Split T5-11b model into shards
#7
by
iarroyof
- opened
Dear all, I have recently facing problem with this model to load it into multiple GPUs. I realized that its weights are stored into a unique file of 45Gb, which I think does not allow for using device_map='auto'. Please let me know wheter it is possible to download smaller shards or what is the correct way to solve this problem.
Thank you in advance.