Adding `safetensors` variant of this model (#2)

tokenizer: fix number of extra ids

readme: add some interesting details about umT5

readme: mention conversion script

readme: add initial version

model: add latest PyTorch version

tokenizer: bump number of extra ids

tokenizer: add config

tokenizer: add special tokens map (imported from mT5)

model: fix copy paste bug in conversion script!!

readme: currently do not maintain PyTorch weights (but they are coming back!)

model: update latest version

config: adjust architecture

config: prepare for upcoming umT5 integration

model: revert transpose change

model: transpose changes

config: adjust vocab size (slightly larger than mT5)

model: add converted PyTorch model

vocab: add original spm model

model: add converted model

initial commit