lsht.yaml

tag:
  - longbench
task: longbench_lsht
dataset_path: THUDM/LongBench
test_split: test
dataset_name: lsht
doc_to_text: '请判断给定新闻的类别，下面是一些例子。\n\n{{context}}\n{{input}}'
doc_to_target: '{{answers}}'
process_results: !function metrics.get_classification_score
generation_kwargs:
  max_gen_toks: 64
  temperature: 1
  do_sample: True
  until: ["\n"]
metric_list:
  - metric: "classification_score"
    aggregation: mean
    higher_is_better: True
metadata:
  version: 3.0