import gradio_client.utils as gc_utils _original_json_schema_to_python_type = gc_utils._json_schema_to_python_type def patched_json_schema_to_python_type(schema, defs=None): if isinstance(schema, bool): return {} return _original_json_schema_to_python_type(schema, defs) gc_utils._json_schema_to_python_type = patched_json_schema_to_python_type import logging import os os.makedirs("tmp", exist_ok=True) os.environ['TMP_DIR'] = "tmp" import subprocess import shutil import glob import gradio as gr import numpy as np from apscheduler.schedulers.background import BackgroundScheduler import json from io import BytesIO from src.radial.radial import create_plot from gradio_leaderboard import Leaderboard, SelectColumns from gradio_space_ci import enable_space_ci from src.display.about import INTRODUCTION_TEXT, TITLE, LLM_BENCHMARKS_TEXT from src.display.css_html_js import custom_css from src.display.utils import AutoEvalColumn, fields from src.envs import API, H4_TOKEN, HF_HOME, REPO_ID, RESET_JUDGEMENT_ENV from src.leaderboard.build_leaderboard import build_leadearboard_df, download_openbench, download_dataset import huggingface_hub os.environ["GRADIO_ANALYTICS_ENABLED"] = "false" logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s") enable_space_ci() def handle_file_upload(file_bytes): """ Read the uploaded bytes and parse JSON directly, avoiding ephemeral disk paths or file read issues. """ logging.info("File uploaded (bytes). Size: %d bytes", len(file_bytes)) v = json.loads(file_bytes.decode("utf-8")) return v def submit_file(v, mn): """ We removed file_path because we no longer need it (no ephemeral path). 'v' is the loaded JSON object. """ print('START SUBMITTING!!!') if 'results' not in v: return "Invalid JSON: missing 'results' key" new_file = v['results'] new_file['model'] = mn columns = [ 'mmlu_translated_kk', 'kk_constitution_mc', 'kk_dastur_mc', 'kazakh_and_literature_unt_mc', 'kk_geography_unt_mc', 'kk_world_history_unt_mc', 'kk_history_of_kazakhstan_unt_mc', 'kk_english_unt_mc', 'kk_biology_unt_mc', 'kk_human_society_rights_unt_mc' ] for column in columns: if column not in new_file or not isinstance(new_file[column], dict): return f"Missing or invalid column: {column}" if 'acc,none' not in new_file[column]: return f"Missing 'acc,none' key in column: {column}" new_file[column] = new_file[column]['acc,none'] if 'config' not in v or 'model_dtype' not in v['config']: return "Missing 'config' or 'model_dtype' in JSON" new_file['model_dtype'] = v['config']["model_dtype"] new_file['ppl'] = 0 print('WE READ FILE: ', new_file) buf = BytesIO() buf.write(json.dumps(new_file).encode('utf-8')) buf.seek(0) API.upload_file( path_or_fileobj=buf, path_in_repo="model_data/external/" + mn.replace('/', '__') + ".json", repo_id="kz-transformers/s-openbench-eval", repo_type="dataset", ) os.environ[RESET_JUDGEMENT_ENV] = "1" return "Success!" def restart_space(): API.restart_space(repo_id=REPO_ID) download_openbench() def update_plot(selected_models): return create_plot(selected_models) def build_demo(): download_openbench() demo = gr.Blocks(title="Kaz LLM LB", css=custom_css) leaderboard_df = build_leadearboard_df() with demo: gr.HTML(TITLE) gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text") with gr.Tabs(elem_classes="tab-buttons"): with gr.TabItem("🏅 LLM Benchmark", elem_id="llm-benchmark-tab-table", id=0): Leaderboard( value=leaderboard_df, datatype=[c.type for c in fields(AutoEvalColumn)], select_columns=SelectColumns( default_selection=[c.name for c in fields(AutoEvalColumn) if c.displayed_by_default], cant_deselect=[c.name for c in fields(AutoEvalColumn) if c.never_hidden or c.dummy], label="Select Columns to Display:", ), search_columns=[AutoEvalColumn.model.name], ) with gr.TabItem("🚀 Submit ", elem_id="llm-benchmark-tab-table", id=3): with gr.Row(): gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text") with gr.Row(): gr.Markdown("# ✨ Submit your model here!", elem_classes="markdown-text") with gr.Column(): model_name_textbox = gr.Textbox(label="Model name") file_output = gr.File( label="Drag and drop JSON file judgment here", type="binary" ) uploaded_file = gr.State() with gr.Row(): with gr.Column(): out = gr.Textbox("Submission Status") submit_button = gr.Button("Submit File", variant='primary') file_output.upload( fn=handle_file_upload, inputs=file_output, outputs=uploaded_file ) submit_button.click( fn=submit_file, inputs=[uploaded_file, model_name_textbox], outputs=[out] ) with gr.TabItem("📊 Analytics", elem_id="llm-benchmark-tab-table", id=4): with gr.Column(): model_dropdown = gr.Dropdown( choices=leaderboard_df["model"].tolist(), label="Models", value=leaderboard_df["model"].tolist(), multiselect=True, info="Select models" ) with gr.Column(): plot = gr.Plot(update_plot(model_dropdown.value)) model_dropdown.change( fn=update_plot, inputs=[model_dropdown], outputs=[plot] ) return demo def aggregate_leaderboard_data(): download_dataset("kz-transformers/s-openbench-eval", "m_data") data_list = [ { "model_dtype": "torch.float16", "model": "dummy-random-baseline", "ppl": 0, "mmlu_translated_kk": 0.22991508817766165, "kk_constitution_mc": 0.25120772946859904, "kk_dastur_mc": 0.24477611940298508, "kazakh_and_literature_unt_mc": 0.2090443686006826, "kk_geography_unt_mc": 0.2019790454016298, "kk_world_history_unt_mc": 0.1986970684039088, "kk_history_of_kazakhstan_unt_mc": 0.19417177914110428, "kk_english_unt_mc": 0.189804278561675, "kk_biology_unt_mc": 0.22330729166666666, "kk_human_society_rights_unt_mc": 0.242152466367713, }, { "model_dtype": "torch.float16", "model": "gpt-4o-mini", "ppl": 0, "mmlu_translated_kk": 0.5623775310254735, "kk_constitution_mc": 0.79, "kk_dastur_mc": 0.755, "kazakh_and_literature_unt_mc": 0.4953071672354949, "kk_geography_unt_mc": 0.5675203725261933, "kk_world_history_unt_mc": 0.6091205211726385, "kk_history_of_kazakhstan_unt_mc": 0.47883435582822087, "kk_english_unt_mc": 0.6763768775603095, "kk_biology_unt_mc": 0.607421875, "kk_human_society_rights_unt_mc": 0.7309417040358744, }, { "model_dtype": "api", "model": "gpt-4o", "ppl": 0, "mmlu_translated_kk": 0.7419986936642717, "kk_constitution_mc": 0.841, "kk_dastur_mc": 0.798, "kazakh_and_literature_unt_mc": 0.6785409556313993, "kk_geography_unt_mc": 0.629802095459837, "kk_world_history_unt_mc": 0.6783387622149837, "kk_history_of_kazakhstan_unt_mc": 0.6785276073619632, "kk_english_unt_mc": 0.7410104688211198, "kk_biology_unt_mc": 0.6979166666666666, "kk_human_society_rights_unt_mc": 0.7937219730941704, }, { "model_dtype": "torch.float16", "model": "nova-pro-v1", "ppl": 0, "mmlu_translated_kk": 0.6792945787067276, "kk_constitution_mc": 0.7753623188405797, "kk_dastur_mc": 0.718407960199005, "kazakh_and_literature_unt_mc": 0.4656569965870307, "kk_geography_unt_mc": 0.5541327124563445, "kk_world_history_unt_mc": 0.6425081433224755, "kk_history_of_kazakhstan_unt_mc": 0.5, "kk_english_unt_mc": 0.6845698680018206, "kk_biology_unt_mc": 0.6197916666666666, "kk_human_society_rights_unt_mc": 0.7713004484304933, }, { "model_dtype": "torch.float16", "model": "gemini-1.5-pro", "ppl": 0, "mmlu_translated_kk": 0.7380796864794252, "kk_constitution_mc": 0.8164251207729468, "kk_dastur_mc": 0.7383084577114428, "kazakh_and_literature_unt_mc": 0.5565273037542662, "kk_geography_unt_mc": 0.6065192083818394, "kk_world_history_unt_mc": 0.6669381107491856, "kk_history_of_kazakhstan_unt_mc": 0.5791411042944785, "kk_english_unt_mc": 0.7114246700045517, "kk_biology_unt_mc": 0.6673177083333334, "kk_human_society_rights_unt_mc": 0.7623318385650224, }, { "model_dtype": "torch.float16", "model": "gemini-1.5-flash", "ppl": 0, "mmlu_translated_kk": 0.6335728282168517, "kk_constitution_mc": 0.748792270531401, "kk_dastur_mc": 0.7054726368159204, "kazakh_and_literature_unt_mc": 0.4761092150170648, "kk_geography_unt_mc": 0.5640279394644936, "kk_world_history_unt_mc": 0.5838762214983714, "kk_history_of_kazakhstan_unt_mc": 0.43374233128834355, "kk_english_unt_mc": 0.6681838871187984, "kk_biology_unt_mc": 0.6217447916666666, "kk_human_society_rights_unt_mc": 0.7040358744394619, }, { "model_dtype": "torch.float16", "model": "claude-3-5-sonnet", "ppl": 0, "mmlu_translated_kk": 0.7335075114304376, "kk_constitution_mc": 0.8623188405797102, "kk_dastur_mc": 0.7950248756218905, "kazakh_and_literature_unt_mc": 0.6548634812286689, "kk_geography_unt_mc": 0.6431897555296857, "kk_world_history_unt_mc": 0.6669381107491856, "kk_history_of_kazakhstan_unt_mc": 0.6251533742331289, "kk_english_unt_mc": 0.7291761492944925, "kk_biology_unt_mc": 0.6686197916666666, "kk_human_society_rights_unt_mc": 0.8026905829596412, }, { "model_dtype": "torch.float16", "model": "yandex-gpt", "ppl": 0, "mmlu_translated_kk": 0.39777922926192033, "kk_constitution_mc": 0.7028985507246377, "kk_dastur_mc": 0.6159203980099502, "kazakh_and_literature_unt_mc": 0.3914249146757679, "kk_geography_unt_mc": 0.4912689173457509, "kk_world_history_unt_mc": 0.5244299674267101, "kk_history_of_kazakhstan_unt_mc": 0.4030674846625767, "kk_english_unt_mc": 0.5844333181611289, "kk_biology_unt_mc": 0.4368489583333333, "kk_human_society_rights_unt_mc": 0.6995515695067265, }, ] files_list = glob.glob("./m_data/model_data/external/*.json") logging.info(f'FILES LIST: {files_list}') for file in files_list: logging.info(f'Trying to read external submit file: {file}') try: with open(file) as f: data = json.load(f) if not isinstance(data, dict): logging.warning(f"File {file} is not a dict, skipping") continue required_keys = {'model_dtype', 'model', 'ppl', 'mmlu_translated_kk'} if not required_keys.issubset(data.keys()): logging.warning(f"File {file} missing required keys, skipping") continue logging.info(f'Successfully read: {file}, got {len(data)} keys') data_list.append(data) except Exception as e: logging.error(f"Error reading file {file}: {e}") continue logging.info("Combined data_list length: %d", len(data_list)) with open("genned.json", "w") as f: json.dump(data_list, f) API.upload_file( path_or_fileobj="genned.json", path_in_repo="leaderboard.json", repo_id="kz-transformers/kaz-llm-lb-metainfo", repo_type="dataset", ) def update_board(): need_reset = os.environ.get(RESET_JUDGEMENT_ENV) logging.info("Updating the judgement (scheduled update): %s", need_reset) if need_reset != "1": pass os.environ[RESET_JUDGEMENT_ENV] = "0" aggregate_leaderboard_data() restart_space() def update_board_(): logging.info("Updating the judgement at startup") aggregate_leaderboard_data() if __name__ == "__main__": os.environ[RESET_JUDGEMENT_ENV] = "1" from apscheduler.schedulers.background import BackgroundScheduler scheduler = BackgroundScheduler() update_board_() scheduler.add_job(update_board, "interval", minutes=10) scheduler.start() demo_app = build_demo() demo_app.launch(debug=True, share=False, show_api=False)