINT_SHUTDOWN")
         if shutdown_url:
@@ -23,13 +40,15 @@ def run_peft_eval():
     except Exception as e:
         print("Shutdown failed:", e, flush=True)
 
-    return f"PEFT-Bench completed at {datetime.datetime.utcnow():%Y-%m-%d %H:%M UTC}"
+    return f"✅ PEFT-Bench completed at {datetime.datetime.utcnow():%Y-%m-%d %H:%M UTC}"
 
 with gr.Blocks() as demo:
-    gr.Markdown("PEFT-Bench GPU Evaluator")
-    gr.Markdown("Click the button below to start evaluation. The Space will shut down when done.")
+    gr.Markdown("### PEFT-Bench GPU Evaluator")
+    gr.Markdown("Click the button below to run `run_eval.py`. The Space will automatically shut down when complete.")
+    
     status = gr.Textbox(label="Status", lines=2)
     btn = gr.Button("Start Evaluation")
+    
     btn.click(fn=run_peft_eval, outputs=status)
 
 demo.launch(show_error=True)