NER Remaining Work

[luisquintanilla]

• Provide E2E sample
• Look into Tokenizer questions / concerns
  • Data needs to be tokenized
  • Choosing a different tokenizer

The tokenizer may not be possible because it's what the model expects. However, you should be able to use a different vocabulary.

[luisquintanilla]

Blocking Model Builder as well.

cc @zewditu

[zewditu]

It appears that we must adhere to a specific version: libtorch-cpu version 1.13.0.1, I think the is the main blocker for GPU usage in Model Builder.