Update app.py

app.py

@@ -79,49 +79,52 @@ def read_logs():
     with open(log_file, "r") as f:
         return f.read()
 
-with gr.Blocks() as demo:
-    gr.Markdown("# BigCodeBench Evaluation App")
-    
-    with gr.Row():
-        jsonl_file = gr.File(label="Upload JSONL file", file_types=[".jsonl"])
-        split = gr.Dropdown(choices=["complete", "instruct"], label="Split", value="complete")
-        subset = gr.Dropdown(choices=["full", "hard"], label="Subset", value="full")
-    
-    with gr.Row():
-        save_pass_rate = gr.Checkbox(label="Save Pass Rate")
-        parallel = gr.Number(label="Parallel (optional)", precision=0)
-        min_time_limit = gr.Number(label="Min Time Limit", value=1, precision=1)
-        max_as_limit = gr.Number(label="Max AS Limit", value=128*1024, precision=0)
-    
-    with gr.Row():
-        max_data_limit = gr.Number(label="Max Data Limit", value=4*1024, precision=0)
-        max_stack_limit = gr.Number(label="Max Stack Limit", value=5, precision=0)
-        check_gt_only = gr.Checkbox(label="Check GT Only")
-        no_gt = gr.Checkbox(label="No GT")
-    
-    command_output = gr.Textbox(label="Command", lines=2)
-    submit_btn = gr.Button("Run Evaluation")
-    log_output = gr.Textbox(label="Execution Logs", lines=10)
-    
-    def update_command(*args):
-        return generate_command(*args)
-    
-    input_components = [
-        jsonl_file, split, subset, save_pass_rate, parallel,
-        min_time_limit, max_as_limit, max_data_limit, max_stack_limit,
-        check_gt_only, no_gt
-    ]
-    
-    for component in input_components:
-        component.change(update_command, inputs=input_components, outputs=command_output)
-    
-    def on_submit(command):
-        threading.Thread(target=run_bigcodebench, args=(command,), daemon=True).start()
-        return "Evaluation started. Please wait for the logs to update..."
-    
-    submit_btn.click(on_submit, inputs=[command_output], outputs=[log_output])
-    
-    demo.load(read_logs, None, log_output, every=1)
+def run()
+    with gr.Blocks() as demo:
+        gr.Markdown("# BigCodeBench Evaluation App")
+        
+        with gr.Row():
+            jsonl_file = gr.File(label="Upload JSONL file", file_types=[".jsonl"])
+            split = gr.Dropdown(choices=["complete", "instruct"], label="Split", value="complete")
+            subset = gr.Dropdown(choices=["full", "hard"], label="Subset", value="full")
+        
+        with gr.Row():
+            save_pass_rate = gr.Checkbox(label="Save Pass Rate")
+            parallel = gr.Number(label="Parallel (optional)", precision=0)
+            min_time_limit = gr.Number(label="Min Time Limit", value=1, precision=1)
+            max_as_limit = gr.Number(label="Max AS Limit", value=128*1024, precision=0)
+        
+        with gr.Row():
+            max_data_limit = gr.Number(label="Max Data Limit", value=4*1024, precision=0)
+            max_stack_limit = gr.Number(label="Max Stack Limit", value=5, precision=0)
+            check_gt_only = gr.Checkbox(label="Check GT Only")
+            no_gt = gr.Checkbox(label="No GT")
+        
+        command_output = gr.Textbox(label="Command", lines=2)
+        submit_btn = gr.Button("Run Evaluation")
+        log_output = gr.Textbox(label="Execution Logs", lines=10)
+        
+        def update_command(*args):
+            return generate_command(*args)
+        
+        input_components = [
+            jsonl_file, split, subset, save_pass_rate, parallel,
+            min_time_limit, max_as_limit, max_data_limit, max_stack_limit,
+            check_gt_only, no_gt
+        ]
+        
+        for component in input_components:
+            component.change(update_command, inputs=input_components, outputs=command_output)
+        
+        def on_submit(command):
+            threading.Thread(target=run_bigcodebench, args=(command,), daemon=True).start()
+            return "Evaluation started. Please wait for the logs to update..."
+        
+        submit_btn.click(on_submit, inputs=[command_output], outputs=[log_output])
+        
+        demo.load(read_logs, None, log_output, every=1)
 
+    demo.queue().launch()
+    
 if __name__ == "__main__":
-    demo.queue().launch()
+    run()