updated transformers version

update publishing info

added facade functionality for prepare_inputs_for_generation

hack so that newer version of HF would support conditional generation with prefix_length

load_metric to support newer HF versions

Merge remote-tracking branch 'origin/main'

fixed custom weights loading for newer transformers versions

Merge pull request #3 from UriSha/patch-1

link fix

support for loading finetuned (SLED) model weights from the hub

fix to do_predict only

run prediction directly from a remote model card

run prediction directly from a remote model card

udpated "generation_max_length" in all data configs

Merge pull request #2 from urialon/main