Inference on TPU-v3-32
#69
by
ybelkada
HF staff
- opened
Hi @zhiG !
Let's move the discussion here!
For BLOOM tpu inference you can check this repo: https://github.com/huggingface/bloom-jax-inference where we use Flax model partitioning
original discussion: https://huggingface.co/bigscience/bloom/discussions/68
Thank you very much for your information, I'll try to use it in TPU!
It seems like the JAX checkpoints of 176B BLOOM model are not available now, so we should wait.