5 ms·
When was this? I use JAX+TPUs to train LLMs and haven't experienced many issues. IMO it was way easier to set up distributed training, sharding, etc compared to
by StarCyan 3y ago
When was this? I use JAX+TPUs to train LLMs and haven't experienced many issues. IMO it was way easier to set up distributed training, sharding, etc compared to Pytorch+GPUs.