import json import gradio as gr from pathlib import Path from src.hf_utils import load_leaderboard_data, upload_submission, check_name_exists from src.eval import start_background_evaluation def handle_upload(submission_name, uploaded_file, progress=gr.Progress()): """Handle file upload and start evaluation.""" if not uploaded_file: return "No file uploaded. Please upload a valid submission file." # normalize the submission name submission_name = submission_name.strip().replace(" ", "_").lower() # keep only alphanumeric characters and underscores, restrict to 30 characters submission_name = "".join( c for c in submission_name if c.isalnum() or c == "_" )[:30] if not submission_name or submission_name.strip() == "": return "Submission name is required" if check_name_exists(submission_name): return f"Submission name '{submission_name}' already exists. Please choose a different name." try: progress(0.3, "Uploading to Hugging Face...") # Check if the file is a valid JSONL file if not uploaded_file.name.endswith(".jsonl"): return "Invalid file format. Please upload a .jsonl file." # Check that the keys in the JSONL file are correct ('id' and 'model') with open(uploaded_file.name, "r") as file: found_one = False for line in file: found_one = True json_object = json.loads(line) if not all(key in json_object for key in ["id", "model"]): return "Invalid content. Each line must contain 'id' and 'model' keys." if not found_one: return "Empty file. Please upload a valid JSONL file." success, result = upload_submission(uploaded_file, submission_name) if not success: return f"Upload failed: {result}" progress(0.7, "Starting evaluation...") # Start evaluation start_background_evaluation(result) progress(1.0, "Process complete") return f"Upload complete. Evaluation started for: {submission_name}. Refresh the leaderboard to see results. Do not worry if the leaderboard does not update immediately; it may take some time for the results to appear." except Exception as e: return f"Error processing upload: {str(e)}" def create_ui(): """Create and return Gradio UI.""" with gr.Blocks(title="CP-Bench Leaderboard") as demo: gr.Markdown("# CP-Bench Leaderboard") with gr.Row(): with gr.Column(scale=1): gr.Markdown("## 📤 Upload Submission") submission_name = gr.Textbox( label="Submission Name (required)", placeholder="Enter a unique name for your submission", interactive=True, info="This name will appear on the leaderboard" ) upload_button = gr.UploadButton("Click to Upload Submission", file_count="single") status_box = gr.Textbox(label="Status", interactive=False) with gr.Column(scale=3): gr.Markdown("## 🏆 Results Leaderboard") leaderboard = gr.DataFrame(value=load_leaderboard_data, label="Leaderboard", interactive=False) refresh_button = gr.Button("🔄 Refresh Leaderboard") # Event handlers upload_button.upload( fn=handle_upload, inputs=[submission_name, upload_button], outputs=[status_box], show_progress="full", ) refresh_button.click( fn=load_leaderboard_data, inputs=None, outputs=[leaderboard] ) return demo