Sanjin2024

/

TinyRecursiveModels-ARC-AGI-1

like 0

Hi,

I am wondering if you could add the following print statement in pretrain.py and upload the results for, say, 1000 steps? Thanks. I am not able to reproduce the results. This will help me debug.

           wandb.log(metrics, step=train_state.step)
           print(f"metrics= {metrics}, step= {train_state.step}, rank= {RANK}", flush=True)