rodrigomasini/data_only_hallucination_leaderboard

1import json2import os3import pprint4import re5from datetime import datetime, timezone6 7import click8from colorama import Fore9from huggingface_hub import HfApi, snapshot_download10 11EVAL_REQUESTS_PATH = "eval-queue"12QUEUE_REPO = "hallucinations-leaderboard/requests"13 14precisions = ("float16", "bfloat16", "8bit (LLM.int8)", "4bit (QLoRA / FP4)", "GPTQ")15model_types = ("pretrained", "fine-tuned", "RL-tuned", "instruction-tuned")16weight_types = ("Original", "Delta", "Adapter")17 18 19def get_model_size(model_info, precision: str):20    size_pattern = size_pattern = re.compile(r"(\d\.)?\d+(b|m)")21    try:22        model_size = round(model_info.safetensors["total"] / 1e9, 3)23    except (AttributeError, TypeError):24        try:25            size_match = re.search(size_pattern, model_info.modelId.lower())26            model_size = size_match.group(0)27            model_size = round(float(model_size[:-1]) if model_size[-1] == "b" else float(model_size[:-1]) / 1e3, 3)28        except AttributeError:29            return 0  # Unknown model sizes are indicated as 0, see NUMERIC_INTERVALS in app.py30 31    size_factor = 8 if (precision == "GPTQ" or "gptq" in model_info.modelId.lower()) else 132    model_size = size_factor * model_size33    return model_size34 35 36def main():37    api = HfApi()38    current_time = datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")39    snapshot_download(repo_id=QUEUE_REPO, revision="main", local_dir=EVAL_REQUESTS_PATH, repo_type="dataset")40 41    model_name = click.prompt("Enter model name")42    revision = click.prompt("Enter revision", default="main")43    precision = click.prompt("Enter precision", default="float32", type=click.Choice(precisions))44    model_type = click.prompt("Enter model type", type=click.Choice(model_types))45    weight_type = click.prompt("Enter weight type", default="Original", type=click.Choice(weight_types))46    base_model = click.prompt("Enter base model", default="")47    status = click.prompt("Enter status", default="FINISHED")48 49    try:50        model_info = api.model_info(repo_id=model_name, revision=revision)51    except Exception as e:52        print(f"{Fore.RED}Could not find model info for {model_name} on the Hub\n{e}{Fore.RESET}")53        return 154 55    model_size = get_model_size(model_info=model_info, precision=precision)56 57    try:58        license = model_info.cardData["license"]59    except Exception:60        license = "?"61 62    eval_entry = {63        "model": model_name,64        "base_model": base_model,65        "revision": revision,66        "private": False,67        "precision": precision,68        "weight_type": weight_type,69        "status": status,70        "submitted_time": current_time,71        "model_type": model_type,72        "likes": model_info.likes,73        "params": model_size,74        "license": license,75    }76 77    user_name = ""78    model_path = model_name79    if "/" in model_name:80        user_name = model_name.split("/")[0]81        model_path = model_name.split("/")[1]82 83    pprint.pprint(eval_entry)84 85    if click.confirm("Do you want to continue? This request file will be pushed to the hub"):86        click.echo("continuing...")87 88        out_dir = f"{EVAL_REQUESTS_PATH}/{user_name}"89        os.makedirs(out_dir, exist_ok=True)90        out_path = f"{out_dir}/{model_path}_eval_request_{False}_{precision}_{weight_type}.json"91 92        with open(out_path, "w") as f:93            f.write(json.dumps(eval_entry))94 95        api.upload_file(96            path_or_fileobj=out_path,97            path_in_repo=out_path.split(f"{EVAL_REQUESTS_PATH}/")[1],98            repo_id=QUEUE_REPO,99            repo_type="dataset",100            commit_message=f"Add {model_name} to eval queue",101        )102    else:103        click.echo("aborting...")104 105 106if __name__ == "__main__":107    main()108