What are `mmlu_continuation` and `mmlu_generative`?

[shizhediao]

What are mmlu_continuation and mmlu_generative? Where can I find their description?

I am going to test mmlu in the cloze way. Like the following illustration:

Thank you!

[baberabb]

Hi! mmlu_generative is a generation variant of mmlu (so can be used when log probs aren't available). mmlu_continuation is cloze style multiple-choice but it does not put the choices in the context. So for example the prompt will be <question> <answer_choice_a>, <question> <answer_choice_b>... (compared to default MMLU where it's <question> <choices> <A.>, <question> <choices> <B.>).
If you want the choices in context as well, I think the simplest way is probably replacing the doc_to_choice in the default config with {{choices}}

[shizhediao]

I see. It is quite clear. Thank you so much!

[shizhediao]

Hi, Just to confirm, is my following command correct?

dataset_path: hails/mmlu_no_train # a copy of `cais/mmlu` with no auxiliary_train split
test_split: test
fewshot_split: dev
fewshot_config:
  sampler: first_n
output_type: multiple_choice
doc_to_text: "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:"
doc_to_choice: {{choices}}
doc_to_target: answer
metric_list:
  - metric: acc
    aggregation: mean
    higher_is_better: true
metadata:
  version: 1.0
dataset_kwargs:
  trust_remote_code: true

[haileyschoelkopf]

Hi @shizhediao , this looks correct to me, although if you wanted the

"The following are multiple-choice questions (with answers) about ..." you would need to add that to the description config field!

[shizhediao]

Thank you!