Kamal Raj Kanakarajan

@luheng Did you use the complete ontonotes 5.0 dataset or CoNLL 2012 train/test/dev split ?

SPM Tokenizer

1

What is the training parameters used for the SPM tokenizer ? How to initialize a new SPM tokenizer for training from scratch? @BigBird01

+ better readability for the metadata written

HF pipeline based inference

9

Hi @pommedeterresautee, Please let me know your thoughts on converting the text classification also to HF pipeline, similar to Token classification and QA pipeline. I can work on this feature....

Running Bloom

5

What kind of machine is required to just run the inference on the 176B model? https://huggingface.co/bigscience/bloom

Huggingface hub integration

2

Any plans on integrating the model download with huggingface hub. Instead of all models in single repo, we can have 2 repo , one each for `mini` and `mega`. I...

Why Tagging Task segment ids are ones instead of zeros? https://github.com/google-research/electra/blob/8a46635f32083ada044d7e9ad09604742600ee7b/finetune/tagging/tagging_tasks.py#L144 Tagging task only contains the first segment and it should be zeros, right? @clarkkev

ConstrainedGenerator

1

Hi @renqianluo, What is special in ConstrainedGenerator ? Thanks Kamal