Reproduce Test

[SuAnita520]

I'm using Ubuntu 24.04 and CUDA 12.1 (verified with nvcc -V). I'm having trouble running the battle_vs_human.sh script. It's giving me an error related to local_rank:
[ERROR ] [Core ] option --local-rank not recognized.

So, I put default config and checkpoint parameters back in test.py, like following:

def parse_args():
    parser = argparse.ArgumentParser(
        description='MMPreTrain test (and eval) a model')
    parser.add_argument('--config', default='./configs/go_battle_vs_human.py')
    parser.add_argument('--checkpoint', default='./work_dirs/go_battle.pth')
    parser.add_argument(
        '--work-dir',
        help='the directory to save the file containing evaluation metrics')
    parser.add_argument('--out', help='the file to output results.')
    ......
    cont. 

but I'm encountering another issue. The code attempts to access files in falcon/datasets/pipelines/tokenizer.py, specifically referencing folders like llama-300m and gpt2_medium, which I don't have. Could you tell me where to download these files? A diagram or description of the final project directory structure would also be very helpful

[MaverickRen]

All the weights are located at https://huggingface.co/maverickrzw/VideoWorld-GoBattle/tree/main/

[SuAnita520]

I downloaded the Intern-300 to work_dirs\Init, then rename it to Llama-300m,

But I get error:

[MaverickRen]

You don't need to rename Intern-300 to Llama-300m. We are only using the tokenizer for Llama-300m. If you encountered an error before indicating that Llama-300m was missing, it was due to a missing weights file. This issue has now been corrected.

[Jona3L]

I also encountered the similar problem, and I replaced it with the original Llama provided on huggingface repo. However, I can still see we are missing iter_990000.pth for inference.