`ValueError` when following finetuning `mpt-7b-arc-easy--gpu.yaml` example with different default batch size about llm-foundry HOT 2 CLOSED

It looks like your tokenizer is producing token_type_ids, which it is not supposed to. This is generally specified on the tokenizer class (https://github.com/huggingface/transformers/blob/06901162b53918c302306b65b90c6a8ec99e0d0b/src/transformers/models/llama/tokenization_llama.py#L139), but the default is to include it (https://github.com/huggingface/transformers/blob/06901162b53918c302306b65b90c6a8ec99e0d0b/src/transformers/tokenization_utils_base.py#L1559). If you want to specify not to include it at run time, you can override with kwargs like so

tokenizer:
  kwargs:
    model_input_names:
      - input_ids
      - attention_mask

from llm-foundry.

It worked! Thanks Daniel. 😊

from llm-foundry.