Closed bugface closed 2 years ago
https://github.com/huggingface/transformers/issues/10321 - model parallel is not implemented yet so loading tensor model parallel model (multi ranks) is not available in HF
supported in commit 8bdc9f59e78c326cc50ca3a4c2722b3b2db539a3, will be merged into main soon
follow up with https://github.com/huggingface/transformers/pull/10911 to include megatron-bert for NER (the PR has not been merged yet)