JingerAI

Attempting to reproduce the effect of 1.6b step1 SFT using the default single_node script configuration resulted in slow training on 4 V100 32G GPUs. It took 6 hours to complete,...