import json import os from datetime import datetime import pandas as pd def generate_request(model_id, precision, model_type, params, index): data = { "model": model_id, "base_model": "", "revision": "main", "private": False, "precision": precision, "weight_type": "Original", "status": "FINISHED", "submitted_time": (datetime.now() + pd.Timedelta(hours=index)).strftime( "%Y-%m-%dT%H:%M:%SZ" ), "model_type": f"\ud83d\udfe2 : {model_type} if model_type == 'pretrained' else model_type", "likes": 0, "params": params, "license": "custom", "architecture": "", "sender": "mariagrandury", } os.makedirs(f"{model_id}", exist_ok=True) with open(f"{model_id}_eval_request_False_{precision}_Original.json", "w") as f: json.dump(data, f) def generate_requests(selection: str): df = pd.read_csv("scripts/models.csv") df = df[["model_id", "precision", "model_type", "params", "iberobench"]] if selection == "pretrained": df = df[df["model_type"] == "pretrained"] elif selection == "pretrained_new": df = df[df["model_type"] == "pretrained"] df = df[df["iberobench"] == False] elif selection == "instruction": df = df[df["model_type"] == "instruction-tuned"] for index, row in df.iterrows(): model_id, precision, model_type, params, iberobench = row generate_request( model_id=model_id, precision=precision, model_type=model_type, params=params, index=index, ) if __name__ == "__main__": import argparse parser = argparse.ArgumentParser(description="Generate model requests.") parser.add_argument("--pretrained", action="store_true") parser.add_argument("--pretrained_new", action="store_true") parser.add_argument("--instruction", action="store_true") args = parser.parse_args() if args.pretrained: generate_requests("pretrained") elif args.pretrained_new: generate_requests("pretrained_new") elif args.instruction: generate_requests("instruction") else: generate_requests()