import argparse
import json
from pathlib import Path

from huggingface_hub import snapshot_download
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer


ROOT_FILES = [
    'adapter_config.json',
    'adapter_model.safetensors',
    'tokenizer.json',
    'tokenizer_config.json',
    'chat_template.jinja',
]


def load_adapter(subfolder=None, revision=None):
    """Load the root MuSiQue 1.7B SLS adapter, or a checkpoint subfolder."""
    repo_id = 'SJTUFreshman/collaborativelearning'
    subfolder = (subfolder or '').replace('\\', '/').strip('/')
    if subfolder == '.':
        subfolder = ''
    tokenizer_folder = str(Path(subfolder).parent).replace('\\', '/') if subfolder.endswith('/policy_adapter') else subfolder
    patterns = (
        [subfolder+'/*', tokenizer_folder+'/tokenizer*', tokenizer_folder+'/chat_template.jinja', tokenizer_folder+'/special_tokens_map.json']
        if subfolder else ROOT_FILES
    )
    snapshot = snapshot_download(repo_id, revision=revision, allow_patterns=patterns)
    adapter_dir = Path(snapshot)/subfolder
    config = json.loads((adapter_dir/'adapter_config.json').read_text())
    tokenizer = AutoTokenizer.from_pretrained(Path(snapshot)/tokenizer_folder)
    base = AutoModelForCausalLM.from_pretrained(config['base_model_name_or_path'], torch_dtype='auto', device_map='auto')
    model = PeftModel.from_pretrained(base, adapter_dir)
    return model, tokenizer


if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('subfolder', nargs='?', help='Checkpoint directory; default: root MuSiQue 1.7B SLS adapter')
    parser.add_argument('--revision')
    args = parser.parse_args()
    model, tokenizer = load_adapter(args.subfolder, args.revision)
    print('Loaded', args.subfolder or 'root (MuSiQue/SLS/Qwen3-1.7B/final)')